release: 0.101.0 (improvement/e36-delivery)

- The fast join gives each ffconcat entry its own media-only format whitelist, so a library file
  can no longer be opened as a nested concat list (the list's whitelist was handed down to it).
- A file the guarded probe cannot read is a download-only file, not a guess from yt-dlp's info.
- A conversion's or an ingest's commit that raises is re-read in a fresh session before any file
  is removed: a commit that landed keeps the file its row now names.
- An original that cannot be removed after conversion is logged; the upload stays ready.
- An upload whose conversion is cancelled or deleted midway is freed once nothing holds it; a
  paused one is kept.
- A shared link says "preparing" only while its video's job is queued or running.
- The admin storage view counts every file on disk ("Files on disk"), not only ready ones.
- A clip's failure title, unknown failure and cancel dialog, and a clip's and an upload's delete
  dialog speak of what the job made; an upload's quota refusal explains that the original is kept.
- /watch announces "preparing" as an alert and moves focus to the new Retry after an action.

.dockerignore

@@ -4,6 +4,7 @@
 *.env.local
 **/__pycache__/
 **/*.pyc
+**/.coverage
 .venv/
 venv/
 # Docker patterns are anchored at the context root with NO implicit `**/` (unlike .gitignore, where

.env.example

@@ -28,7 +28,7 @@ ALLOWED_EMAILS=
 # Admin emails (subset of the above) get the admin role.
 ADMIN_EMAILS=
 
-# Logging (E22). LOG_LEVEL is the BOOTSTRAP level — the level a process starts at and the only one
+# Logging. LOG_LEVEL is the BOOTSTRAP level — the level a process starts at and the only one
 # the worker/download child read; the API's running level is the admin Configuration page's
 # `log_level` knob, which takes effect without a restart. TRACE | DEBUG | INFO | WARNING | ERROR, in
 # any letter case; any other value stops the boot with an error. LOG_FORMAT is "json" (one object per

.git-blame-ignore-revs

@@ -2,8 +2,11 @@
 # whole-repo reformatting does not bury the real authorship of each line.
 # Configure once per clone:  git config blame.ignoreRevsFile .git-blame-ignore-revs
 #
-# Prettier one-time repo-wide format pass (R2 S1b, 2026-07-21)
+# Prettier one-time repo-wide format pass (2026-07-21)
 51c36c2071728d2cd7ae580b7bfdf21f6867a347
 
 # ruff format one-time backend + scripts pass (line-length 100, 2026-08-23)
 b1832afd180866b922521148603db447020c57a4
+
+# Plan-code purge from comments, one-time (2026-10-04)
+4aaaaa786017278460e0b4f94cc0d3374e0e37e2

.gitea/workflows/check.yml

@@ -1,119 +0,0 @@
-# Server-side mirror of the local `siftlode check` gate (E1 S3 / DX-1), so `main` is always green
-# independent of any one machine. Two jobs match the two local lanes:
-#   frontend — tsc (3 projects) + eslint + prettier --check + knip + vitest
-#   backend  — ruff + the full pytest suite (pure-logic + the DB-backed lane) against a real Postgres
-#
-# They are declared side by side but they DO NOT run in parallel: the lab has a single act_runner
-# with one slot, so they queue. Measured across runs #86-#91 — the backend job starts at the exact
-# second the frontend one ends, every time, and the pair costs ~4 + ~6 = **~10 minutes from the
-# push**. Anything that waits on this gate has to budget for the sum, not the max; the ops
-# `release` read-back timed out on its first real run for believing otherwise.
-#
-# Runner: the lab act_runner (LXC 115) mounts the Docker socket, so service containers work. The backend
-# runs pytest NATIVELY against a `postgres` service (not the local docker-compose.test.yml, whose
-# bind-mount is unreliable under a mounted-socket runner); fidelity is kept by installing the SAME pinned
-# requirements, Postgres 16, and ffmpeg, and by setting SIFTLODE_TEST_REQUIRE_DB=1 so a DB/ffmpeg that
-# failed to come up is a hard error, never a silent skip.
-#
-# The frontend container's node major is NOT free: it must equal the prod build image's and the dev
-# machine's, because vitest/tsc read the engine's own Intl and web-API implementations and those split
-# between majors. `.nvmrc` is the single source of truth and
-# `frontend/src/lib/nodeEngine.contract.test.ts` fails this gate if any of them drift apart.
-name: gate
-
-# Branch pushes only. A release pushes `main` and then the `v<ver>` tag, and a bare `on: [push]`
-# treated the tag as a second push of the IDENTICAL commit — every release ran this whole gate
-# twice, to the same verdict, for nothing (runs #81-#90 are five such pairs). `branches: ['**']`
-# keeps every branch covered and drops tag refs.
-#
-# It has to be `branches`, NOT the `tags-ignore` this obviously wants to be. Gitea's matchPushEvent
-# requires EVERY named filter to match, and each one bails on the wrong ref kind: `tags-ignore`
-# starts with `if !refName.IsTag() { break }`, so a filter meant to skip tags rejects every BRANCH
-# push and the gate stops running entirely. `branches` bails the same way on a tag, which is the
-# direction we want. (Measured in go-gitea v1.27.3 modules/actions/workflows.go, not assumed — it
-# differs from GitHub, where `tags-ignore` alone leaves branches unfiltered.)
-#
-# If that filter were ever wrong the gate would not run at all — which is why `siftlode release`
-# now ends by reading the verdict back for the sha it pushed, and says so out loud when there is
-# no run for it. A gate that silently stops running is the failure this repo already lived through.
-on:
-  push:
-    branches: ["**"]
-  pull_request:
-
-jobs:
-  frontend:
-    runs-on: ubuntu-latest
-    container: node:24-bookworm
-    defaults:
-      run:
-        working-directory: frontend
-    steps:
-      - uses: actions/checkout@v4
-      - run: npm ci
-      - run: npm run typecheck
-      - run: npm run lint
-      - run: npm run format:check
-      - run: npm run knip
-      - run: npm run test
-
-  backend:
-    runs-on: ubuntu-latest
-    # A combined python+node image: python 3.12 (matches the prod Dockerfile) for the tests, and node
-    # so the JS actions (checkout) run. actions/setup-python is NOT usable here — the runner can't reach
-    # GitHub's python-versions download, so it must be baked into the image.
-    #
-    # This node runs the checkout action, never our code, so it is deliberately OUTSIDE the engine
-    # contract the frontend job is held to — it tracks `.nvmrc` as hygiene (node 20 went EOL in April
-    # 2026), but a missing upstream tag here must never be able to block the gate.
-    container: nikolaik/python-nodejs:python3.12-nodejs24
-    services:
-      postgres:
-        image: postgres:16-alpine
-        env:
-          POSTGRES_USER: siftlode
-          POSTGRES_PASSWORD: siftlode
-          POSTGRES_DB: siftlode_test
-        options: >-
-          --health-cmd "pg_isready -U siftlode -d siftlode_test"
-          --health-interval 2s
-          --health-timeout 3s
-          --health-retries 30
-    env:
-      DATABASE_URL: postgresql+psycopg://siftlode:siftlode@postgres:5432/siftlode_test
-      SIFTLODE_TEST_REQUIRE_DB: "1"
-    steps:
-      - uses: actions/checkout@v4
-      # ffmpeg for the poster/thumbnail tests; postgresql-client for the readiness wait below.
-      - run: apt-get update && apt-get install -y --no-install-recommends ffmpeg postgresql-client
-      - name: Wait for Postgres
-        run: until pg_isready -h postgres -U siftlode -d siftlode_test; do sleep 1; done
-      - run: pip install --no-cache-dir -r backend/requirements.lock -r backend/requirements-dev.txt
-      - name: ruff
-        run: python -m ruff check --config backend/ruff.toml backend scripts
-      - name: ruff format --check
-        run: python -m ruff format --check --config backend/ruff.toml backend scripts
-      # The ninth lane (E27 S3). Run from backend/ so it resolves exactly as the local lane does —
-      # pyrightconfig.json's include/extraPaths are relative to the config, but the baseline is
-      # found relative to the working directory. New errors fail; the ~200 pre-existing ones are in
-      # .basedpyright/baseline.json and burn down epic by epic. Matching is per (file, rule, column
-      # span) plus a count, so a new error of the SAME shape in a file where a sibling was just
-      # fixed is absorbed — burn the hot files down first.
-      #
-      # The second step is not redundant. basedpyright TIGHTENS the baseline in place when the count
-      # drops, with no --writebaseline, and CI throws its workspace away — so without this a
-      # burn-down is green here and red on the developer's `siftlode check`, which has the same
-      # guard. Two gates over one tree must not answer differently.
-      - name: basedpyright
-        run: python -m basedpyright --project pyrightconfig.json
-        working-directory: backend
-      - name: basedpyright baseline is committed
-        run: git diff --exit-code -- backend/.basedpyright/baseline.json
-      - name: pytest
-        run: python -m pytest
-        working-directory: backend
-      # Dependency CVE audit over the pinned lock. Non-blocking for now (informational) — a real CVE
-      # shows up in the run without failing the gate; tighten to a hard fail once the noise is known.
-      - name: pip-audit (advisory)
-        run: pip install --no-cache-dir pip-audit && pip-audit -r backend/requirements.lock
-        continue-on-error: true

.gitignore

@@ -9,6 +9,7 @@ __pycache__/
 venv/
 .mypy_cache/
 .pytest_cache/
+.coverage
 .ruff_cache/
 
 # Node / frontend
@@ -41,3 +42,4 @@ Thumbs.db
 # TypeScript incremental build artifacts — machine-local, absolute-path-bearing, and no gate lane
 # reads or regenerates them, so a committed one is stale from the commit that adds it.
 *.tsbuildinfo
+frontend/e2e/.speed/

.trivyignore.yaml

@@ -0,0 +1,15 @@
+# The CVE gate's allowlist. The publish gate scans the built image and the committed tree with
+# Trivy and fails on any HIGH or CRITICAL finding that has a fix available. A finding that cannot be
+# fixed yet is let through ONLY by an entry here, and every entry carries both:
+#   statement   — why it is acceptable for now (what is exposed, what is not)
+#   expired_at  — the date it stops being accepted; after it, the gate fails again on its own
+# A contract test (frontend/src/lib/gates.contract.test.ts) refuses an entry without either, and
+# docs/gates.md lists them.
+#
+# Example:
+#   vulnerabilities:
+#     - id: CVE-2026-00000
+#       statement: in libfoo's XML path, which nothing in the image calls; no fixed Debian build yet
+#       expired_at: 2026-12-31
+vulnerabilities: []
+secrets: []

CHANGELOG.md

@@ -7,6 +7,88 @@ edit the release notes and re-run the generator.
 
 Upgrading, pinning a version and rolling back: [docs/self-hosting.md](docs/self-hosting.md#upgrading).
 
+## 0.101.0 — 2026-10-06
+
+### New
+
+- A shared link plays on any phone, iPhone and iPad included: downloads, clips and uploads are kept in a format every phone plays, and an upload in another format is converted after it arrives.
+- A shared link opens much faster: its page no longer loads the whole app.
+
+### Fixed
+
+- The YouTube player keeps its size on a short window and when maximised — before, it could shrink to a thin strip. Plus smaller fixes around uploads and clips.
+
+## 0.100.0 — 2026-10-05
+
+### New
+
+- Direct messages are gone: the Messages page, the chat dock and their encryption keys were removed with their stored messages, and the privacy policy no longer describes end-to-end encrypted chat.
+- The shared demo account is gone, with its sign-in entry and its whitelist.
+
+### Fixed
+
+- Sign-in security without email: when a Google sign-in takes over an account someone else registered first, or makes it an admin through the admin list, its password is cleared and its sessions end. The owner sets a new one in Settings.
+- Pressing Back onto a page that no longer exists opens the feed instead of breaking the app.
+- Behind the scenes: tidier checks and comments across the code.
+
+## 0.99.2 — 2026-10-04
+
+### Fixed
+
+- Settings: the download worker's concurrency and the log level now take effect within seconds — before, the worker ignored both until it was rebuilt.
+- The YouTube quota history no longer grows forever: the daily cleanup prunes it after a year (Configuration → Quota → Quota history kept).
+- The live-search button no longer says a search spends the API quota: with the default website source the search itself is free. It names the cost when the official API is selected.
+
+## 0.99.1 — 2026-10-04
+
+### Fixed
+
+- Behind the scenes: every release is now checked by one script on the build machine, including the security audit of every dependency.
+
+## 0.99.0 — 2026-10-04
+
+### New
+
+- Nearly every screen is now compiled by the React Compiler, so lists and menus re-render less.
+- The channel list's row menu opens and closes without redrawing the table.
+
+### Fixed
+
+- A failed or empty answer from YouTube or Plex no longer empties a playlist, a mirror or your Plex watch history: Siftlode keeps what it had until a complete answer arrives.
+- Plex watch states un-watched on Plex now sync for each library on its own, even if another library is briefly unreachable.
+- A channel with no uploads no longer fails its backfill on every run.
+- “This playlist no longer exists” appears only for a list you picked yourself.
+- A filter on a deleted tag matches nothing and its chip says so, and the tag cleanup keeps tags a saved view filters on.
+- A channel filter names its channels from the first moment, and a playlist reopened from the cache keeps its rows.
+- The bug report opens on its title, and a dialog whose form loads late gives the form the focus once.
+- The Plex backdrop crossfade setting is respected even with reduced motion on.
+- Smaller fixes to keyboard focus, undo and loading states.
+
+## 0.98.0 — 2026-10-03
+
+### New
+
+- Search is much faster: a short or common word now answers in a fraction of a second, with the newest matches ranked by relevance.
+- The feed, the channel list and the Plex pages load faster, and Plex cards fetch a picture sized to the card.
+- While a new search counts its results, the count dims instead of showing the old number.
+
+### Fixed
+
+- A list whose next page failed to load says so and offers Retry, instead of asking again and again.
+- The filter rail's tag counts follow the current search.
+- Smaller fixes to layout, keyboard focus and wording.
+
+## 0.97.0 — 2026-10-01
+
+### New
+
+- The app speaks Hungarian throughout: Plex genres and statuses, numbers, units and A→Z order follow your language.
+- Person pages show a Hungarian biography where TMDB has one, and open faster, with headshots kept on the server.
+
+### Fixed
+
+- Smaller fixes to wording and the Plex pages.
+
 ## 0.96.0 — 2026-09-27
 
 ### New

CLAUDE.md

@@ -12,13 +12,16 @@ router-free navigation provider.
 
 ## The gate — never merge or ship red
 
-The local gate is **ruff (`check` + `format --check`) + basedpyright (backend) + typecheck (3 tsc projects) + eslint + prettier
---check + knip + vitest + pytest**. It runs on every `publish`, which **refuses to build on failure — there is deliberately no
-`--skip` flag.** Run it and get it green **before** any merge, release, or handing work to review.
+The local gate is **ruff (`check` + `format --check`) + basedpyright (backend) + the lock check + typecheck (3 tsc projects) +
+eslint + prettier --check + knip + vitest + vite build + pytest + pip-audit + npm audit**. It runs
+before every `release` merges and on every `publish`, which **refuses to build on failure — there is deliberately no `--skip`
+flag**; a release also requires the Playwright suite, and the image must pass a Trivy scan before it is pushed. Run it and get it
+green **before** any merge, release, or handing work to review.
 
-- Full gate (agents with the private ops dispatcher): `siftlode check`.
-- Directly, without the dispatcher: `cd frontend && npm run typecheck && npm run lint && npm run knip && npm run test`, plus backend `ruff check backend` + `ruff format --check backend` and `pytest` (the DB-lane needs Postgres).
-- **The backend type lane runs in the test image, not on the host** — `requirements.lock` cannot be installed on Windows (uvloop), and a type checker without the real packages reports nonsense. It fails on NEW errors only; the pre-existing ones sit in `backend/.basedpyright/baseline.json` and burn down epic by epic. Fixed some? Regenerate with `--writebaseline`.
+- Full gate: `bash scripts/check.sh` (`lint` for the static lanes only); the private ops dispatcher's `siftlode check` runs the same script. Single lanes: [`docs/development.md`](docs/development.md) § The gate.
+- **There is no CI** (owner, 2026-10-04): this gate, run here, is the only one — green before every hand-over, and `release` refuses to merge without it.
+- **A gate with any real hit is never switched off or narrowed** — not a lint rule, a compiler flag, a lane or a release step. Fix the sites, or suppress each one inline naming its invariant (`-- why`, lint-enforced). What stays narrowed is in [`docs/gates.md`](docs/gates.md), and a contract test fails when a narrowing is missing from it. The effective ESLint config of every scope and the gate/lane lines are pinned in `frontend/src/lib/gates.snapshot.json`: any gate change fails until you regenerate it (`GATE_SNAPSHOT=update`, see gates.md) — then read that diff as a gate change. A comment saying "X is enforced by Y" is checked by running Y (E32: one such comment hid 95 functions the React Compiler skipped).
+- **The backend type lane runs in the test image, not on the host** — `requirements.lock` cannot be installed on Windows (uvloop), and a type checker without the real packages reports nonsense. It fails on NEW errors only; the pre-existing ones sit in `backend/.basedpyright/baseline.json` (no burn-down mandate, but none was ever triaged: a `reportOptional*` entry in code you touch may be a real bug). A run that fixed some rewrites it — commit that.
 - New DB-backed test? It must actually run, not silently skip — assume `SIFTLODE_TEST_REQUIRE_DB=1`.
 
 ## Workflow
@@ -60,9 +63,10 @@ how it gets built. Concretely:
 
 `routes/*` (HTTP only: validation, auth-dependency, serialization) → domain packages `sync/`, `plex/`,
 `downloads/`, `youtube/`, `issues/` (shared by routes + scheduler + worker) → flat core (`auth`, `quota`,
-`sysconfig`, `state`, `audit`, `notifications`, `userscope`, `realtime`, `ratelimit`, `bodylimit`, `progress`,
+`sysconfig`, `state`, `audit`, `notifications`, `userscope`, `ratelimit`, `bodylimit`, `progress`, `collation`,
 `security`, `logs`, `logsink`, `diagnostics`, `debugsession`) → `models/db/config`. **Dependency direction is one-way: core never imports routes.**
-Frontend: `main.tsx` provider tree → `App.tsx` shell → `moduleRegistry` route-chunks → module
+Frontend: `main.tsx` provider tree → `App.tsx` shell (a shared `/watch` link boots from its own
+entry, `watch.html` → `watchMain.tsx`, held app-free by a contract test) → `moduleRegistry` route-chunks → module
 components → the **platform layer `components/ui/`** (domain-free primitives: the popover engine and
 its menus, `Modal`/`Overlay`, `Button`/`form`, `DataTable`/`VirtualGrid`, `QueryState`, …). That
 boundary is one-way and **lint-enforced**, not just described: a platform component may not import a
@@ -77,37 +81,39 @@ CQRS, a repository/DI framework, a router library, a generic CRUD-router factory
 
 ## Canonical patterns — when you have a choice, pick THIS
 
-| Area                                 | Use                                                                                                                     | Not                                                                                  |
-| ------------------------------------ | ----------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------ |
-| ORM                                  | SQLAlchemy 2.0 `select()`                                                                                               | `db.query(...)`                                                                      |
-| Request body                         | a Pydantic model                                                                                                        | `payload: dict`                                                                      |
-| Request SIZE                         | nothing — `bodylimit` caps every route (SEC-20); need more? add a ceiling to `ROUTE_MAX_BODY_BYTES`                     | `capped_body` on a route that does not read raw bytes itself                         |
-| Ownership check                      | `owned_or_404` / `get_or_404`                                                                                           | ad-hoc `_own_playlist`/`_owned`/`_own_job`                                           |
-| Admin gate import                    | `app.auth`                                                                                                              | the `app.routes.admin` re-export                                                     |
-| Upstream failure status              | `422` with JSON detail                                                                                                  | `502` (reserve 502 for raw byte-proxies)                                             |
-| User-facing error                    | structured `detail` (`{code, reason}`)                                                                                  | a bare English string under a bilingual UI                                           |
-| Mutation response                    | `{"ok": true, …}` (`{"deleted": id}`), test-held                                                                        | a bare `{"unsubscribed"}`/`{"cleared": n}`, no `ok`                                  |
-| Transaction commit                   | the handler commits; helpers `flush` (test-held list)                                                                   | helpers that commit (only `quota` owns its session)                                  |
-| Admin knob storage                   | `sysconfig` (values)                                                                                                    | env / `app_state` / `scheduler_settings` scatter                                     |
-| Job status word                      | the `downloads/status.py` StrEnum                                                                                       | loose `"canceled"`/`"cancelled"` literals                                            |
-| Current time                         | inline `datetime.now(timezone.utc)`                                                                                     | new `now_utc`/`_now` helpers                                                         |
-| FE per-account state                 | `usePref`                                                                                                               | raw `readAccount`/`writeAccount` or bespoke hooks                                    |
-| FE button                            | `ui/Button` (a `SIZES` rung) · `IconButton` · `Pressable` (a clickable card/row) · `Chip` · `DragHandle`                | a raw `<button>` or `FocusSafeButton` outside the control primitives (lint-enforced) |
-| FE keyboard shortcut                 | a row in `lib/keymap.ts` + `dispatch()`                                                                                 | a private `keydown` listener with its own key literals                               |
-| FE list/grid keys                    | `NavGrid`/`keyboardNav` + `navItemAttrs` + `data-nav-open`                                                              | a private arrow-key / roving-tabindex handler                                        |
-| FE view push (history)               | `rememberNavPosition()` right before `history.pushState`                                                                | a push that forgets: Back strands focus on `<body>`                                  |
-| FE text field / checkbox / slider    | `ui/Input` · `Textarea` · `InnerInput` in a `fieldClass` frame · `Checkbox`/`CheckRow` · `Slider`                       | a raw `<input>`/`<textarea>` anywhere but those primitives (lint-enforced)           |
-| FE glass surface                     | a role: `SURFACE.panel` / `card` / `menu` / `popover` / `tooltip` (`ui/surfaces.ts`)                                    | a bare `glass`/`glass-card`/`glass-menu` beside a hand-picked radius (lint-enforced) |
-| FE static label / button-like link   | `Badge` (flat, no edge) · `LinkButton` / `IconLink`                                                                     | a glass pill that reads as a chip; an `<a>` hand-painted like a button               |
-| FE control                           | `SegmentedToggle`; a long or user's list: `FilterSelect`                                                                | a native `<select>` (lint-enforced)                                                  |
-| FE loading/empty/error               | `QueryState` / `StateMessage`                                                                                           | ad-hoc conditionals                                                                  |
-| FE failed write                      | `meta.failTitle`, else `notifyActionError(err, title)`                                                                  | a bare `catch {}` or `notify()` with no error attached                               |
-| FE network call                      | `req`/`uploadReq`/`beacon`, or `lib/http.ts` `send` (public pages)                                                      | a bare `fetch`/`XMLHttpRequest`/`WebSocket` (lint-enforced)                          |
-| FE diagnostics                       | `lib/log.ts` (`log.info("event.code", msg, attrs)`); a filter change → `lib/filterLog.ts`, a player → `lib/mediaLog.ts` | `console.*` (lint-enforced); a value only a higher level may carry, logged unguarded |
-| FE state that MIRRORS a prop/query   | derive it, or hold a draft (`draft ?? server`)                                                                          | a copy an effect re-syncs — that paints the stale value first                        |
-| FE state reset on a key change       | adjust it DURING render against a remembered key                                                                        | `useEffect(() => setX(init), [key])`                                                 |
-| FE effect that must not re-subscribe | `useEffectEvent` (React 19.2+), else a latest-value ref                                                                 | an excused dep list asserting the same thing in prose                                |
-| FE hook suppression                  | an inline `eslint-disable-next-line` NAMING the invariant                                                               | `eslint-suppressions.json` (deliberately absent — E27 S2)                            |
+| Area                                 | Use                                                                                                                                                                                               | Not                                                                                                                                                     |
+| ------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| ORM                                  | SQLAlchemy 2.0 `select()`                                                                                                                                                                         | `db.query(...)`                                                                                                                                         |
+| Request body                         | a Pydantic model                                                                                                                                                                                  | `payload: dict`                                                                                                                                         |
+| Request SIZE                         | nothing — `bodylimit` caps every route (SEC-20); need more? add a ceiling to `ROUTE_MAX_BODY_BYTES`                                                                                               | `capped_body` on a route that does not read raw bytes itself                                                                                            |
+| Ownership check                      | `owned_or_404` / `get_or_404`                                                                                                                                                                     | ad-hoc `_own_playlist`/`_owned`/`_own_job`                                                                                                              |
+| Admin gate import                    | `app.auth`                                                                                                                                                                                        | the `app.routes.admin` re-export                                                                                                                        |
+| Upstream failure status              | `422` with JSON detail                                                                                                                                                                            | `502` (reserve 502 for raw byte-proxies)                                                                                                                |
+| User-facing error                    | structured `detail` (`{code, reason}`)                                                                                                                                                            | a bare English string under a bilingual UI                                                                                                              |
+| Mutation response                    | `{"ok": true, …}` (`{"deleted": id}`), test-held                                                                                                                                                  | a bare `{"unsubscribed"}`/`{"cleared": n}`, no `ok`                                                                                                     |
+| Transaction commit                   | the handler commits; helpers `flush` (test-held list)                                                                                                                                             | helpers that commit (only `quota` owns its session)                                                                                                     |
+| Admin knob storage                   | `sysconfig` (values)                                                                                                                                                                              | env / `app_state` / `scheduler_settings` scatter                                                                                                        |
+| Job status word                      | the `downloads/status.py` StrEnum                                                                                                                                                                 | loose `"canceled"`/`"cancelled"` literals                                                                                                               |
+| Current time                         | inline `datetime.now(timezone.utc)`                                                                                                                                                               | new `now_utc`/`_now` helpers                                                                                                                            |
+| FE per-account state                 | `usePref`                                                                                                                                                                                         | raw `readAccount`/`writeAccount` or bespoke hooks                                                                                                       |
+| FE keyboard shortcut                 | a row in `lib/keymap.ts` + `dispatch()`                                                                                                                                                           | a private `keydown` listener with its own key literals                                                                                                  |
+| FE list/grid keys                    | `NavGrid`/`keyboardNav` + `navItemAttrs` + `data-nav-open`                                                                                                                                        | a private arrow-key / roving-tabindex handler                                                                                                           |
+| FE view push (history)               | `rememberNavPosition()` right before `history.pushState`                                                                                                                                          | a push that forgets: Back strands focus on `<body>`                                                                                                     |
+| FE static label / button-like link   | `Badge` (flat, no edge) · `LinkButton` / `IconLink`                                                                                                                                               | a glass pill that reads as a chip; an `<a>` hand-painted like a button                                                                                  |
+| FE loading/empty/error               | `QueryState` / `StateMessage`                                                                                                                                                                     | ad-hoc conditionals                                                                                                                                     |
+| FE failed write                      | `meta.failTitle`, else `notifyActionError(err, title)`                                                                                                                                            | a bare `catch {}` or `notify()` with no error attached                                                                                                  |
+| FE diagnostics                       | `lib/log.ts` (`log.info("event.code", msg, attrs)`); a filter change → `lib/filterLog.ts`, a player → `lib/mediaLog.ts`                                                                           | `console.*` (lint-enforced); a value only a higher level may carry, logged unguarded                                                                    |
+| FE state that MIRRORS a prop/query   | derive it, or hold a draft (`draft ?? server`)                                                                                                                                                    | a copy an effect re-syncs — that paints the stale value first                                                                                           |
+| FE state reset on a key change       | adjust it DURING render against a remembered key                                                                                                                                                  | `useEffect(() => setX(init), [key])`                                                                                                                    |
+| FE effect that must not re-subscribe | `useEffectEvent` (React 19.2+); a handler or timer: `useLatest` / `useStableCallback`                                                                                                             | a render-phase `ref.current = x` — the React Compiler skips the component                                                                               |
+| FE hook suppression                  | an inline `eslint-disable-next-line` NAMING the invariant                                                                                                                                         | `eslint-suppressions.json` (deliberately absent — E27 S2)                                                                                               |
+| FE popover / scroll-fade ref         | destructure it: `const { setReference } = pop.refs`, then `ref={setReference}`                                                                                                                    | `ref={pop.refs.setReference}` / `ref={fade.ref}` — the compiler then treats the whole object as a ref and skips                                         |
+| FE async handler with cleanup        | `try { … } catch { … }`, then the cleanup after it (steps that `return` early go in their own function); a conditional, loop or `?.` the `try` needs goes before it or into a module-level helper | `try/finally`, or a `?:` / `??` / loop / `?.` inside `try` — the React Compiler skips the whole component (and `react-hooks/todo` misses most of these) |
+
+Not repeated here, because a lint rule refuses the wrong one and its message names the right one:
+buttons (`ui/Button`, `IconButton`, `Pressable`, `Chip`, `DragHandle`), text fields and native
+controls (`ui/Input`, `Checkbox`, `Slider`, …), glass surfaces (`SURFACE.*`), a native `<select>`
+(`SegmentedToggle` / `FilterSelect`), and network calls (`req` / `uploadReq` / `beacon`, `lib/http.ts`).
 
 ## House rules that bite (learned from real bugs)
 
@@ -120,7 +126,8 @@ CQRS, a repository/DI framework, a router library, a generic CRUD-router factory
   "no results". Reads degrade honestly; writes are conditional (`UPDATE … WHERE status='running'`).
 - **i18n parity.** HU/EN, zero key drift. Every user-facing string is translated.
 - **Comment the "why", keep it current.** This codebase uses comments as its architecture doc; a stale
-  "why" is a liability, so update it when the behavior changes.
+  "why" is a liability, so update it when the behavior changes. Say the why itself, never the plan
+  step that produced it: no epic, sprint or review-round ids in the code (`planCodes.contract.test.ts`).
 
 ---
 

Dockerfile

@@ -11,7 +11,7 @@ ARG BUILD_DATE=
 
 # Stage 1: build the frontend SPA
 # The major is pinned to `.nvmrc` and held there by frontend/src/lib/nodeEngine.contract.test.ts:
-# this image, the CI gate's image and the dev machine must be ONE engine (see that test for why).
+# this image and the dev machine the gate runs on must be ONE engine (see that test for why).
 FROM node:24-alpine AS frontend
 WORKDIR /fe
 # Deps first, keyed on the lockfile, so editing app source never re-resolves node_modules.
@@ -47,14 +47,17 @@ RUN adduser --disabled-password --gecos "" appuser \
     && mkdir -p /downloads \
     && chown appuser /app /downloads
 
-# ffmpeg: merge separate video+audio streams + postprocessing, and the Plex VAAPI transcode (E6 S1).
+# ffmpeg: merge separate video+audio streams + postprocessing, and the Plex VAAPI transcode.
 # intel-media-va-driver (iHD): the VA driver ffmpeg's h264_vaapi needs — libva2 ships in the base but
 # no driver, so without this `vainfo`/`h264_vaapi` fail and transcode silently falls back to libx264.
 # Harmless on hosts with no /dev/dri (the driver just goes unused). Deno: the JavaScript runtime yt-dlp
 # uses to solve YouTube's "n" signature challenge for the web player clients (required alongside the PO
 # token, else only storyboards are offered). yt-dlp auto-detects deno on PATH.
+# `apt-get upgrade`: the base image is a snapshot, and the Debian security fixes released since it
+# was cut are only in the archive — the first Trivy scan found one (pcre2) fixed upstream but not in it.
 ARG DENO_VERSION=v2.9.1
 RUN apt-get update \
+    && apt-get upgrade -y --no-install-recommends \
     && apt-get install -y --no-install-recommends ffmpeg intel-media-va-driver unzip curl ca-certificates \
     && curl -fsSL "https://github.com/denoland/deno/releases/download/${DENO_VERSION}/deno-x86_64-unknown-linux-gnu.zip" -o /tmp/deno.zip \
     && unzip -o /tmp/deno.zip -d /usr/local/bin \
@@ -76,11 +79,14 @@ RUN printf 'Types: deb\nURIs: http://deb.debian.org/debian\nSuites: bookworm\nCo
     && rm -rf /var/lib/apt/lists/*
 
 # Install from the PINNED lock (requirements.lock, generated from requirements.txt) so the image is
-# byte-reproducible and the exact versions are the ones pip-audit scanned in CI — not whatever floats
-# up from the floor ranges on build day. Regenerate the lock when requirements.txt changes (see its header).
+# byte-reproducible and the exact versions are the ones the gate's pip-audit scanned — not whatever
+# floats up from the floor ranges on build day. Regenerate the lock when requirements.txt changes (see its header).
+# pip itself is removed once it has installed the lock: nothing runs it at runtime, and it carries
+# its own VENDORED copies of urllib3, msgpack and setuptools, which Trivy reports as the image's own —
+# an attack surface and a CVE stream for a tool the running app never calls.
 COPY backend/requirements.lock .
 RUN --mount=type=cache,target=/root/.cache/pip \
-    pip install --upgrade pip && pip install -r requirements.lock
+    pip install --upgrade pip && pip install -r requirements.lock && pip uninstall -y pip
 
 COPY --chown=appuser backend/ .
 # Read at runtime by config.py as the app_version (a committed single source of truth).

VERSION

@@ -1 +1 @@
-0.96.0
+0.101.0

backend/.basedpyright/baseline.json

@@ -548,64 +548,6 @@
                 }
             }
         ],
-        "./app/plex/stream.py": [
-            {
-                "code": "reportOptionalMemberAccess",
-                "range": {
-                    "startColumn": 51,
-                    "endColumn": 55,
-                    "lineCount": 1
-                }
-            },
-            {
-                "code": "reportOptionalMemberAccess",
-                "range": {
-                    "startColumn": 23,
-                    "endColumn": 25,
-                    "lineCount": 1
-                }
-            },
-            {
-                "code": "reportOptionalMemberAccess",
-                "range": {
-                    "startColumn": 18,
-                    "endColumn": 31,
-                    "lineCount": 1
-                }
-            },
-            {
-                "code": "reportOptionalMemberAccess",
-                "range": {
-                    "startColumn": 75,
-                    "endColumn": 79,
-                    "lineCount": 1
-                }
-            },
-            {
-                "code": "reportOptionalMemberAccess",
-                "range": {
-                    "startColumn": 10,
-                    "endColumn": 23,
-                    "lineCount": 1
-                }
-            },
-            {
-                "code": "reportOptionalMemberAccess",
-                "range": {
-                    "startColumn": 10,
-                    "endColumn": 12,
-                    "lineCount": 1
-                }
-            },
-            {
-                "code": "reportOptionalMemberAccess",
-                "range": {
-                    "startColumn": 10,
-                    "endColumn": 17,
-                    "lineCount": 1
-                }
-            }
-        ],
         "./app/plex/sync.py": [
             {
                 "code": "reportArgumentType",
@@ -638,22 +580,6 @@
                     "endColumn": 89,
                     "lineCount": 1
                 }
-            },
-            {
-                "code": "reportArgumentType",
-                "range": {
-                    "startColumn": 27,
-                    "endColumn": 33,
-                    "lineCount": 1
-                }
-            },
-            {
-                "code": "reportArgumentType",
-                "range": {
-                    "startColumn": 17,
-                    "endColumn": 18,
-                    "lineCount": 1
-                }
             }
         ],
         "./app/plex/tmdb.py": [
@@ -898,96 +824,6 @@
                 }
             }
         ],
-        "./app/routes/messages.py": [
-            {
-                "code": "reportCallIssue",
-                "range": {
-                    "startColumn": 11,
-                    "endColumn": 43,
-                    "lineCount": 1
-                }
-            },
-            {
-                "code": "reportArgumentType",
-                "range": {
-                    "startColumn": 23,
-                    "endColumn": 27,
-                    "lineCount": 1
-                }
-            },
-            {
-                "code": "reportCallIssue",
-                "range": {
-                    "startColumn": 13,
-                    "endColumn": 5,
-                    "lineCount": 7
-                }
-            },
-            {
-                "code": "reportArgumentType",
-                "range": {
-                    "startColumn": 8,
-                    "endColumn": 15,
-                    "lineCount": 5
-                }
-            },
-            {
-                "code": "reportReturnType",
-                "range": {
-                    "startColumn": 15,
-                    "endColumn": 72,
-                    "lineCount": 1
-                }
-            },
-            {
-                "code": "reportCallIssue",
-                "range": {
-                    "startColumn": 30,
-                    "endColumn": 48,
-                    "lineCount": 1
-                }
-            },
-            {
-                "code": "reportArgumentType",
-                "range": {
-                    "startColumn": 41,
-                    "endColumn": 44,
-                    "lineCount": 1
-                }
-            },
-            {
-                "code": "reportOptionalMemberAccess",
-                "range": {
-                    "startColumn": 36,
-                    "endColumn": 38,
-                    "lineCount": 1
-                }
-            },
-            {
-                "code": "reportOptionalMemberAccess",
-                "range": {
-                    "startColumn": 71,
-                    "endColumn": 73,
-                    "lineCount": 1
-                }
-            },
-            {
-                "code": "reportArgumentType",
-                "range": {
-                    "startColumn": 30,
-                    "endColumn": 39,
-                    "lineCount": 1
-                }
-            },
-            {
-                "code": "reportOptionalMemberAccess",
-                "range": {
-                    "startColumn": 27,
-                    "endColumn": 29,
-                    "lineCount": 1
-                }
-            }
-        ],
         "./app/routes/playlists.py": [
             {
                 "code": "reportOptionalOperand",
@@ -1658,24 +1494,6 @@
                     "endColumn": 36,
                     "lineCount": 1
                 }
-            },
-            {
-                "code": "reportArgumentType",
-                "range": {
-                    "startColumn": 46,
-                    "endColumn": 52,
-                    "lineCount": 1
-                }
-            }
-        ],
-        "./app/youtube/rss.py": [
-            {
-                "code": "reportGeneralTypeIssues",
-                "range": {
-                    "startColumn": 10,
-                    "endColumn": 15,
-                    "lineCount": 1
-                }
             }
         ]
     }

backend/alembic/versions/0091_plex_tagline_hu.py

@@ -0,0 +1,27 @@
+"""add tagline_hu to plex_items + plex_shows — the HU overlay's tagline (E30 S1)
+
+Revision ID: 0091_plex_tagline_hu
+Revises: 0090_debug_session_draft
+Create Date: 2026-09-29
+
+The Hungarian overlay section lists a tagline beside the title and summary; the info page showed the
+English one to a Hungarian reader. NULL until the next overlay sync fills it (serving falls back to EN).
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+revision: str = "0091_plex_tagline_hu"
+down_revision: str | None = "0090_debug_session_draft"
+branch_labels: str | None = None
+depends_on: str | None = None
+
+
+def upgrade() -> None:
+    op.add_column("plex_items", sa.Column("tagline_hu", sa.Text(), nullable=True))
+    op.add_column("plex_shows", sa.Column("tagline_hu", sa.Text(), nullable=True))
+
+
+def downgrade() -> None:
+    op.drop_column("plex_shows", "tagline_hu")
+    op.drop_column("plex_items", "tagline_hu")

backend/alembic/versions/0092_plex_person_biography_hu.py

@@ -0,0 +1,31 @@
+"""add biography_hu to plex_people — TMDB's Hungarian biography (E30 S4)
+
+Revision ID: 0092_plex_person_biography_hu
+Revises: 0091_plex_tagline_hu
+Create Date: 2026-09-29
+
+The person page showed a Hungarian reader the English biography although TMDB has a Hungarian one for
+most people. The person fetch now appends `translations` and keeps the Hungarian biography beside the
+English one. Already-enriched people are un-stamped, so their next page open refetches them once (the
+same lazy path as a first open) and fills it; until then the English biography stands in.
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+revision: str = "0092_plex_person_biography_hu"
+down_revision: str | None = "0091_plex_tagline_hu"
+branch_labels: str | None = None
+depends_on: str | None = None
+
+
+def upgrade() -> None:
+    op.add_column("plex_people", sa.Column("biography_hu", sa.Text(), nullable=True))
+    op.execute(
+        "UPDATE plex_people SET tmdb_enriched_at = NULL "
+        "WHERE tmdb_person_id IS NOT NULL AND tmdb_enriched_at IS NOT NULL"
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("plex_people", "biography_hu")

backend/alembic/versions/0093_plex_person_known_for_votes.py

@@ -0,0 +1,29 @@
+"""re-fetch linked people — "Known for" ranked by votes, without guest visits (E30 S6)
+
+Revision ID: 0093_plex_person_known_for_votes
+Revises: 0092_plex_person_biography_hu
+Create Date: 2026-09-30
+
+A person's "Known for" pool was their credits ranked by TMDB popularity, so talk-show visits as
+themselves crowded out the films (Tom Cruise: 14 of 15 were talk shows, none in the library). The pool is
+now ranked by vote count, without guest visits, with directed works, and deeper. Already-enriched people
+are un-stamped so the people_enrich job fetches them again; their biography stays on the row meanwhile.
+"""
+
+from alembic import op
+
+revision: str = "0093_plex_person_known_for_votes"
+down_revision: str | None = "0092_plex_person_biography_hu"
+branch_labels: str | None = None
+depends_on: str | None = None
+
+
+def upgrade() -> None:
+    op.execute(
+        "UPDATE plex_people SET tmdb_enriched_at = NULL "
+        "WHERE tmdb_person_id IS NOT NULL AND tmdb_enriched_at IS NOT NULL"
+    )
+
+
+def downgrade() -> None:
+    pass  # the watermark only says "fetched"; a refetch under the old ranking is not worth undoing to

backend/alembic/versions/0094_plex_genres_hu_everywhere.py

@@ -0,0 +1,45 @@
+"""Hungarian genres on every Plex row, not only on overlaid ones (E30 review)
+
+Revision ID: 0094_plex_genres_hu_everywhere
+Revises: 0093_plex_person_known_for_votes
+Create Date: 2026-09-30
+
+The HU genre facets and filter read `genres_hu`, which only the overlay pass wrote — so a library with
+no HU twin, or a guid-less row, offered English genres a Hungarian filter could never match. The sync
+now writes `genres_hu` for every row (plex/sync.py `_apply_facet_fields`); this fills the rows it has
+not reached yet, from their own English list through the same map.
+"""
+
+import json
+
+import sqlalchemy as sa
+from alembic import op
+
+from app.plex.genres import hu_genres
+
+revision: str = "0094_plex_genres_hu_everywhere"
+down_revision: str | None = "0093_plex_person_known_for_votes"
+branch_labels: str | None = None
+depends_on: str | None = None
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+    for table in ("plex_items", "plex_shows"):
+        rows = conn.execute(
+            sa.text(
+                f"SELECT id, genres FROM {table} "
+                "WHERE genres_hu IS NULL AND jsonb_typeof(genres) = 'array'"
+            )
+        ).fetchall()
+        updates = [
+            {"id": rid, "g": json.dumps(hu)} for rid, genres in rows if (hu := hu_genres(genres))
+        ]
+        if updates:
+            conn.execute(
+                sa.text(f"UPDATE {table} SET genres_hu = CAST(:g AS jsonb) WHERE id = :id"), updates
+            )
+
+
+def downgrade() -> None:
+    pass  # derived from `genres`; the sync writes the same values again

backend/alembic/versions/0095_videos_channel_cover_index.py

@@ -0,0 +1,83 @@
+"""videos: the per-channel index covers what the channel manager and the feed sidebar read (E31 S4d)
+
+The channel manager (`GET /api/channels`) sums each subscribed channel's videos: count, last upload, total
+duration. The feed sidebar (`/feed/summary`, `/feed/years`) counts the reader's videos per channel and per
+year under the default filters: available, not via explore, not a short, not live. Every one of them found
+its rows through `ix_videos_channel_pub_desc` (channel_id, published_at DESC NULLS LAST, id) and then read
+each video's heap page for the rest: ~270k rows at ~2.5 KB each, 681 MB on prod. That stays in cache only
+while nothing else needs the memory. Measured on prod (2026-10-02): `/api/channels` 51–61 ms warm, but
+2.25 s and 1.96 s on a first open in the browser and 6.4 s under EXPLAIN with 30k blocks read from disk;
+`/feed/summary` and `/feed/years` ~140 ms warm, ~800 ms cold.
+
+The same key with those columns INCLUDEd makes them Index Only Scans. On a prod-copy localdev, after a
+VACUUM, the manager's sums read 0 heap pages and the summary reads the 25 MB index instead of the heap.
+The visibility map is what lets them skip the heap; prod's autovacuum kept it complete (relallvisible ==
+relpages). The key columns are unchanged, so every reader of the old index (the per-channel feed, the FK
+cascade, the sync upsert, channel detail) rides this one. The old index is dropped.
+tests/test_channel_list_index_only.py holds the queries and this column list together.
+
+CONCURRENTLY (autocommit_block) builds it without blocking a live prod. IF (NOT) EXISTS keeps the drift
+test's downgrade → re-upgrade round trip idempotent. A re-run after a build that failed part-way rebuilds
+the INVALID leftover, and the old index is dropped only once the new one is valid: a re-run that kept the
+leftover and dropped the old one left `videos` with no usable channel_id index (E31 S5 review).
+"""
+
+from alembic import op
+from sqlalchemy import text
+
+revision: str = "0095_videos_channel_cover_index"
+down_revision: str | None = "0094_plex_genres_hu_everywhere"
+branch_labels: str | None = None
+depends_on: str | None = None
+
+
+def _valid(name: str) -> bool | None:
+    """The index's `indisvalid`, or None when there is no index of that name."""
+    row = (
+        op.get_bind()
+        .execute(
+            text(
+                "SELECT i.indisvalid FROM pg_index i JOIN pg_class c ON c.oid = i.indexrelid "
+                "WHERE c.relname = :name"
+            ),
+            {"name": name},
+        )
+        .first()
+    )
+    return None if row is None else bool(row[0])
+
+
+def _build(name: str, ddl: str) -> None:
+    """CREATE INDEX CONCURRENTLY `name` (`ddl` is what follows the name), re-runnable. A build that
+    failed part-way leaves an INVALID index under the name, and `IF NOT EXISTS` would keep it: the
+    planner never uses it, and nothing says so (E31 S5 review). Such a leftover is dropped and built
+    again."""
+    if _valid(name) is False:
+        op.execute(f"DROP INDEX CONCURRENTLY IF EXISTS {name}")
+    op.execute(f"CREATE INDEX CONCURRENTLY IF NOT EXISTS {name} {ddl}")
+
+
+def _swap(build: str, ddl: str, drop: str) -> None:
+    _build(build, ddl)
+    if not _valid(build):
+        raise RuntimeError(f"{build} did not build valid; {drop} is kept")
+    op.execute(f"DROP INDEX CONCURRENTLY IF EXISTS {drop}")
+
+
+def upgrade() -> None:
+    with op.get_context().autocommit_block():
+        _swap(
+            "ix_videos_channel_pub_cover",
+            "ON videos (channel_id, published_at DESC NULLS LAST, id) "
+            "INCLUDE (duration_seconds, unavailable_since, via_explore, is_short, live_status)",
+            "ix_videos_channel_pub_desc",
+        )
+
+
+def downgrade() -> None:
+    with op.get_context().autocommit_block():
+        _swap(
+            "ix_videos_channel_pub_desc",
+            "ON videos (channel_id, published_at DESC NULLS LAST, id ASC)",
+            "ix_videos_channel_pub_cover",
+        )

backend/alembic/versions/0096_videos_sort_indexes.py

@@ -0,0 +1,86 @@
+"""videos: an index per feed sort, so a sorted page stops after its rows (E31 S4e)
+
+The feed's views, duration and title sorts had no index. Every page (the first and each next cursor)
+read the whole videos heap, filtered it and kept a top-N: on prod ~350 ms a page for views/duration,
+550–780 ms for title (the ICU comparison on every row), 2.3 s for page 2 when the copies of one request
+queued. With an index in the exact ORDER BY (`feedquery.order_by`: the key, NULLS LAST, then id) the
+planner walks it and stops after the page. On a prod-copy localdev every plain sort fell from 180–310 ms
+to ~1 ms, deep pages (20) to 3–5 ms, and narrow filters (a tag, a channel, a year, nothing found) stayed
+at 0.3–25 ms.
+
+One index per direction: a backward scan of `DESC NULLS LAST` gives `ASC NULLS FIRST`, not the
+`ASC NULLS LAST` the asc sorts ask for. The title pair is per reader language (`app.collation.ICU`):
+an index only serves an ORDER BY in its own collation.
+
+Cost: ~130 MB (localdev: 10 + 10 + 8 + 8 + 4 × 23). Writes stay cheap: view_count and duration are
+written by the enrich, which rewrites indexed columns already (the revalidation job only touches
+last_checked_at), so no HOT update is lost. A search combined with one of these sorts can now walk the
+index too: a term matching ~2 % of the videos took 130–200 ms instead of 7 (still under the plain sort it
+replaces); rarer and broader terms stay fast. Fencing the index off under a search made broad terms
+(`m`, `the`) take 2 s, so it is left to the planner.
+
+The title indexes are the schema's first ones on an ICU collation: an ICU upgrade in the Postgres image
+reorders them, and `app.collation.heal_drift` rebuilds them at startup when it sees one.
+tests/test_feed_sort_indexes.py holds each sort and its index together.
+
+CONCURRENTLY (autocommit_block) builds them without blocking a live prod; IF (NOT) EXISTS keeps the drift
+test's downgrade → re-upgrade round trip idempotent.
+"""
+
+from alembic import op
+from sqlalchemy import text
+
+revision: str = "0096_videos_sort_indexes"
+down_revision: str | None = "0095_videos_channel_cover_index"
+branch_labels: str | None = None
+depends_on: str | None = None
+
+# name → the indexed ORDER BY (feedquery.sort_keys + order_by: key NULLS LAST, then id).
+INDEXES = {
+    "ix_videos_views_desc": "view_count DESC NULLS LAST, id",
+    "ix_videos_views_asc": "view_count ASC NULLS LAST, id",
+    "ix_videos_duration_desc": "duration_seconds DESC NULLS LAST, id",
+    "ix_videos_duration_asc": "duration_seconds ASC NULLS LAST, id",
+    "ix_videos_title_en": '(lower(title) COLLATE "en-x-icu") ASC NULLS LAST, id',
+    "ix_videos_title_en_desc": '(lower(title) COLLATE "en-x-icu") DESC NULLS LAST, id',
+    "ix_videos_title_hu": '(lower(title) COLLATE "hu-x-icu") ASC NULLS LAST, id',
+    "ix_videos_title_hu_desc": '(lower(title) COLLATE "hu-x-icu") DESC NULLS LAST, id',
+}
+
+
+def _valid(name: str) -> bool | None:
+    """The index's `indisvalid`, or None when there is no index of that name."""
+    row = (
+        op.get_bind()
+        .execute(
+            text(
+                "SELECT i.indisvalid FROM pg_index i JOIN pg_class c ON c.oid = i.indexrelid "
+                "WHERE c.relname = :name"
+            ),
+            {"name": name},
+        )
+        .first()
+    )
+    return None if row is None else bool(row[0])
+
+
+def _build(name: str, ddl: str) -> None:
+    """CREATE INDEX CONCURRENTLY `name` (`ddl` is what follows the name), re-runnable. A build that
+    failed part-way leaves an INVALID index under the name, and `IF NOT EXISTS` would keep it: the
+    planner never uses it, and nothing says so (E31 S5 review). Such a leftover is dropped and built
+    again."""
+    if _valid(name) is False:
+        op.execute(f"DROP INDEX CONCURRENTLY IF EXISTS {name}")
+    op.execute(f"CREATE INDEX CONCURRENTLY IF NOT EXISTS {name} {ddl}")
+
+
+def upgrade() -> None:
+    with op.get_context().autocommit_block():
+        for name, columns in INDEXES.items():
+            _build(name, f"ON videos ({columns})")
+
+
+def downgrade() -> None:
+    with op.get_context().autocommit_block():
+        for name in INDEXES:
+            op.execute(f"DROP INDEX CONCURRENTLY IF EXISTS {name}")

backend/alembic/versions/0097_drop_messages.py

@@ -0,0 +1,69 @@
+"""messages: drop direct messaging (the owner's decision, 2026-10-04)
+
+Chat — end-to-end encrypted direct messages plus the server's welcome note — was never used: prod held
+two welcome notes, no message between people, and one key. Its tables go with its code. The rows are
+not migrated anywhere; the nightly backup holds them.
+
+The downgrade restores the schema as it stood (0026 + 0027), empty: a message can't be brought back
+from a migration, and the E2EE key blobs were only ever readable by their owner's browser anyway.
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+revision: str = "0097_drop_messages"
+down_revision: str | None = "0096_videos_sort_indexes"
+branch_labels: str | None = None
+depends_on: str | None = None
+
+
+def upgrade() -> None:
+    op.drop_table("message_keys")
+    op.drop_index("ix_messages_created_at", table_name="messages")
+    op.drop_index("ix_messages_recipient_id", table_name="messages")
+    op.drop_index("ix_messages_sender_id", table_name="messages")
+    op.drop_table("messages")
+
+
+def downgrade() -> None:
+    op.create_table(
+        "messages",
+        sa.Column("id", sa.Integer(), primary_key=True),
+        sa.Column(
+            "sender_id", sa.Integer(), sa.ForeignKey("users.id", ondelete="CASCADE"), nullable=True
+        ),
+        sa.Column(
+            "recipient_id",
+            sa.Integer(),
+            sa.ForeignKey("users.id", ondelete="CASCADE"),
+            nullable=False,
+        ),
+        sa.Column("body", sa.Text(), nullable=True),
+        sa.Column("read_at", sa.DateTime(timezone=True), nullable=True),
+        sa.Column(
+            "created_at", sa.DateTime(timezone=True), server_default=sa.func.now(), nullable=False
+        ),
+        sa.Column("kind", sa.String(length=16), nullable=False, server_default="user"),
+        sa.Column("ciphertext", sa.Text(), nullable=True),
+        sa.Column("iv", sa.String(length=32), nullable=True),
+    )
+    op.create_index("ix_messages_sender_id", "messages", ["sender_id"])
+    op.create_index("ix_messages_recipient_id", "messages", ["recipient_id"])
+    op.create_index("ix_messages_created_at", "messages", ["created_at"])
+    op.create_table(
+        "message_keys",
+        sa.Column(
+            "user_id",
+            sa.Integer(),
+            sa.ForeignKey("users.id", ondelete="CASCADE"),
+            primary_key=True,
+        ),
+        sa.Column("public_key", sa.Text(), nullable=False),
+        sa.Column("wrapped_private_key", sa.Text(), nullable=False),
+        sa.Column("salt", sa.String(length=64), nullable=False),
+        sa.Column("wrap_iv", sa.String(length=32), nullable=False),
+        sa.Column("key_check", sa.Text(), nullable=False),
+        sa.Column(
+            "created_at", sa.DateTime(timezone=True), server_default=sa.func.now(), nullable=False
+        ),
+    )

backend/alembic/versions/0098_drop_demo_account.py

@@ -0,0 +1,55 @@
+"""demo: drop the shared demo account (the owner's decision, 2026-10-04)
+
+The demo account — one shared user that whitelisted emails entered from the login page, reset on a
+schedule — goes with its code. Prod held the demo user and 10 whitelist rows.
+
+The upgrade deletes the demo user first (its playlists, states and subscriptions cascade; audit and
+quota rows keep their history with the actor nulled, as for any deleted account), then the traces
+that only the demo code could render: its audit rows (whitelist edits, manual resets, and every row
+about the reset job — its runs and its interval changes), the reset job's interval override and its
+completion notices. Nothing is migrated anywhere; the nightly backup holds the rows.
+
+The downgrade restores the schema as it stood (0014), empty: the demo user is created lazily on its
+first sign-in, so an empty `demo_whitelist` is the whole of its state.
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+revision: str = "0098_drop_demo_account"
+down_revision: str | None = "0097_drop_messages"
+branch_labels: str | None = None
+depends_on: str | None = None
+
+
+def upgrade() -> None:
+    op.execute("DELETE FROM users WHERE is_demo")
+    op.execute(
+        "DELETE FROM audit_log WHERE action IN ('demo_add', 'demo_remove', 'demo_reset')"
+        " OR (target_type = 'job' AND target_id = 'demo_reset')"
+    )
+    op.execute("DELETE FROM scheduler_settings WHERE job_id = 'demo_reset'")
+    op.execute("DELETE FROM notifications WHERE data->>'job_id' = 'demo_reset'")
+    op.drop_index("ix_demo_whitelist_email", table_name="demo_whitelist")
+    op.drop_table("demo_whitelist")
+    op.drop_index("ix_users_is_demo", table_name="users")
+    op.drop_column("users", "is_demo")
+
+
+def downgrade() -> None:
+    op.add_column(
+        "users",
+        sa.Column("is_demo", sa.Boolean(), nullable=False, server_default=sa.false()),
+    )
+    op.create_index("ix_users_is_demo", "users", ["is_demo"])
+    op.create_table(
+        "demo_whitelist",
+        sa.Column("id", sa.Integer(), primary_key=True),
+        sa.Column("email", sa.String(length=320), nullable=False),
+        sa.Column("note", sa.Text(), nullable=True),
+        sa.Column("added_by", sa.String(length=320), nullable=True),
+        sa.Column(
+            "created_at", sa.DateTime(timezone=True), server_default=sa.func.now(), nullable=False
+        ),
+    )
+    op.create_index("ix_demo_whitelist_email", "demo_whitelist", ["email"], unique=True)

backend/app/README.md

@@ -1,7 +1,7 @@
 # Backend conventions
 
 The rules below are already followed across `backend/app` — they lived only in a few module
-docstrings (`demo.py`, `userscope.py`, `routes/_common.py`) until this file collected them (E3 S3,
+docstrings (`userscope.py`, `routes/_common.py`) until this file collected them (E3 S3,
 B-core-20). When you add code, match these so the next reader doesn't have to guess "which pattern do
 I copy".
 
@@ -21,9 +21,10 @@ Three layers, by *who* runs the code — not by feature:
    `issues/bundle.py` is the one definition of a bug report's files, for the zip export and the tracker.
 
 3. **Flat `app/*.py` core modules — cross-cutting primitives.** `auth`, `sysconfig`, `quota`, `audit`,
-   `notifications`, `userscope`, `realtime`, `ratelimit`, `progress`, `logs` (the one logging
+   `notifications`, `userscope`, `ratelimit`, `progress`, `logs` (the one logging
    configuration), `logsink` (its `log_events` table), `diagnostics` (a user's consented log
-   level) and `debugsession` (a recording that tags its user's lines, E16), … are single-file primitives many layers import. They depend on nothing above them (a core
+   level), `debugsession` (a recording that tags its user's lines, E16) and `collation` (A→Z in the
+   reader's alphabet, E30), … are single-file primitives many layers import. They depend on nothing above them (a core
    module must never import a `routes/…` module).
 
 **Extraction trigger: the SECOND consumer.** Keep new logic inline in its first caller. The moment a
@@ -71,7 +72,9 @@ mutation it records, so an audit entry can never exist without its change (or vi
 ## The gate
 
 Everything below must be green before a change is done (run via `siftlode check`):
-`ruff` (backend) · `tsc` ×3 + `eslint` + `prettier` + `knip` (frontend) · `vitest` · `pytest`. The
-pytest lane runs in `backend/docker-compose.test.yml` — an ephemeral Postgres + the backend tree
-bind-mounted — so the DB-backed integration tests (the `db` fixture) actually run; `SIFTLODE_TEST_
-REQUIRE_DB=1` there turns a would-be silent skip into a hard failure. See `docs/development.md`.
+`ruff` + `basedpyright` (backend) · the lock check + `tsc` ×3 + `eslint` + `prettier` + `knip` +
+`vite build` (frontend) · `vitest` · `pytest`. The pytest lane runs in `backend/docker-compose.test.yml`
+— an ephemeral Postgres + the backend tree bind-mounted — so the DB-backed integration tests (the `db`
+fixture) actually run; a missing database is a hard failure, not a silent skip, unless a hand run asks
+for pure-logic only with `SIFTLODE_TEST_REQUIRE_DB=0`. See `docs/development.md`, and `docs/gates.md`
+for what each gate deliberately leaves out.

backend/app/audit.py

@@ -27,7 +27,7 @@ class AuditAction:
     USER_SUSPEND = "user_suspend"
     USER_UNSUSPEND = "user_unsuspend"
     USER_DELETE = "user_delete"
-    # E9.5 S5: an admin let a stuck registrant in, or re-sent their confirmation link. Both are
+    # An admin let a stuck registrant in, or re-sent their confirmation link. Both are
     # recorded because both are the admin acting ON someone's access, not merely reading a list.
     USER_ACTIVATE = "user_activate"
     USER_VERIFY_RESEND = "user_verify_resend"
@@ -36,12 +36,9 @@ class AuditAction:
     INVITE_DENY = "invite_deny"
     INVITE_ADD = "invite_add"
     INVITE_DELETE = "invite_delete"
-    # E9.5 S6: the admin chose to tell someone their address is whitelisted. Adding the address is
+    # The admin chose to tell someone their address is whitelisted. Adding the address is
     # silent, so this is the act that actually reaches a person — and the one worth a record.
     INVITE_SEND = "invite_send"
-    DEMO_ADD = "demo_add"
-    DEMO_REMOVE = "demo_remove"
-    DEMO_RESET = "demo_reset"
     DISCOVERY_PURGE = "discovery_purge"
     # Config
     CONFIG_SET = "config_set"
@@ -58,7 +55,7 @@ class AuditAction:
     DOWNLOAD_CREDENTIAL_DELETE = "download_credential_delete"
     # The log itself was cleared (a single row survives the clear as tamper-evidence)
     AUDIT_CLEAR = "audit_clear"
-    # E16: a user sent a debug session's issue to the tracker — the one act that takes their log
+    # A user sent a debug session's issue to the tracker — the one act that takes their log
     # off this instance, so it is recorded (target = the session, summary = the issue's URL).
     ISSUE_CREATE = "issue_create"
 

backend/app/auth.py

@@ -24,8 +24,6 @@ def operator_contact() -> str | None:
     return next(iter(sorted(settings.admin_email_set)), None)
 
 
-DEMO_GOOGLE_SUB = "__demo__"
-DEMO_EMAIL = "demo@siftlode.local"
 MAX_SESSION_ACCOUNTS = 6
 
 # OAuth scope vocabulary — the atoms + the request bundles, kept together as core. has_*_scope check
@@ -69,7 +67,7 @@ def has_write_scope(user: User) -> bool:
 def env_allowed_canonicals() -> set[str]:
     """The env bootstrap allow-list, in CANONICAL form — the only shape it may be compared in.
 
-    S3 canonicalised the Invite branch of `is_allowed` and left this one matching the literal string,
+    Canonicalising only the Invite branch of `is_allowed` left this one matching the literal string,
     which is the 2026-09-03 lockout with a different source of truth: an operator who writes
     `first.last@gmail.com` into ALLOWED_EMAILS is refused when Google answers with the undotted
     primary, and a `+tag`ged entry can never match at all, because a Google account cannot contain a
@@ -129,7 +127,7 @@ def can_activate(db: Session, user: User) -> bool:
     Lives in core so the admin surface and the /api/me badge count answer from ONE definition; a
     button that appears and a number that counts must never disagree about who is stuck.
     """
-    if user.is_demo or user.is_suspended or user.is_active:
+    if user.is_suspended or user.is_active:
         return False
     return is_allowed(db, user.email)
 
@@ -147,7 +145,6 @@ def count_pending_activations(db: Session) -> int:
     rows = (
         db.execute(
             select(User).where(
-                User.is_demo.is_(False),
                 User.is_suspended.is_(False),
                 User.is_active.is_(False),
             )
@@ -188,24 +185,6 @@ def find_user_by_email(db: Session, email: str) -> User | None:
     return candidates[0] if len(candidates) == 1 else None
 
 
-def get_or_create_demo_user(db: Session) -> User:
-    """The one shared demo user, created on first use. No OAuth token / YouTube scope, so
-    has_read_scope/has_write_scope are False and every YouTube-touching path is closed to it."""
-    user = db.query(User).filter(User.is_demo.is_(True)).one_or_none()
-    if user is None:
-        user = User(
-            google_sub=DEMO_GOOGLE_SUB,
-            email=DEMO_EMAIL,
-            display_name="Demo",
-            role="user",
-            is_demo=True,
-            preferences={"language": "en"},
-        )
-        db.add(user)
-        db.commit()
-    return user
-
-
 def revoke_google_token(token: str | None) -> None:
     """Best-effort revocation of a Google OAuth grant at Google's endpoint. Revoking the refresh
     token tears down the whole grant, so a later sign-in starts from a clean consent (no scopes
@@ -278,8 +257,7 @@ ACTIVE_ACCOUNT_HEADER = "x-siftlode-account"
 
 def resolved_user_id(conn: HTTPConnection, *, strict: bool = True) -> tuple[int | None, bool]:
     """Which account this connection acts as, and whether that's the session's default account.
-    Takes any HTTPConnection — an HTTP Request OR a WebSocket — so the WS push channel shares this
-    exact per-tab resolution instead of re-implementing it.
+    Takes an HTTPConnection rather than a Request: it reads only the session and the headers.
 
     The signed cookie holds the *wallet* (`account_ids`, every account signed into this browser)
     plus a default `user_id`. A caller may name its account with the X-Siftlode-Account header (or
@@ -297,8 +275,8 @@ def resolved_user_id(conn: HTTPConnection, *, strict: bool = True) -> tuple[int
     """
     default_id = conn.session.get("user_id")
     wallet = conn.session.get("account_ids") or []
-    # Header for normal XHR; ?account= for contexts that can't set headers (WebSocket, and a
-    # plain <a> file download). Both are wallet-gated below.
+    # Header for normal XHR; ?account= for a context that can't set headers (a plain <a> file
+    # download). Both are wallet-gated below.
     hdr = conn.headers.get(ACTIVE_ACCOUNT_HEADER) or conn.query_params.get("account")
     if hdr:
         try:
@@ -362,12 +340,12 @@ def _session_user(request: Request, db: Session, user_id: int | None, is_default
     if default_id and default_id not in accounts:
         accounts.append(default_id)
         request.session["account_ids"] = accounts[-MAX_SESSION_ACCOUNTS:]
-    # Attribute this request's log lines to the account (E22). Here, not in the request middleware:
+    # Attribute this request's log lines to the account. Here, not in the request middleware:
     # this is the one place a session becomes an account, and the middleware runs outside both the
     # session middleware and this task's context. `request.state` is shared across middleware layers
     # (it lives on the ASGI scope), so the access-log line can read it back. The user's consented
-    # diagnostics level rides along: their requests are logged — and stored — in that detail (S5).
-    # So does a debug session the request names (E16): only the caller's own open one counts, and a
+    # diagnostics level rides along: their requests are logged — and stored — in that detail.
+    # So does a debug session the request names: only the caller's own open one counts, and a
     # header naming anything else is ignored rather than refused — a stale tab must keep working.
     session = debugsession.from_header(db, user, request.headers.get(debugsession.HEADER))
     level, until = debugsession.level_for(user, session)
@@ -403,25 +381,10 @@ def optional_current_user(request: Request, db: Session = Depends(get_db)) -> Us
         return None
 
 
-def require_human(user: User = Depends(current_user)) -> User:
-    """Reject the shared demo account from actions that need a real Google/YouTube identity
-    or spend the shared quota. Most YouTube paths are already closed to it implicitly (it has
-    no token, so has_read_scope/has_write_scope are False); this makes the boundary explicit
-    where there's no scope check to lean on."""
-    if user.is_demo:
-        # Structured, because `req` leaves 403 to the caller and `notifyActionError` only shows a
-        # STRUCTURED detail — a plain string here reached the demo user as their caller's generic
-        # "couldn't do that", with no hint that the demo account is the reason (round 6). Affordances
-        # this gate sits behind are not all UI-gated on `is_demo` (the comfort shelf is not).
-        raise HTTPException(status_code=403, detail={"code": "demo"})
-    return user
-
-
-def require_write_scope(user: User = Depends(require_human)) -> User:
+def require_write_scope(user: User = Depends(current_user)) -> User:
     """Dependency: require the YouTube write scope (subscribe/unsubscribe, playlist export/edit).
     Single source of truth for the write gate — replaces the inline `if not has_write_scope(user)`
-    checks that had drifted into three different 403 messages across channels/playlists. Layers on
-    `require_human` because writing to YouTube needs a real, non-demo identity."""
+    checks that had drifted into three different 403 messages across channels/playlists."""
     if not has_write_scope(user):
         raise refused(403, "youtube_editing_off")
     return user

backend/app/bodylimit.py

@@ -59,9 +59,9 @@ DEFAULT_MAX_BODY_BYTES = 1024 * 1024
 # a say.
 #
 # Both listed routes now satisfy that. The screenshot route always did (`capped_body` streams it,
-# and the owner check resolves first). The upload route did NOT until P1: it declared
+# and the owner check resolves first). The upload route did NOT at first: it declared
 # `file: UploadFile = File(...)`, so an anonymous caller could make the process write the whole
-# per-file allowance to temp before `require_human` refused them — this middleware bounded that
+# per-file allowance to temp before the auth dependency refused them — this middleware bounded that
 # number but could not stop the writing. It now reads its own body from a dependency declared after
 # the auth one, so its in-handler cap measures bytes as they arrive, like everything else here.
 #
@@ -73,12 +73,12 @@ DEFAULT_MAX_BODY_BYTES = 1024 * 1024
 # `POST /api/diagnostics/events` is absent: its `capped_body` limit is 512 KiB, already below the
 # default, so listing it could only widen what it accepts.
 ROUTE_MAX_BODY_BYTES: dict[tuple[str, str], int] = {
-    # Ingesting a local file. The body IS the file since P1 — no multipart envelope to allow for —
+    # Ingesting a local file. The body IS the file now — no multipart envelope to allow for —
     # so this ceiling and the route's own cap count exactly the same bytes. `sysconfig`'s
     # `download_upload_max_bytes` is bounded by this same constant (see sysconfig.SPECS), so an
     # admin can only ever LOWER the route's cap below the edge, never raise it past one.
     #
-    # WHICH LAYER ANSWERS, measured rather than assumed (deep-review round 1):
+    # WHICH LAYER ANSWERS, measured rather than assumed:
     #   * knob LOWERED (an override below the default) — the route's cap is the smaller number, so
     #     `capped_stream_to` trips first and answers 422 `upload_too_large` carrying the admin's own
     #     figure, after draining the rest of the body so the client can read it.
@@ -219,7 +219,7 @@ class BodyLimitMiddleware:
 
         try:
             await self.app(scope, counting_receive, guarded_send)
-        except Exception:  # noqa: BLE001 — only swallowed when WE caused it, see below
+        except Exception:  # only swallowed when WE caused it, see below
             # ClientDisconnect (or whatever a handler raises on a truncated read) is the expected
             # consequence of the disconnect we just faked, and the caller must get the 413 rather
             # than a 500. Anything raised when we did NOT refuse is a real failure and is re-raised

backend/app/collation.py

@@ -0,0 +1,83 @@
+"""Alphabetical order in the reader's language.
+
+Why not the database default: the Postgres image is Alpine, and musl's `en_US.utf8` collation is BYTE
+order — every accented initial sorted after "Z" ("Zorró < Életrajzi < Ötödik"), in every A→Z list the
+server orders. Postgres ships the ICU collations on the same image (dev, test, prod alike), so a text
+sort names one: Hungarian puts "Ö" after "O" and "Cs" after "Cu", English folds the accents in.
+
+`collated` wraps an ORDER BY expression — and because a keyset cursor compares with the SAME expression
+(`feedquery.sort_keys`), the cursor pages in the same order it sorted in. `sorted_words` orders a Python
+list the same way through the database, so a list built in Python matches one built in SQL exactly.
+
+The feed's title sorts ride indexes built in these collations (migration 0096), and an index keeps the
+order of the ICU that built it. `postgres:16-alpine` is a floating tag: an image update can bring a newer
+ICU, after which those indexes disagree with the comparisons a query makes (a keyset page skips or
+repeats a title). `heal_drift` rebuilds them at startup once Postgres reports the version moved."""
+
+from typing import Any
+
+from sqlalchemy import Engine, SQLColumnExpression, column, func, select, text, values
+from sqlalchemy.orm import Session
+from sqlalchemy.sql.elements import ColumnElement
+from sqlalchemy.types import String
+
+from app import logs
+from app.mailtext import normalize
+
+log = logs.get_logger(__name__)
+
+ICU: dict[str, str] = {"en": "en-x-icu", "hu": "hu-x-icu"}
+
+
+def collation(lang: str | None) -> str:
+    """The ICU collation for a UI language; anything unknown sorts as English."""
+    return ICU[normalize(lang)]
+
+
+def collated(expr: SQLColumnExpression[Any], lang: str | None) -> ColumnElement[Any]:
+    """`expr` compared in the reader's alphabet. Case-folded, so "apple" and "Apple" sit together."""
+    return func.lower(expr).collate(collation(lang))
+
+
+def sorted_words(db: Session, words: list[str], lang: str | None) -> list[str]:
+    """`words` in the reader's alphabetical order, by the same ICU collation as `collated`."""
+    if not words:
+        return []
+    w = values(column("w", String), name="words").data([(x,) for x in words])
+    return list(db.scalars(select(w.c.w).order_by(collated(w.c.w, lang), w.c.w)))
+
+
+def heal_drift(engine: Engine) -> list[str]:
+    """Rebuild every index on an ICU collation whose library version moved, then record the new version.
+    Returns the rebuilt indexes. CONCURRENTLY: the feed keeps reading while it runs."""
+    with engine.connect().execution_options(isolation_level="AUTOCOMMIT") as conn:
+        drifted = conn.execute(
+            text(
+                "SELECT oid, collname FROM pg_collation WHERE collname = ANY(:names) "
+                "AND collversion IS DISTINCT FROM pg_collation_actual_version(oid)"
+            ),
+            {"names": list(ICU.values())},
+        ).all()
+        if not drifted:
+            return []
+        indexes = list(
+            conn.execute(
+                text(
+                    "SELECT DISTINCT d.objid::regclass::text FROM pg_depend d "
+                    "JOIN pg_class c ON c.oid = d.objid AND c.relkind = 'i' "
+                    "WHERE d.classid = 'pg_class'::regclass "
+                    "AND d.refclassid = 'pg_collation'::regclass AND d.refobjid = ANY(:oids)"
+                ),
+                {"oids": [row.oid for row in drifted]},
+            ).scalars()
+        )
+        log.warning(
+            "ICU collation version changed (%s): rebuilding %d indexes",
+            ", ".join(row.collname for row in drifted),
+            len(indexes),
+        )
+        for name in indexes:
+            conn.execute(text(f"REINDEX INDEX CONCURRENTLY {name}"))
+        for row in drifted:
+            conn.execute(text(f'ALTER COLLATION "{row.collname}" REFRESH VERSION'))
+        return indexes

backend/app/config.py

@@ -146,8 +146,6 @@ class Settings(BaseSettings):
     autotag_interval_minutes: int = 30
     subscriptions_resync_minutes: int = 360
     playlist_sync_minutes: int = 360
-    # How often the shared demo account is auto-reset to a clean baseline (communal sandbox).
-    demo_reset_minutes: int = 720
     # --- Maintenance / validation job ---
     # Runs once a day by default (admin-tunable via the Scheduler dashboard). Grace periods
     # are in days because that's the granularity that matters; the rolling re-validation
@@ -244,7 +242,7 @@ class Settings(BaseSettings):
     download_default_max_bytes: int = 5_368_709_120  # 5 GiB
     download_default_max_concurrent: int = 2
     download_default_max_jobs: int = 50
-    # Per-file cap for local uploads (E7). Unlike a download (whose size is unknown until the worker
+    # Per-file cap for local uploads. Unlike a download (whose size is unknown until the worker
     # finishes), an upload's size is known at ingest, so this bounds a single file directly; the
     # user's overall byte footprint is still governed by download_default_max_bytes.
     # ge=1: this number is also the EDGE ceiling (`bodylimit.ROUTE_MAX_BODY_BYTES`), where 0 is a
@@ -260,12 +258,15 @@ class Settings(BaseSettings):
     # Admin audit log: how long to keep rows (days; 0 = keep forever) + how often the GC runs.
     audit_retention_days: int = 180
     audit_gc_minutes: int = 1440
+    # YouTube quota history (quota_events): days kept (0 = forever). A year, so the admin usage view's
+    # longest window (365 days) stays whole; the same daily cleanup job prunes it.
+    quota_retention_days: int = 365
     # On-disk layout: "plex" (Channel/Season YYYY/… [id].ext + .nfo/poster) or "flat".
     download_layout: DownloadLayout = "plex"
     # Default SponsorBlock (skip sponsor segments) for new custom profiles.
     sponsorblock_default: bool = False
 
-    # --- Diagnostics / logging (E22) ---
+    # --- Diagnostics / logging ---
     # BOOTSTRAP level only: the running level is the `log_level` admin knob (sysconfig), applied at
     # startup and on change without a restart. This is what a process logs at before it can reach
     # the database — and the only knob the worker/runner children read, since they are short-lived.
@@ -298,7 +299,7 @@ class Settings(BaseSettings):
     plex_path_map: str = ""
     # Which Plex library section keys are exposed (comma-separated). Empty = all enabled sections.
     plex_libraries: str = ""
-    # HU-overlay mapping (E5 S5): comma-separated `huSectionKey:enSectionKey` pairs. Each named HU
+    # HU-overlay mapping: comma-separated `huSectionKey:enSectionKey` pairs. Each named HU
     # section is synced as a non-browsable overlay whose Hungarian metadata is written onto the EN
     # twin section's rows (matched by guid). Empty = no overlays. E.g. "3:4,5:1".
     plex_overlay_map: str = ""
@@ -308,7 +309,7 @@ class Settings(BaseSettings):
     plex_sync_interval_min: int = 360
     # How often the incremental Plex→Siftlode watch-state reconcile runs, in minutes (Phase C):
     # a cheap history + on-deck pull plus a re-push of any states that failed their immediate push.
-    # 15 min (S2): with Plex now the source of truth (prompt write-back on the ~10s checkpoint +
+    # 15 min: with Plex now the source of truth (prompt write-back on the ~10s checkpoint +
     # Plex-authoritative pull), a faster incremental pull keeps Siftlode close to Plex-side changes.
     plex_watch_sync_interval_min: int = 15
     # How often the FULL watch-state reconcile runs, in minutes (Phase C). A whole-section rescan
@@ -323,7 +324,7 @@ class Settings(BaseSettings):
     # each one is a whole segmented copy of the film on the scratch volume, written faster than
     # realtime. One account holds at most two regardless (plex/stream.py `_MAX_SESSIONS_PER_USER`).
     plex_max_sessions: int = 8
-    # --- Transcode (playable="transcode") re-encode tuning (E6 S1) ---
+    # --- Transcode (playable="transcode") re-encode tuning ---
     # Hardware-encoder policy: "auto" = VAAPI (h264_vaapi) when the iGPU render node is present AND the
     # source codec is hardware-decodable, else libx264 software; "on" = force VAAPI wherever the render
     # node exists (skip the decodability gate); "off" = always software. See plex/stream.py `_select_encode`.
@@ -346,7 +347,7 @@ class Settings(BaseSettings):
     # local disk, so the default is fine there.
     plex_hls_dir: str = ""
 
-    # --- TMDB enrichment (E5 S2; optional — enriches the Plex catalog from TMDB) ---
+    # --- TMDB enrichment (optional — enriches the Plex catalog from TMDB) ---
     # TMDB API key (v3 api_key, 32 hex chars). Server-side only; stored encrypted when set via the
     # admin Config page. Empty → the TMDB enrichment job no-ops. Feeds keywords / crew / budget /
     # revenue / TMDb score / recommendations / backdrops onto the mirrored Plex rows, matched by the
@@ -356,7 +357,7 @@ class Settings(BaseSettings):
     # Metadata language for TMDB fetches (keyword/overview locale). Stored EN by default.
     tmdb_language: str = "en-US"
 
-    # --- Bug reports (E16): where a debug session's issue goes. All three empty = no tracker; the
+    # --- Bug reports: where a debug session's issue goes. All three empty = no tracker; the
     # composer still exports a zip. The repository must be PRIVATE — an issue carries the reporter's
     # log — and app/issues/gitea.py refuses a public one. Runtime knobs (sysconfig, admin UI). ---
     issues_gitea_url: str = ""
@@ -366,6 +367,11 @@ class Settings(BaseSettings):
     tmdb_enrich_interval_min: int = 120
     # Titles fetched per TMDB enrichment run (TMDB is fast + generously rate-limited).
     tmdb_enrich_batch: int = 100
+    # How often the people job fetches a batch of person bios from TMDB and portraits through Plex,
+    # in minutes, and how many of each per run: ~18k linked people at 300 per 10 min is about ten hours
+    # once, then only the newcomers.
+    people_enrich_interval_min: int = 10
+    people_enrich_batch: int = 300
 
     @property
     def allowed_email_set(self) -> set[str]:

backend/app/debugsession.py

@@ -1,4 +1,4 @@
-"""Debug sessions (E16): Start → reproduce → Stop, and the evidence a bug report is built from.
+"""Debug sessions: Start → reproduce → Stop, and the evidence a bug report is built from.
 
 While a session is open its user's browser names it in `X-Debug-Session` on every request. The auth
 dependency accepts the header only for the caller's OWN open session (anything else is ignored, never
@@ -269,7 +269,7 @@ def set_delivery(db: Session, session: DebugSession, state: dict, token: str) ->
     every step. It is stamped from `token` here instead, and only after the ROW confirms the claim
     is still ours. Without that check the write is a full-column overwrite with no WHERE: a send
     that lost its lease came back and stomped the take-over send's committed links, then put its
-    own dead token back on an idle session (deep-review round 4). `release_send` has always made
+    own dead token back on an idle session. `release_send` has always made
     the same check; this is the other half of it."""
     if claim_owner(db, session) != token:
         raise SendClaimLost(session.id)

backend/app/demo.py

@@ -1,82 +0,0 @@
-"""Demo-account service: reset the shared demo sandbox to a clean, seeded baseline.
-
-Shared by the admin's manual reset endpoint (routes/admin.py) and the scheduled demo-reset job
-(scheduler.py). Lives in the service layer, not routes, so the scheduler can call it without an
-`app.routes -> app.scheduler` import cycle (the reason its import used to be lazy).
-"""
-
-from sqlalchemy import delete, func, select
-from sqlalchemy.orm import Session
-
-from app.models import Playlist, PlaylistItem, Subscription, User, Video, VideoState
-
-
-def _seed_demo_playlists(db: Session, demo: User) -> int:
-    """Seed a few sample playlists from the shared catalog so a freshly-reset demo isn't
-    empty — gives visitors something to play with. Best-effort: skips silently if the
-    catalog is too small."""
-    recent = (
-        db.execute(
-            select(Video.id)
-            .where(Video.is_short.is_(False), Video.live_status == "none")
-            .order_by(Video.published_at.desc().nullslast())
-            .limit(60)
-        )
-        .scalars()
-        .all()
-    )
-    if not recent:
-        return 0
-    samples = [
-        ("Fresh picks", recent[0:10]),
-        ("Quick watch", recent[10:18]),
-        ("Saved for later", recent[18:26]),
-    ]
-    created = 0
-    for pos, (name, vids) in enumerate(samples):
-        if not vids:
-            continue
-        pl = Playlist(user_id=demo.id, name=name, position=pos)
-        db.add(pl)
-        db.flush()
-        for ipos, vid in enumerate(vids):
-            db.add(PlaylistItem(playlist_id=pl.id, video_id=vid, position=ipos))
-        created += 1
-    return created
-
-
-def _seed_demo_subscriptions(db: Session, demo: User, n: int = 14) -> int:
-    """Seed the demo with a handful of catalog channels (the ones with the most videos) so its
-    Channel manager isn't empty — gives visitors something to explore and matches the landing-page
-    screenshot. Idempotent: clears the demo's subscriptions first, then re-adds the top N."""
-    db.execute(delete(Subscription).where(Subscription.user_id == demo.id))
-    top = (
-        db.execute(
-            select(Video.channel_id)
-            .where(Video.channel_id.is_not(None))
-            .group_by(Video.channel_id)
-            .order_by(func.count().desc())
-            .limit(n)
-        )
-        .scalars()
-        .all()
-    )
-    for channel_id in top:
-        db.add(Subscription(user_id=demo.id, channel_id=channel_id))
-    return len(top)
-
-
-def reset_demo_state(db: Session, demo: User) -> int:
-    """Wipe the demo account's per-user state (watch/save/hide states, playlists, preferences)
-    back to a clean baseline and re-seed a few sample playlists + channel subscriptions. Returns
-    the seeded playlist count. Shared by the admin's manual reset button and the scheduled demo
-    reset (see scheduler)."""
-    db.execute(delete(VideoState).where(VideoState.user_id == demo.id))
-    # Playlist items cascade on playlist delete (ondelete="CASCADE").
-    db.execute(delete(Playlist).where(Playlist.user_id == demo.id))
-    demo.preferences = {"language": "en"}
-    db.commit()
-    seeded = _seed_demo_playlists(db, demo)
-    _seed_demo_subscriptions(db, demo)
-    db.commit()
-    return seeded

backend/app/diagnostics.py

@@ -1,4 +1,4 @@
-"""A user's diagnostics level (E22 S5): what they consented to, and for how long.
+"""A user's diagnostics level: what they consented to, and for how long.
 
 Three levels, named in the UI Alap / Részletes / Teljes:
 
@@ -10,7 +10,6 @@ Three levels, named in the UI Alap / Részletes / Teljes:
 
 A raised level always lapses after `ELEVATION_HOURS`: the expiry is computed HERE at consent time,
 never taken from the client, and an expired level reads as INFO without anything having to clear it.
-The shared demo account cannot raise its level — its log would mix every visitor's session.
 
 Core layer: imports `models` and `sysconfig`.
 """
@@ -27,7 +26,7 @@ DiagnosticsLevel = Literal["INFO", "DEBUG", "TRACE"]
 LEVELS: tuple[str, ...] = get_args(DiagnosticsLevel)
 BASE: DiagnosticsLevel = "INFO"
 ELEVATION_HOURS = 24
-# How long a debug session records at most (E16, app/debugsession.py). Here, beside the other
+# How long a debug session records at most (app/debugsession.py). Here, beside the other
 # consent window, because the start dialog's consent text states it and reads it from `state`.
 SESSION_MINUTES = 60
 

backend/app/discover/__init__.py

@@ -1,4 +1,4 @@
-"""E24 — the read-model behind the Home surface ("Mit nézzek most?").
+"""The read-model behind the Home surface ("Mit nézzek most?").
 
 The anti-paralysis answer layer: one hero suggestion, a cross-source resume rail, and the user's own
 comfort shelf, all served from our own Postgres with no upstream call at serve time (the same rule

backend/app/discover/budget.py

@@ -1,4 +1,4 @@
-"""The time budget — E24's replacement for "what mood are you in?".
+"""The time budget — the replacement for "what mood are you in?".
 
 `random-gondolatok` 15/D asked for a mood questionnaire before the recommendations. The same file's
 later BA-interview note (#23) says the mood half is not what makes this work: the friction is

backend/app/discover/hero.py

@@ -48,6 +48,7 @@ def pick(
     shelf: list[dict] | None = None,
     exclude: set[str] | None = None,
     start: int = 0,
+    lang: str,
 ) -> dict | None:
     """`{"tier": ..., "card": ...}`, or None when nothing at all fits.
 
@@ -70,8 +71,8 @@ def pick(
     skip = exclude or set()
 
     def _resume() -> dict | None:
-        for card in rails.continue_watching(db, user) if resume is None else resume:
-            playable = rails.playable(db, user.id, card)
+        for card in rails.continue_watching(db, user, lang=lang) if resume is None else resume:
+            playable = rails.playable(db, user.id, card, lang=lang)
             # Excluded on the card the USER saw (the show), not only on what it resolves to, so a
             # rerolled-past show does not come straight back as its own next episode.
             if cards.key(card) in skip or (playable is not None and cards.key(playable) in skip):
@@ -82,7 +83,7 @@ def pick(
 
     def _comfort() -> dict | None:
         if shelf is None:
-            own = pins.shelf(db, user, include_plex=rails.plex_available(db))
+            own = pins.shelf(db, user, include_plex=rails.plex_available(db), lang=lang)
         else:
             own = shelf
         for card in pins.playables(own):
@@ -93,7 +94,7 @@ def pick(
     def _surprise() -> dict | None:
         # A wider draw than the one card we need: the shuffle is seeded, so without the headroom an
         # excluded pick would leave the tier empty rather than move past it.
-        for card in rails.surprise(db, user, budget, seed, limit=SURPRISE_DRAW):
+        for card in rails.surprise(db, user, budget, seed, limit=SURPRISE_DRAW, lang=lang):
             if cards.key(card) not in skip:
                 return card
         return None

backend/app/discover/pins.py

@@ -46,8 +46,8 @@ def resolve_ref(db: Session, user: User, kind: str, ref: str) -> tuple[str, obje
     translation belongs here rather than in five call sites.
 
     The COLUMN half of every answer comes from `PIN_COLUMNS`; only the LOOKUP differs per kind
-    (a direct `db.get`, a rating-key translation, an int parse plus an ownership check). Before the
-    E24 review the branches spelled the column names out again and `PIN_COLUMNS` had no readers at
+    (a direct `db.get`, a rating-key translation, an int parse plus an ownership check). Before a
+    review the branches spelled the column names out again and `PIN_COLUMNS` had no readers at
     all — a constant that called itself the one vocabulary while nothing spoke it."""
     col = PIN_COLUMNS.get(kind)
     if col is None:
@@ -170,7 +170,7 @@ class _Batch:
         return b
 
 
-def _shelf_card(db: Session, user: User, pin: ComfortPin, batch: _Batch) -> dict | None:
+def _shelf_card(db: Session, user: User, pin: ComfortPin, batch: _Batch, lang: str) -> dict | None:
     """One pin as a unified card with its resolved `next`. None when the pinned row has gone (the FK
     cascade removes the pin itself, so this is only the window between a delete and a refetch)."""
     card: dict | None = None
@@ -196,7 +196,7 @@ def _shelf_card(db: Session, user: User, pin: ComfortPin, batch: _Batch) -> dict
         if it is None:
             return None
         st = rollup.item_state_map(db, user.id, [it.id]).get(it.id)
-        card = cards.from_plex(leaf_card(it, st))
+        card = cards.from_plex(leaf_card(it, st, lang=lang))
         nxt = dict(card)  # a movie or a single episode plays itself
 
     elif pin.plex_show_id is not None:
@@ -204,8 +204,8 @@ def _shelf_card(db: Session, user: User, pin: ComfortPin, batch: _Batch) -> dict
         if sh is None:
             return None
         status = rollup.state_map(db, user.id, [sh.id]).get(sh.id, "new")
-        card = cards.from_plex(show_card(sh, status))
-        ep = queries.next_episode(db, user.id, sh.id)
+        card = cards.from_plex(show_card(sh, status, lang=lang))
+        ep = queries.next_episode(db, user.id, sh.id, lang=lang)
         nxt = cards.from_plex(ep) if ep else None
 
     elif pin.playlist_id is not None:
@@ -229,7 +229,7 @@ def _shelf_card(db: Session, user: User, pin: ComfortPin, batch: _Batch) -> dict
 _PLEX_COLUMNS = ("plex_item_id", "plex_show_id")
 
 
-def shelf(db: Session, user: User, *, include_plex: bool = True) -> list[dict]:
+def shelf(db: Session, user: User, *, include_plex: bool = True, lang: str) -> list[dict]:
     """The whole shelf, in the user's own order.
 
     `include_plex=False` is the Home surface with the deployment's Plex toggle off (the owner's call,
@@ -243,7 +243,7 @@ def shelf(db: Session, user: User, *, include_plex: bool = True) -> list[dict]:
         pins = [p for p in pins if all(getattr(p, c) is None for c in _PLEX_COLUMNS)]
     batch = _Batch.load(db, user, pins)
     for pin in pins:
-        card = _shelf_card(db, user, pin, batch)
+        card = _shelf_card(db, user, pin, batch, lang)
         if card is not None:
             out.append(card)
     return out

backend/app/discover/queries.py

@@ -40,7 +40,7 @@ from app.models import (
     VideoState,
 )
 from app.plex import rollup
-from app.plex.cards import leaf_card, movie_card, show_card
+from app.plex.cards import display_title, leaf_card, movie_card, show_card
 from app.plex.query import apply_show_state_filter, content_libs
 
 # How many rows a single rail may carry — Home's rails and the Plex module's Discover rails alike
@@ -221,7 +221,9 @@ def _unseen_movie_conds(uid: int):
     return ~seen
 
 
-def plex_in_progress(db: Session, uid: int, limit: int = RAIL_LIMIT) -> list[tuple[object, dict]]:
+def plex_in_progress(
+    db: Session, uid: int, limit: int = RAIL_LIMIT, *, lang: str
+) -> list[tuple[object, dict]]:
     """In-progress movies + partially-watched shows as (progressed_at, card).
 
     Mirrors `routes/plex/discover.py::_continue_watching` — same `PlexState` predicate for movies (a
@@ -249,7 +251,7 @@ def plex_in_progress(db: Session, uid: int, limit: int = RAIL_LIMIT) -> list[tup
             .limit(limit)
         )
         for it, st in db.execute(mq).all():
-            out.append((st.progress_updated_at, movie_card(it, st)))
+            out.append((st.progress_updated_at, movie_card(it, st, lang=lang)))
 
     if show_libs:
         prog = rollup.agg_subquery(db, uid, include_resumed=True).subquery()
@@ -263,12 +265,14 @@ def plex_in_progress(db: Session, uid: int, limit: int = RAIL_LIMIT) -> list[tup
         for sh, resumed_at in db.execute(
             sq.order_by(prog.c.resumed_at.desc().nullslast()).limit(limit)
         ).all():
-            out.append((resumed_at, show_card(sh, "in_progress")))
+            out.append((resumed_at, show_card(sh, "in_progress", lang=lang)))
 
     return out
 
 
-def plex_surprise(db: Session, uid: int, budget: str, seed: int, limit: int) -> list[dict]:
+def plex_surprise(
+    db: Session, uid: int, budget: str, seed: int, limit: int, *, lang: str
+) -> list[dict]:
     """Unseen playable movies inside the budget, in seeded-shuffle order.
 
     Movies only. A show is not a thing you can start playing, and resolving every candidate show to
@@ -300,14 +304,14 @@ def plex_surprise(db: Session, uid: int, budget: str, seed: int, limit: int) ->
         q = q.where(PlexItem.duration_s <= hi)
     items = db.execute(q).scalars().all()
     states = rollup.item_state_map(db, uid, [it.id for it in items])
-    return [movie_card(it, states.get(it.id)) for it in items]
+    return [movie_card(it, states.get(it.id), lang=lang) for it in items]
 
 
 # Watch states that take an episode out of the running for "what would play next".
 _DONE = ("watched", "hidden")
 
 
-def next_episode(db: Session, uid: int, show_id: int) -> dict | None:
+def next_episode(db: Session, uid: int, show_id: int, *, lang: str) -> dict | None:
     """What "put that show on" means: the episode in progress, else the first unwatched one, else the
     first — over the same `(season, episode)` broadcast ordering `detail.py::_episode_nav` uses.
 
@@ -357,5 +361,9 @@ def next_episode(db: Session, uid: int, show_id: int) -> dict | None:
     ep, st = pick
     show = db.get(PlexShow, show_id)
     return leaf_card(
-        ep, st, show_title=show.title if show else None, show_rk=show.rating_key if show else None
+        ep,
+        st,
+        show_title=display_title(show, lang) if show else None,
+        show_rk=show.rating_key if show else None,
+        lang=lang,
     )

backend/app/discover/rails.py

@@ -22,7 +22,7 @@ def plex_available(db: Session) -> bool:
     return sysconfig.get_bool(db, "plex_enabled")
 
 
-def playable(db: Session, uid: int, card: dict) -> dict | None:
+def playable(db: Session, uid: int, card: dict, *, lang: str) -> dict | None:
     """The thing that would actually START from this card — itself for anything with a runtime, and
     for a Plex SHOW the episode that is next (in progress, else the first unwatched, else the first).
 
@@ -40,11 +40,11 @@ def playable(db: Session, uid: int, card: dict) -> dict | None:
     sh = by_rating_key(db, PlexShow, card["id"])
     if sh is None:
         return None
-    ep = queries.next_episode(db, uid, sh.id)
+    ep = queries.next_episode(db, uid, sh.id, lang=lang)
     return cards.from_plex(ep) if ep else None
 
 
-def continue_watching(db: Session, user: User, limit: int = RAIL_LIMIT) -> list[dict]:
+def continue_watching(db: Session, user: User, limit: int = RAIL_LIMIT, *, lang: str) -> list[dict]:
     """One resume rail across both sources, most recently progressed first.
 
     A Plex SHOW enters as a show CARD — that is what the Plex rail shows, and it is the artwork the
@@ -57,17 +57,22 @@ def continue_watching(db: Session, user: User, limit: int = RAIL_LIMIT) -> list[
         (ts, cards.from_video(r)) for ts, r in queries.yt_in_progress(db, user, limit)
     ]
     if plex_available(db):
-        rows += [(ts, cards.from_plex(c)) for ts, c in queries.plex_in_progress(db, user.id, limit)]
+        rows += [
+            (ts, cards.from_plex(c))
+            for ts, c in queries.plex_in_progress(db, user.id, limit, lang=lang)
+        ]
     rows.sort(key=lambda r: r[0] or NO_TS, reverse=True)
     out = [card for _, card in rows[:limit]]
     # Resolve AFTER the slice: only the rows that survive it are worth a query.
     for card in out:
         if card["kind"] == "show":
-            card["next"] = playable(db, user.id, card)
+            card["next"] = playable(db, user.id, card, lang=lang)
     return out
 
 
-def surprise(db: Session, user: User, budget: str, seed: int, limit: int = 3) -> list[dict]:
+def surprise(
+    db: Session, user: User, budget: str, seed: int, limit: int = 3, *, lang: str
+) -> list[dict]:
     """The hero's draw pool: unseen things that fit the budget, from both sources under ONE seed.
 
     Interleaved rather than concatenated — the same alternation `routes/plex/discover.py` uses for its
@@ -83,7 +88,9 @@ def surprise(db: Session, user: User, budget: str, seed: int, limit: int = 3) ->
         (cards.from_video, queries.yt_surprise(db, user, budget, seed, limit)),
         (
             cards.from_plex,
-            queries.plex_surprise(db, user.id, budget, seed, limit) if plex_available(db) else [],
+            queries.plex_surprise(db, user.id, budget, seed, limit, lang=lang)
+            if plex_available(db)
+            else [],
         ),
     ]
     if seed % 2:

backend/app/downloads/adapters.py

@@ -156,7 +156,7 @@ class YouTubeAdapter(SourceAdapter):
 
 class FacebookAdapter(SourceAdapter):
     key = "facebook"
-    # Measured (E11.5, yt-dlp 2026.07.04): with the user's cookie, some reels answer a plain request
+    # Measured (yt-dlp 2026.07.04): with the user's cookie, some reels answer a plain request
     # with "Cannot parse data" and parse only when the request looks like Chrome. Impersonation is a
     # SECOND attempt behind that one error, not the default: the links that already download with the
     # cookie keep their proven path, and only the ones stuck at the wall go through curl_cffi. Without
@@ -166,7 +166,7 @@ class FacebookAdapter(SourceAdapter):
     _POST_PAGE_MAX_BYTES = 16 * 1024 * 1024
 
     def resolve_url(self, url: str, ydl_opts: dict) -> str:
-        # Measured (E11.5 S4): given a group post's link, yt-dlp either finds no formats or — worse —
+        # Measured: given a group post's link, yt-dlp either finds no formats or — worse —
         # downloads ANOTHER video from the page's feed data. The post's own video is the one Video node
         # in the page scripts that carry the post id; its watch link downloads (plain first, then the
         # Chrome retry above). Needs the cookie: signed out, the page is the login wall.
@@ -176,7 +176,7 @@ class FacebookAdapter(SourceAdapter):
         with yt_dlp.YoutubeDL(ydl_opts) as ydl:
             raw = ydl.urlopen(url).read(self._POST_PAGE_MAX_BYTES)
         page = raw.decode("utf-8", "replace")
-        # Measured (E11.5 review): a refused or expired cookie gets the login page (~450 KB, a login
+        # Measured: a refused or expired cookie gets the login page (~450 KB, a login
         # form, no post id at all), which is not a post without a video — the user's cookie is the cause.
         if _quoted(post_id) not in page:
             raise StatedFailure(POST_NOT_SHOWN_MESSAGE)

backend/app/downloads/credentials.py

@@ -249,8 +249,8 @@ def usable_cookie(
 
     If (2) fails the asset is the SHARED (unsalted) one — attaching the cookie would poison the
     global cache with a private-group video, so we attach nothing. An unreadable blob (rotated key →
-    InvalidToken) also yields None; the download then fails honestly on the login wall (S3's
-    `login_required`)."""
+    InvalidToken) also yields None; the download then fails honestly on the login wall (the
+    `login_required` category)."""
     host_key = host_key_for_url(url)
     if host_key is None:
         return None

backend/app/downloads/edit.py

@@ -188,6 +188,51 @@ def clip_duration(spec: dict, source_duration: int | None) -> int | None:
 
 # --- ffmpeg: trim / crop --------------------------------------------------------------------
 
+# Every libx264 / aac re-encode in this package uses these, so a re-encoded file is always the
+# phone-safe shape (`playable`): `-pix_fmt yuv420p` pins 8-bit 4:2:0 — without it libx264 keeps a
+# 10-bit or 4:4:4 source's pixel format and writes H.264 High 10 / High 4:4:4, which browsers decode
+# only patchily (the one safe H.264 is 8-bit 4:2:0).
+X264_ARGS = ["-c:v", "libx264", "-preset", "veryfast", "-crf", "20", "-pix_fmt", "yuv420p"]
+AAC_ARGS = ["-c:a", "aac", "-b:a", "192k"]
+# The last step of every video filter chain a libx264 re-encode runs: 4:2:0 halves the chroma, so
+# libx264 refuses an odd width or height ("width not divisible by 2") — measured on an odd source
+# that is not 4:2:0 already (a 4:4:4 or RGB screen capture, 1365x767). A crop of a 4:2:0 source is
+# rounded to its chroma grid by ffmpeg itself; this holds every other path to the encoder to the same
+# even frame. Rounding down to even costs at most one pixel row/column.
+EVEN_FRAME = "scale=trunc(iw/2)*2:trunc(ih/2)*2"
+
+# The demuxers ffmpeg/ffprobe may use to open a library file — single-file containers only. Without
+# it ffmpeg picks the demuxer from the file's CONTENT and NAME, and an upload's name is the user's:
+# a playlist-type demuxer (HLS, DASH, concat) opens whatever local paths its entries name, so a
+# crafted upload could copy another user's clip or a Plex file into its own converted asset. A file
+# no demuxer here can open fails the probe, and an unprobeable upload is a download-only 'file'.
+MEDIA_DEMUXERS = (
+    "mov",  # mp4, m4a, mov, 3gp
+    "matroska",  # mkv, webm
+    "mpegts",
+    "mpeg",
+    "avi",
+    "asf",
+    "flv",
+    "ogg",
+    "mp3",
+    "aac",
+    "ac3",
+    "eac3",
+    "flac",
+    "wav",
+    "w64",
+    "aiff",
+    "caf",
+    "amr",
+)
+INPUT_GUARD = ["-format_whitelist", ",".join(MEDIA_DEMUXERS)]
+# A downloaded thumbnail is the remote site's bytes under the suffix yt-dlp gave it (`.png`, `.webp`,
+# `.jpg` — `pipeline._THUMB_SUFFIXES`): only the single-image demuxers those open, so a "thumbnail"
+# that is really a playlist cannot pull a local file into the poster. Anything else fails, and the
+# poster falls back to a cut frame (`ensure_poster`).
+THUMB_GUARD = ["-format_whitelist", "png_pipe,jpeg_pipe,webp_pipe"]
+
 
 def build_edit_cmd(src: Path, dest: Path, spec: dict, out_ext: str) -> list[str]:
     """ffmpeg command for a trim/crop edit. `-progress pipe:1` streams machine-readable progress.
@@ -214,7 +259,7 @@ def build_edit_cmd(src: Path, dest: Path, spec: dict, out_ext: str) -> list[str]
     ]
     if start:
         cmd += ["-ss", f"{start:.3f}"]
-    cmd += ["-i", str(src)]
+    cmd += [*INPUT_GUARD, "-i", str(src)]
     if end is not None:
         dur = end - start
         if dur > 0:
@@ -222,21 +267,9 @@ def build_edit_cmd(src: Path, dest: Path, spec: dict, out_ext: str) -> list[str]
 
     if reencode:
         # Input -ss + re-encode is frame-accurate (decodes from the prior keyframe, discards up to
-        # the exact cut). A crop adds the filter; an accurate trim re-encodes with no filter.
-        if crop:
-            cmd += ["-vf", f"crop={crop['w']}:{crop['h']}:{crop['x']}:{crop['y']}"]
-        cmd += [
-            "-c:v",
-            "libx264",
-            "-preset",
-            "veryfast",
-            "-crf",
-            "20",
-            "-c:a",
-            "aac",
-            "-b:a",
-            "192k",
-        ]
+        # the exact cut). A crop adds the filter; an accurate trim only evens the frame.
+        cropf = f"crop={crop['w']}:{crop['h']}:{crop['x']}:{crop['y']}," if crop else ""
+        cmd += ["-vf", cropf + EVEN_FRAME, *X264_ARGS, *AAC_ARGS]
     else:
         cmd += ["-c", "copy", "-avoid_negative_ts", "make_zero"]
 
@@ -284,11 +317,14 @@ def build_concat_plan(src: Path, dest: Path, spec: dict, out_ext: str, staging:
         parts, labels = [], []
         for i, s in enumerate(segs):
             st, en = s["start_s"], s["end_s"]
-            parts.append(f"[0:v]trim=start={st}:end={en},setpts=PTS-STARTPTS{cropf}[v{i}]")
+            parts.append(
+                f"[0:v]trim=start={st}:end={en},setpts=PTS-STARTPTS{cropf},{EVEN_FRAME}[v{i}]"
+            )
             parts.append(f"[0:a]atrim=start={st}:end={en},asetpts=PTS-STARTPTS[a{i}]")
             labels.append(f"[v{i}][a{i}]")
         parts.append(f"{''.join(labels)}concat=n={len(segs)}:v=1:a=1[v][a]")
         cmd = head + [
+            *INPUT_GUARD,
             "-i",
             str(src),
             "-filter_complex",
@@ -297,26 +333,23 @@ def build_concat_plan(src: Path, dest: Path, spec: dict, out_ext: str, staging:
             "[v]",
             "-map",
             "[a]",
-            "-c:v",
-            "libx264",
-            "-preset",
-            "veryfast",
-            "-crf",
-            "20",
-            "-c:a",
-            "aac",
-            "-b:a",
-            "192k",
+            *X264_ARGS,
+            *AAC_ARGS,
         ]
     else:
         listp = staging / "concat.txt"
-        lines = []
+        lines = ["ffconcat version 1.0"]
         for s in segs:
             lines.append(f"file '{_concat_quote(src)}'")
+            # ffmpeg hands the list's whitelist (which must name `concat` to read the list) down to
+            # the file it names, so a library file whose bytes are a concat list would be read as one
+            # and open the paths inside it. Each entry's own whitelist replaces the inherited one.
+            lines.append(f"option format_whitelist {INPUT_GUARD[1]}")
             lines.append(f"inpoint {s['start_s']}")
             lines.append(f"outpoint {s['end_s']}")
         prep[listp] = "\n".join(lines) + "\n"
-        cmd = head + ["-f", "concat", "-safe", "0", "-i", str(listp), "-c", "copy"]
+        guard = ["-format_whitelist", "concat," + INPUT_GUARD[1]]
+        cmd = head + ["-f", "concat", "-safe", "0", *guard, "-i", str(listp), "-c", "copy"]
 
     if out_ext in ("mp4", "m4a", "mov"):
         cmd += ["-movflags", "+faststart"]
@@ -400,7 +433,7 @@ def escalate_kill(proc: subprocess.Popen, terminate, kill, grace_s: float, label
     try:
         proc.wait(timeout=grace_s)
     except subprocess.TimeoutExpired:
-        # noqa TRY400: the exception IS the message here. A TimeoutExpired's traceback is this
+        # TRY400: the exception IS the message here. A TimeoutExpired's traceback is this
         # `wait` call and nothing else — it would add three frames and no fact the line lacks.
         log.error("%s pid %s survived SIGKILL", label, proc.pid)  # noqa: TRY400
 
@@ -456,7 +489,7 @@ def run_ffmpeg(
         try:
             ret = proc.wait(timeout=_FFMPEG_EXIT_GRACE_S)
         except subprocess.TimeoutExpired:  # pipes closed but the process lingers
-            raise StatedFailure("ffmpeg did not exit after closing its output")
+            raise StatedFailure("ffmpeg did not exit after closing its output") from None
     except BaseException:
         # ONE exit path stops the process, so the invariant is unconditional: nothing leaves this
         # function with ffmpeg still running. Not just our own EditAborted/stall — `on_progress`
@@ -474,14 +507,7 @@ def run_ffmpeg(
         raise StatedFailure(f"ffmpeg failed ({ret}): {''.join(err_tail)[-300:]}")
 
 
-# --- ffmpeg: universal browser playability --------------------------------------------------
-# A downloaded file must play in EVERY browser, whatever the source shipped. Universal = H.264
-# video + AAC/MP3 audio in an mp4 container: H.264 is the only video codec all browsers decode
-# (VP9/AV1 fail on Safari/iOS; HEVC fails on Chrome/Firefox), and mp4/AAC is the safe pairing.
-# Anything else is fixed post-download: re-encode the offending stream(s), or just remux when the
-# codecs are already fine but the container isn't.
-_BROWSER_SAFE_VCODEC = ("h264",)
-_BROWSER_SAFE_ACODEC = ("aac", "mp3")
+# --- ffprobe ---------------------------------------------------------------------------------
 
 
 def run_ffprobe_json(
@@ -493,10 +519,10 @@ def run_ffprobe_json(
 ) -> dict | None:
     """One ffprobe call → its parsed JSON dict, or None on any failure (timeout, non-zero exit,
     missing binary, unparseable output). The SINGLE ffprobe-JSON-invocation point in the downloads
-    package: the browser-playability codec probe here and the shared media probe (`probe.probe_media`,
-    used by both the download worker's finalize and the upload ingest) both go through it, so the
+    package: the shared media probe (`probe.probe_json`: the stored facts and the phone-safe rule) goes
+    through it, so the
     subprocess flags/timeout/error handling live in one place instead of drifting copies."""
-    args = ["ffprobe", "-v", "error"]
+    args = ["ffprobe", "-v", "error", *INPUT_GUARD]
     if show_streams:
         args.append("-show_streams")
     if show_format:
@@ -517,35 +543,6 @@ def run_ffprobe_json(
         return None
 
 
-def _codecs_from_streams(streams: list) -> tuple[str | None, str | None]:
-    """(video_codec, audio_codec) lowercased from an ffprobe JSON stream list — the first of each kind.
-
-    Parsed from the JSON `streams` list, NOT ffprobe's flat `default=nw=1` output: there the fields
-    arrive in ffprobe's own struct order, which puts `codec_name` BEFORE `codec_type`, so a line-by-line
-    "remember the last codec_type" parser attributes every name to the PREVIOUS stream. Measured on real
-    downloads: an ordinary H.264+AAC mp4 probed as `v=aac a=None` — the browser-playability check then
-    believed the video stream was AAC and re-encoded a perfectly fine file on every download. Reading each
-    stream dict by its own `codec_type` is immune to that ordering."""
-    vcodec = acodec = None
-    for s in streams:
-        name = (s.get("codec_name") or "").lower() or None
-        ctype = (s.get("codec_type") or "").lower()
-        if ctype == "video" and vcodec is None:
-            vcodec = name
-        elif ctype == "audio" and acodec is None:
-            acodec = name
-    return vcodec, acodec
-
-
-def probe_codecs(path: Path) -> tuple[str | None, str | None]:
-    """Return (video_codec, audio_codec) lowercased via one ffprobe call; None for a missing
-    stream or if ffprobe fails (a failed probe is treated as 'leave it alone' upstream)."""
-    data = run_ffprobe_json(path, entries="stream=codec_type,codec_name")
-    if data is None:
-        return None, None
-    return _codecs_from_streams(data.get("streams") or [])
-
-
 def probe_duration(path: Path) -> int:
     """Media duration in whole seconds via ffprobe, or 0 if unknown — a progress denominator for a
     re-encode when the yt-dlp info dict carried no `duration` (some URL/live extractions)."""
@@ -555,6 +552,7 @@ def probe_duration(path: Path) -> int:
                 "ffprobe",
                 "-v",
                 "error",
+                *INPUT_GUARD,
                 "-show_entries",
                 "format=duration",
                 "-of",
@@ -577,49 +575,6 @@ def probe_duration(path: Path) -> int:
         return 0
 
 
-def browser_transcode_flags(
-    vcodec: str | None, acodec: str | None, is_mp4: bool
-) -> tuple[bool, bool, bool]:
-    """Decide what (if anything) makes a file universally browser-playable. Returns
-    (needs_change, reencode_video, reencode_audio). needs_change=False => already H.264 + AAC/MP3
-    in mp4. A None codec (no such stream, or a failed probe) is never re-encoded."""
-    rv = bool(vcodec) and vcodec not in _BROWSER_SAFE_VCODEC
-    ra = bool(acodec) and acodec not in _BROWSER_SAFE_ACODEC
-    return (rv or ra or not is_mp4), rv, ra
-
-
-def build_browser_safe_cmd(
-    src: Path, dest: Path, reencode_video: bool, reencode_audio: bool, threads: int | None = None
-) -> list[str]:
-    """ffmpeg command that lands `src` as a browser-safe mp4 (faststart): re-encode only the
-    stream(s) that need it (libx264 / aac), stream-copy the rest, always into mp4. `-progress
-    pipe:1` streams machine-readable progress for run_ffmpeg. `threads` caps the encoder so a
-    re-encode can't claim every core on the shared box (see the worker-child de-prioritization)."""
-    cmd = [
-        "ffmpeg",
-        "-y",
-        "-hide_banner",
-        "-loglevel",
-        "error",
-        "-nostdin",
-        "-progress",
-        "pipe:1",
-        "-nostats",
-        "-i",
-        str(src),
-    ]
-    cmd += (
-        ["-c:v", "libx264", "-preset", "veryfast", "-crf", "20"]
-        if reencode_video
-        else ["-c:v", "copy"]
-    )
-    cmd += ["-c:a", "aac", "-b:a", "192k"] if reencode_audio else ["-c:a", "copy"]
-    if threads and threads > 0:
-        cmd += ["-threads", str(threads)]
-    cmd += ["-movflags", "+faststart", str(dest)]
-    return cmd
-
-
 # --- filmstrip (editor scrub track) ---------------------------------------------------------
 
 _SB_COLS = 12
@@ -655,6 +610,7 @@ def build_storyboard_cmd(src: Path, dest: Path, duration_s: int | None, geom: di
         "-loglevel",
         "error",
         "-nostdin",
+        *INPUT_GUARD,
         "-i",
         str(src),
         "-frames:v",
@@ -678,6 +634,7 @@ def build_poster_cmd(src: Path, dest: Path, at_s: float) -> list[str]:
         "-nostdin",
         "-ss",
         f"{max(0.0, at_s):.3f}",
+        *INPUT_GUARD,
         "-i",
         str(src),
         "-frames:v",
@@ -699,6 +656,7 @@ def build_thumbnail_jpg_cmd(src: Path, dest: Path) -> list[str]:
         "-loglevel",
         "error",
         "-nostdin",
+        *THUMB_GUARD,
         "-i",
         str(src),
         "-frames:v",
@@ -734,7 +692,7 @@ def _step_failure(err: BaseException) -> str:
 
 
 def _is_jpeg(path: Path) -> bool:
-    """True if the file starts with the JPEG SOI marker (FF D8 FF)."""
+    """True if the file starts with the JPEG SOI marker (0xFF 0xD8 0xFF)."""
     try:
         with path.open("rb") as f:
             return f.read(3) == b"\xff\xd8\xff"

backend/app/downloads/errors.py

@@ -8,7 +8,7 @@ explanation (and decides whether a retry makes sense). The raw text stays availa
 "details" disclosure. This is the same string-matching approach the worker already trusts for
 `_RETRIABLE_MARKERS` — kept here as the single source of truth for "what kind of failure is this".
 
-**The failure catalog** (`_CATALOG`, E11.5). A category is generic: "Sign-in required" cannot tell a
+**The failure catalog** (`_CATALOG`). A category is generic: "Sign-in required" cannot tell a
 Facebook user that the cookie they already gave was refused, and yt-dlp words several different
 Facebook walls identically (`Cannot parse data`). A catalog entry is scoped to ONE host and may also
 depend on whether a cookie was attached and on the kind of link, and names a `detail` code the UI has
@@ -49,10 +49,11 @@ from app.downloads.credentials import host_key_for_url
 # "produced no output file" have markers here; the urlguard refusal deliberately has none and
 # classifies as `unknown` — it is a refusal, not a diagnosis, and no category would say more.
 #
-# A source adapter's verdicts on a link (`SourceAdapter.resolve_url`, E11.5 S4) are adapters.py's
+# A source adapter's verdicts on a link (`SourceAdapter.resolve_url`) are adapters.py's
 # constants, imported above; each has a rule below that gives its category back, and a catalog entry
 # that gives the site's sentence.
 QUOTA_FULL_MESSAGE = "Your download storage is full — free up space, then try again."
+UPLOAD_MISSING_MESSAGE = "The uploaded file is missing."
 NETWORK_REDACTED_MESSAGE = "The server could not reach that address."
 
 # (marker, category) — first match wins, so order most-specific → most-general. Matching is
@@ -98,6 +99,8 @@ _RULES: tuple[tuple[str, str], ...] = (
     # An edit's source download went away — our own wording, and MORE specific than the generic
     # "no longer available" below, so it must be matched first.
     ("source download is no longer available", "source_missing"),
+    # An upload's original went away before its conversion ran — our wording (UPLOAD_MISSING_MESSAGE).
+    (UPLOAD_MISSING_MESSAGE.lower().rstrip("."), "upload_missing"),
     ("source file is missing", "source_missing"),
     # Private / deleted / removed / terminated.
     ("private video", "unavailable"),
@@ -263,6 +266,15 @@ def classify(msg: str | None) -> str:
     return "unknown"
 
 
+def classify_conversion(msg: str | None) -> str:
+    """The category of an upload's failed conversion. The generic ffmpeg categories speak of a
+    download ("the video downloaded but…"), which an upload never was: a failed encode or an
+    unrecognised failure of one is `convert_failed`; a category of its own (a full quota, a missing
+    original) stays as it is."""
+    code = classify(msg)
+    return "convert_failed" if code in ("postprocess", "unknown") else code
+
+
 def diagnose(
     msg: str | None, *, url: str | None = None, cookie_attached: bool = False
 ) -> Diagnosis:

backend/app/downloads/formats.py

@@ -13,7 +13,7 @@ spec shape (all keys optional; defaults + browser-playability clamp applied by `
     vcodec         always coerced to null → H.264 selected (see `normalize`); no opt-out
     embed_subs     bool     embed_chapters bool     embed_thumbnail bool     sponsorblock bool
 
-`normalize` clamps container/vcodec/audio_format so EVERY download plays in-browser (E7); the
+`normalize` clamps container/vcodec/audio_format so EVERY download plays in-browser; the
 `container`/`vcodec` keys survive only so an old stored profile deserializes without error.
 """
 
@@ -182,7 +182,9 @@ _DEFAULTS = {
 # v4: EVERY download is clamped to the mp4/H.264/AAC compat profile (no more mkv/webm/VP9/AV1/Opus
 # opt-out), so any asset cached under an old non-compat spec — or a stale non-H.264 mp4 that slipped
 # the v3 net — must re-derive through the current browser-safe guarantee.
-_SIG_VERSION = 4
+# v5: the phone-safe rule (`playable`) also judges the pixel format, so a cached 10-bit / 4:4:4 H.264
+# asset (which v4 let through as "H.264") re-derives as 8-bit 4:2:0 instead of being reused.
+_SIG_VERSION = 5
 # Audio-only downloads version on their own, so a change to the audio bytes doesn't re-download every
 # cached video. v5: an extracted .mp3/.m4a is stored as itself instead of remuxed into an .mp4 that
 # iPad's WebKit refuses to play, so a cached remuxed audio asset must re-derive.
@@ -217,14 +219,15 @@ def normalize(spec: dict) -> dict:
     # Audio-only files can't embed subtitles/thumbnails as video; keep the flags meaningful.
     if s["mode"] == "a":
         s["embed_subs"] = False
-    # Browser-playability guarantee (E7): the Download module exists to download → share → PLAY
+    # Browser-playability guarantee: the Download module exists to download → share → PLAY
     # in-browser, so every output must play in EVERY browser, iOS/Safari/WebKit included. The only
-    # universally-safe combo is mp4 + H.264 + AAC/MP3, so we CLAMP the spec to it HERE — the single
-    # point every path (picker, stored profile, raw API) flows through — instead of only hiding
-    # options in the UI. Non-mp4 containers (mkv doesn't play at all; webm/VP9/AV1 fail iOS) and any
-    # explicit vcodec are coerced to the compat profile; Opus audio (Safari-flaky as a bare file)
-    # falls back to AAC/m4a. A source that exposes no H.264 is still caught by the worker's
-    # browser-safe re-encode (`_ensure_browser_playable`), so playback is guaranteed end-to-end.
+    # universally-safe combo is mp4 + H.264 + AAC (an .mp3/.m4a for audio), so we CLAMP the spec to
+    # it HERE — the single point every path (picker, stored profile, raw API) flows through — instead
+    # of only hiding options in the UI. Non-mp4 containers (mkv plays nowhere on iOS; AV1 fails there,
+    # and webm/VP9 plays only on recent devices) and any explicit vcodec are coerced to the compat
+    # profile; Opus audio falls back to AAC/m4a. A source that exposes no H.264 (or a 10-bit one) is
+    # still caught by the worker's phone-safe step (`downloads.playable`), so playback is guaranteed
+    # end-to-end.
     s["container"] = "mp4"
     s["vcodec"] = None
     if s["audio_format"] not in ("m4a", "mp3"):
@@ -274,7 +277,7 @@ def build_ydl_opts(
     `extractor_args` with the player-client set + bgutil PO-token base_url); it is merged onto the
     generic options below. Everything here is source-agnostic.
 
-    `cookiefile` is the acting user's per-host cookies.txt path (S6), attached ONLY by the worker
+    `cookiefile` is the acting user's per-host cookies.txt path, attached ONLY by the worker
     after the host-scope + private-asset gate in `downloads.credentials`. It is set AFTER the `extra`
     merge so a source adapter can never clobber it.
 
@@ -329,7 +332,7 @@ def build_ydl_opts(
     if max_bytes and max_bytes > 0:
         opts["max_filesize"] = max_bytes
 
-    # The acting user's per-host cookie (S6). AFTER the adapter merge on purpose: an adapter must not
+    # The acting user's per-host cookie. AFTER the adapter merge on purpose: an adapter must not
     # be able to override the authenticated session the worker resolved for this specific user.
     if cookiefile:
         opts["cookiefile"] = cookiefile
@@ -351,7 +354,7 @@ def build_ydl_opts(
         # inside mp4 — a VP9/AV1-in-mp4 file shows a broken player there — so H.264 is selected at the
         # SELECTION level, not merely preferred: a sort key can't conjure an H.264 stream a source
         # doesn't expose as video-only. A source that ships only non-H.264 is caught after download by
-        # the worker's browser-safe re-encode (`_ensure_browser_playable`).
+        # the worker's phone-safe step (`downloads.playable`).
         if s["mode"] == "v":
             # Video-only: keep it audio-free. Prefer an H.264 video-only stream, then any video-only,
             # and only fall to a progressive (audio-bearing) `best` as a last resort.

backend/app/downloads/gc.py

@@ -6,12 +6,13 @@ Five passes each run:
   2. Expiry deletion — delete assets past their TTL (files + row). The FK SET NULL nulls the
      holders' jobs' asset_id, so those jobs surface as "expired" (re-downloadable) in the UI.
   3. LRU eviction — if a total cache cap is set and exceeded, delete least-recently-accessed
-     ready assets until back under it.
+     ready assets (never an upload) until back under it.
   4. Orphan reap — delete errored, fileless, unreferenced asset rows (they carry no TTL, so
      passes 1-3 skip them; benign row-count hygiene).
   5. Orphan FILE sweep — delete on-disk files/dirs no surviving asset owns (cruft left when a
-     re-download renames an asset, e.g. old accented channel dirs), skipping system trees and
-     anything newer than an hour so an in-flight download is never touched.
+     re-download renames an asset, e.g. old accented channel dirs, or an upload's stranded file).
+     It skips the dot-prefixed system trees (all but `.uploads`) and anything newer than an hour,
+     so an in-flight download is never touched.
 
 Runs in the API process (which mounts DOWNLOAD_ROOT). Pure disk/DB work — no YouTube quota.
 """
@@ -20,19 +21,19 @@ import time
 from datetime import datetime, timedelta, timezone
 from pathlib import Path
 
-from sqlalchemy import func, select
+from sqlalchemy import func, or_, select
 from sqlalchemy.orm import Session
 
 from app import logs, sysconfig
 from app.config import settings
-from app.downloads import storage
+from app.downloads import quota, storage
 from app.models import DownloadJob, MediaAsset
 from app.notifications import create_notification
 
 log = logs.get_logger(__name__)
-# Dot-prefixed top-level trees that are NOT part of the Plex/flat asset layout — the orphan sweep
-# never touches them (user clips, Plex HLS/img caches, in-progress staging, editor storyboards).
-_SYSTEM_TREE = {".edits", ".plex-hls", ".plex-img-cache", ".staging", ".storyboards"}
+# The one dot-prefixed top-level tree the orphan sweep enters. The others are not part of the asset
+# layout (user clips, Plex HLS/img caches, in-progress staging, editor storyboards) and are skipped.
+_UPLOAD_TREE = ".uploads"
 _ORPHAN_MIN_AGE_S = 3600  # only sweep files older than this, so an in-flight download is never hit
 
 
@@ -67,16 +68,20 @@ def _delete_asset(db: Session, asset: MediaAsset, ntype: str) -> None:
 def _sweep_orphan_files(db: Session, download_root: str) -> int:
     """Delete media/sidecar files (and now-empty dirs) under DOWNLOAD_ROOT that no `ready` asset
     owns — cruft left when a re-download changes an asset's on-disk name (e.g. an old accented
-    channel dir, or a poster from a superseded naming scheme). Skips the dot-prefixed system trees
-    and only removes files older than `_ORPHAN_MIN_AGE_S`, so a just-completed or in-flight download
-    is never deleted, and only files NOT owned by a surviving asset."""
+    channel dir, or a poster from a superseded naming scheme) — or, under `.uploads`, that no upload
+    row names at all. Skips the other dot-prefixed system trees and only removes files older than
+    `_ORPHAN_MIN_AGE_S`, so a just-completed or in-flight download is never deleted, and only files
+    NOT owned by a surviving asset."""
     root = Path(download_root)
     if not root.is_dir():
         return 0
     keep: set[Path] = set()
+    # An upload's file is on disk from ingest on, whatever its state (a queued conversion's original,
+    # a failed one kept for Retry), so every upload row keeps its file — not only a ready one.
     for rel in db.execute(
         select(MediaAsset.rel_path).where(
-            MediaAsset.status == "ready", MediaAsset.rel_path.is_not(None)
+            or_(MediaAsset.status == "ready", MediaAsset.source_kind == "upload"),
+            MediaAsset.rel_path.is_not(None),
         )
     ).scalars():
         media = (root / rel).resolve()
@@ -99,7 +104,10 @@ def _sweep_orphan_files(db: Session, download_root: str) -> int:
 
     unreclaimed = 0
     for entry in root.iterdir():
-        if entry.name.startswith("."):  # .edits / .plex-* / .staging / .storyboards + any dotfile
+        # .edits / .plex-* / .staging / .storyboards + any dotfile — but `.uploads` is swept: a
+        # conversion stopped between placing the new file and the commit that points at it (or the
+        # unlink of the original after it) leaves a file no row names, and nothing else finds it.
+        if entry.name.startswith(".") and entry.name != _UPLOAD_TREE:
             continue
         files = [entry] if entry.is_file() else [p for p in entry.rglob("*") if p.is_file()]
         for f in files:
@@ -188,17 +196,17 @@ def run_download_gc(db: Session) -> dict:
     if cap > 0:
         total = (
             db.execute(
-                select(func.coalesce(func.sum(MediaAsset.size_bytes), 0)).where(
-                    MediaAsset.status == "ready"
-                )
+                select(func.coalesce(func.sum(MediaAsset.size_bytes), 0)).where(quota.ON_DISK)
             ).scalar()
             or 0
         )
         if total > cap:
+            # An upload counts toward the total (it is disk like any other) but is never evicted:
+            # it has no source to fetch again, and it lives until its owner deletes it.
             candidates = (
                 db.execute(
                     select(MediaAsset)
-                    .where(MediaAsset.status == "ready")
+                    .where(MediaAsset.status == "ready", MediaAsset.source_kind != "upload")
                     .order_by(MediaAsset.last_access_at.asc().nulls_first())
                 )
                 .scalars()
@@ -232,6 +240,29 @@ def run_download_gc(db: Session) -> dict:
     if orphans:
         db.commit()
 
+    # 4b. Reap unheld uploads. An upload carries no expires_at, is never evicted and never re-enqueued,
+    #     so the passes above never free one that lost its last holder outside `release_asset` — a
+    #     conversion that finished after its job was cancelled, or one cancelled mid-run (the asset
+    #     was `downloading` then). Its file, ready or still the unconverted original, is at `rel_path`.
+    unheld = (
+        db.execute(
+            select(MediaAsset).where(
+                MediaAsset.source_kind == "upload",
+                MediaAsset.ref_count == 0,
+                MediaAsset.status != "downloading",
+            )
+        )
+        .scalars()
+        .all()
+    )
+    for asset in unheld:
+        if asset.rel_path:  # no holder is left to tell, so no notice — just the file and the row
+            storage.delete_asset_files(settings.download_root, asset.rel_path)
+        db.delete(asset)
+        reaped += 1
+    if unheld:
+        db.commit()
+
     # 5. Orphan FILE sweep — remove on-disk cruft no surviving asset owns (old accented dirs, stale
     #    posters). Runs last so the keep-set reflects the assets that survived passes 2-4.
     swept = _sweep_orphan_files(db, settings.download_root)

backend/app/downloads/og.py

@@ -147,13 +147,14 @@ def _watch_tags(token: str, db: Session) -> str | None:
     return "\n    ".join(tags)
 
 
-def render_watch_html(token: str, index_html: Path, db: Session) -> str | None:
-    """Return index.html with the OG block replaced for this watch token, or None to serve the
-    generic page unchanged (invalid/expired/not-ready link, or a template without the markers)."""
+def render_watch_html(token: str, page_html: Path, db: Session) -> str | None:
+    """Return the watch page (watch.html) with the OG block replaced for this token, or None to
+    serve the generic page unchanged (invalid/expired/not-ready link, or a template without the
+    markers)."""
     tags = _watch_tags(token, db)
     if tags is None:
         return None
-    text = index_html.read_text(encoding="utf-8")
+    text = page_html.read_text(encoding="utf-8")
     if "<!--OG:START-->" not in text:
         return None
     # A function replacement avoids re.sub interpreting backslashes/group refs in `tags`.

backend/app/downloads/pipeline.py

@@ -13,7 +13,8 @@ from app.config import settings
 from app.db import SessionLocal
 from app.downloads import edit as editmod
 from app.downloads import errors as dlerrors
-from app.downloads import adapters, credentials, formats, mediakind, probe, quota, service, storage
+from app.downloads import adapters, credentials, formats, mediakind, playable, probe, quota
+from app.downloads import service, storage
 from app.downloads import urlguard
 from app.downloads.jobdb import (
     _edit_staging_dir,
@@ -58,6 +59,10 @@ _DB_WRITE_INTERVAL_S = 1.0
 _CANCEL_POLL_INTERVAL_S = 1.0
 
 _THUMB_SUFFIXES = (".jpg", ".jpeg", ".png", ".webp")
+# The job phase while a file is brought into the phone-safe shape (`_ensure_phone_safe`) for a
+# download or an upload. Its own word, not yt-dlp's "processing": it reports ffmpeg's real progress,
+# so the queue row shows a % for it, where yt-dlp's post-steps only get an indeterminate pulse.
+CONVERTING = "converting"
 # yt-dlp's working files in the staging dir. None of them is ever the finished media, but several
 # CAN be the biggest file there (a video-only per-format stream, an abandoned fragment run), so the
 # staging fallback must exclude them by name — picking one would store e.g. a silent video-only
@@ -73,9 +78,9 @@ _YTDLP_TEMP_RE = re.compile(
 
 def _notify_download_done(db, user_id: int | None, title: str | None, job_id, *, kind: str) -> None:
     """Post a best-effort 'ready' notice to the owner's inbox — downloads/edits used to finish
-    silently while scheduler jobs and playlist-push already notified (R11 feedback symmetry). A
+    silently while scheduler jobs and playlist-push already notified (feedback symmetry). A
     notification failure must never mask the completed job, so it is isolated + rolled back (mirrors
-    scheduler._notify_done). `kind` ("download"/"edit") lets the inbox pick accurate copy; the
+    scheduler._notify_done). `kind` ("download"/"edit"/"upload") lets the inbox pick accurate copy; the
     English title here is only a fallback for an unknown type. `user_id`/`title` are captured by the
     caller BEFORE its commit, so reading them here needs no post-commit refresh of an expired row."""
     if user_id is None:
@@ -219,7 +224,7 @@ def _ffmpeg_progress_writer(job_id: int, phase: str, *, now=time.monotonic):
     when the whole-percent value changed (the bar shows integers), plus a separate once-per-interval
     heartbeat while the percent is stuck (so a slow re-encode isn't mistaken for a stall)."""
     # Progress writes have their OWN 1/interval budget so a heartbeat can never delay a real write (the
-    # round-1 concern). The heartbeat, in turn, is gated on `last_touch` — the last DB write OR beat —
+    # original concern). The heartbeat, in turn, is gated on `last_touch` — the last DB write OR beat —
     # so a progress write (which already refreshes updated_at) suppresses a redundant heartbeat that
     # same interval, keeping active downloads at ~1 write/interval, not two.
     allow_write = _rate_limiter(_DB_WRITE_INTERVAL_S, now=now)
@@ -341,7 +346,7 @@ def _extract_with_fallback(
     (adapters.POST_*_MESSAGE) instead of letting yt-dlp guess, and a failed page read fails the
     download as itself.
 
-    `cookiefile` (S6) is the acting user's per-host cookies.txt, already gated by
+    `cookiefile` is the acting user's per-host cookies.txt, already gated by
     `credentials.resolve_cookiefile`; passed to every attempt so the fallback stays authenticated.
 
     `max_bytes` (SEC-13) is the owner's remaining storage budget, applied to every attempt."""
@@ -416,48 +421,50 @@ def _download_settings() -> tuple[str, int]:
         )
 
 
-def _ensure_browser_playable(job_id: int, produced: Path, spec: dict, info: dict) -> Path:
-    """Make an mp4-compat-profile download play in every browser: H.264 + AAC/MP3 in mp4. If the
-    download (whatever the source shipped) isn't already that, re-encode the offending stream(s) —
-    or just remux when the codecs are fine but the container isn't. **Scope:** only the mp4 compat
-    profile is normalized; an explicit `vcodec` or a non-mp4 custom profile is the user opting out
-    of the guarantee and is left as-is (so a `webm` profile can still ship VP9/Opus). Returns the
-    file to store (converted, or the original) and corrects `info`'s codec fields to match disk.
+def _ensure_phone_safe(
+    job_id: int,
+    produced: Path,
+    out_dir: Path,
+    phase: str,
+    total_s: int | None = None,
+    *,
+    keep_source: bool = False,
+) -> Path:
+    """Bring a file that is about to become a ready library asset into the phone-safe shape
+    (`playable`), writing any converted copy into `out_dir`. Returns the file to store — `produced`
+    itself when it already is that shape, or isn't media at all — and removes `produced` once a
+    converted copy exists, unless `keep_source` (an upload's original, which must outlive the
+    conversion until the database points at the new file). The job shows `phase` with real ffmpeg
+    progress while it converts.
 
-    An audio download (mode "a") is never touched: yt-dlp already extracted it to .mp3 / .m4a, which
-    every browser plays as audio/mpeg / audio/mp4. Remuxing it into .mp4 served an mp3 as video/mp4,
-    which iPad's WebKit refuses to play (prod job 268)."""
-    if spec.get("mode") == "a" or spec.get("container") != "mp4" or spec.get("vcodec"):
+    Every ready path calls this — a download, an upload's worker conversion and an editor clip — so
+    a share link can only ever serve the one shape every phone and browser plays."""
+    p = playable.plan(probe.probe_json(produced), produced.suffix)
+    if p is None or not p.needs_change:
         return produced
-    vcodec, acodec = editmod.probe_codecs(produced)
-    is_mp4 = produced.suffix.lower() == ".mp4"
-    needs, reenc_v, reenc_a = editmod.browser_transcode_flags(vcodec, acodec, is_mp4)
-    if not needs:
-        return produced
-
-    dest = produced.with_name(produced.stem + ".browser.mp4")
-    action = "re-encode" if (reenc_v or reenc_a) else "remux"
+    dest = out_dir / f"{produced.stem}.safe.{p.ext}"
+    action = "re-encode" if (p.reencode_video or p.reencode_audio) else "remux"
     log.info(
-        "job %s: %s (v=%s a=%s mp4=%s) -> browser-safe mp4", job_id, action, vcodec, acodec, is_mp4
+        "job %s: %s (video=%s audio=%s) -> phone-safe .%s",
+        job_id,
+        action,
+        p.reencode_video,
+        p.reencode_audio,
+        p.ext,
+        extra=logs.event("download.phone_safe", job_id=job_id, action=action, ext=p.ext),
     )
-    _set_job(job_id, phase="processing", progress=0, speed_bps=None, eta_s=None)
-    # Prefer the info duration; probe the file when the extractor didn't provide one, so a long
-    # re-encode still shows real progress instead of a frozen 0%.
-    total_s = int(info["duration"]) if info.get("duration") else editmod.probe_duration(produced)
+    _set_job(job_id, phase=phase, progress=0, speed_bps=None, eta_s=None)
+    # A re-encode with no known length would hold a frozen 0%; probe the file for a denominator.
     editmod.run_ffmpeg(
-        editmod.build_browser_safe_cmd(
-            produced, dest, reenc_v, reenc_a, threads=editmod.encode_thread_cap()
-        ),
-        total_s=total_s,
-        on_progress=_ffmpeg_progress_writer(job_id, "processing"),
+        playable.build_cmd(produced, dest, p, threads=editmod.encode_thread_cap()),
+        total_s=total_s or editmod.probe_duration(produced),
+        on_progress=_ffmpeg_progress_writer(job_id, phase),
         should_cancel=_cancel_poll(job_id),
     )
-    produced.unlink(missing_ok=True)
-    # Keep the stored asset metadata truthful about what's now on disk.
-    if reenc_v:
-        info["vcodec"] = "h264"
-    if reenc_a:
-        info["acodec"] = "aac"
+    if not dest.exists():
+        raise StatedFailure("ffmpeg produced no output file")
+    if not keep_source:
+        produced.unlink(missing_ok=True)
     return dest
 
 
@@ -483,7 +490,7 @@ def _download(
                 urlguard.check_url(url)
             except urlguard.BlockedUrl as e:
                 raise StatedFailure(f"Refusing to download from this address ({e.reason}).") from e
-        # S6: attach the acting user's per-host cookie ONLY when it's host-scoped to this URL AND
+        # Attach the acting user's per-host cookie ONLY when it's host-scoped to this URL AND
         # this asset is that user's private, salted identity (the double gate lives in credentials).
         remaining_bytes: int | None = None
         with SessionLocal() as cdb:
@@ -522,7 +529,13 @@ def _download(
         adapter = adapters.resolve_adapter(source_kind, source_ref, info)
 
         produced = _produced_file(info, staging)
-        produced = _ensure_browser_playable(job_id, produced, spec, info)
+        produced = _ensure_phone_safe(
+            job_id,
+            produced,
+            staging,
+            CONVERTING,
+            int(info["duration"]) if info.get("duration") else None,
+        )
         ext = produced.suffix.lstrip(".").lower()
         # Codecs/dims/duration come from a PROBE of the produced file — the ground truth — NOT yt-dlp's
         # `info` dict, whose fields are absent for some sources (Facebook among them): deriving
@@ -722,9 +735,12 @@ def _log_download_failed(job_id: int, msg: str, code: str | None, exc: BaseExcep
     )
 
 
-def _fail_edit(job_id: int, asset_id: int, msg: str, exc: BaseException | None = None) -> None:
-    """Fail an edit job, and say so ONCE — every way an edit fails comes through here, so this is
-    where its line and its event code live. `exc` is the exception when there was one.
+def _fail_edit(
+    job_id: int, asset_id: int, msg: str, exc: BaseException | None = None, *, what: str = "edit"
+) -> None:
+    """Fail a local ffmpeg job — an edit, or an upload's conversion (`what="upload"`) — and say so
+    ONCE: every way one fails comes through here, so this is where its line and its event code
+    (`<what>.job.failed`) live. `exc` is the exception when there was one.
 
     Its traceback goes with the line unless the message already tells it all (`_failure_stack`).
     Logged after the write, as on the download error path."""
@@ -739,19 +755,22 @@ def _fail_edit(job_id: int, asset_id: int, msg: str, exc: BaseException | None =
             job_id,
             status=JobStatus.ERROR,
             error=msg,
-            error_code=dlerrors.classify(msg),
+            error_code=dlerrors.classify_conversion(msg)
+            if what == "upload"
+            else dlerrors.classify(msg),
             phase=None,
         )
         db.commit()
     # `msg` is the user's copy; ffmpeg's stderr tail in it can name the source file, whose path is
     # built from the channel and the title.
     log.warning(
-        "edit job %s failed: %s",
+        "%s job %s failed: %s",
+        what,
         job_id,
         formats.log_safe(msg),
         extra={
             **logs.event(
-                "edit.job.failed",
+                f"{what}.job.failed",
                 job_id=job_id,
                 asset_id=asset_id,
                 **({"error_type": type(exc).__name__} if exc is not None else {}),
@@ -811,11 +830,17 @@ def _process_edit(job_id: int, asset_id: int) -> None:
         )
         if not dest_staging.exists():
             raise StatedFailure("ffmpeg produced no output file")
+        # A fast trim stream-copies the source's codecs and container, so the clip is only as
+        # playable as its source: hold it to the same rule as every other ready file.
+        dest_staging = _ensure_phone_safe(job_id, dest_staging, staging, "editing", total_s)
+        out_ext = dest_staging.suffix.lstrip(".").lower() or out_ext
 
         rel = storage.edit_rel_path(user_id, asset_id, out_ext)
         storage.place_file(dest_staging, settings.download_root, rel)
         final = storage.abs_path(settings.download_root, rel)
         size = final.stat().st_size if final.exists() else None
+        # The clip's real frame: a re-encode evens an odd crop or source (`editmod.EVEN_FRAME`).
+        _, _, out_w, out_h, _ = probe.probe_media(final)
         _, retention_days = _download_settings()
         expires = datetime.now(timezone.utc) + timedelta(days=retention_days)
 
@@ -827,7 +852,9 @@ def _process_edit(job_id: int, asset_id: int) -> None:
                 asset.size_bytes = size
                 asset.container = out_ext
                 asset.duration_s = total_s or asset.duration_s
-                if crop:
+                if out_w and out_h:
+                    asset.width, asset.height = out_w, out_h
+                elif crop:
                     asset.width = edit_spec["crop"]["w"]
                     asset.height = edit_spec["crop"]["h"]
                 else:
@@ -867,3 +894,163 @@ def _process_edit(job_id: int, asset_id: int) -> None:
         _fail_edit(job_id, asset_id, str(exc)[:500], exc)
     finally:
         shutil.rmtree(staging, ignore_errors=True)
+
+
+def _check_converted_fits(owner_id: int | None, old_size: int, new_size: int) -> None:
+    """Refuse a converted upload that would push its owner over the storage cap: the upload was
+    charged at its original size, and a re-encode can come out larger. The original stays, counted,
+    for the owner to delete or keep."""
+    if owner_id is None:
+        return
+    with SessionLocal() as db:
+        if not quota.fits_after_conversion(db, owner_id, old_size, new_size):
+            raise StatedFailure(dlerrors.QUOTA_FULL_MESSAGE)
+
+
+def _unplace(placed: Path | None) -> None:
+    """Remove a converted upload that was placed at its new rel but never committed: the row still
+    names the original, so nothing else would ever find this file."""
+    if placed is not None:
+        placed.unlink(missing_ok=True)
+
+
+def _conversion_landed(asset_id: int, new_rel: str) -> bool | None:
+    """After the commit that makes a converted upload ready RAISED: did it land anyway? A commit can
+    reach the server and lose its answer (the connection drops after COMMIT). True/False by what the
+    row now says; None when the row cannot be read either."""
+    try:
+        with SessionLocal() as db:
+            asset = db.get(MediaAsset, asset_id)
+            return asset is not None and asset.status == "ready" and asset.rel_path == new_rel
+    except Exception:  # noqa: BLE001 — "cannot tell" is the answer, the caller acts on it
+        # silent-ok: the caller re-raises the commit's own error, which `_fail_edit` reports.
+        return None
+
+
+def _drop_original(job_id: int, src: Path) -> None:
+    """Remove an upload's original once the converted file is committed in its place. Best effort:
+    the job is done and its file is ready, so a failure here must not fail it — the original is
+    named by no row now, and the GC's `.uploads` sweep takes it."""
+    try:
+        src.unlink(missing_ok=True)
+    except OSError as exc:
+        log.warning(
+            "upload job %s: original not removed, left to the GC sweep: %s",
+            job_id,
+            exc,
+            extra=logs.event("upload.original.unlink_failed", job_id=job_id),
+        )
+
+
+def _free_unheld_upload(asset_id: int) -> bool:
+    """Free an upload no job holds any more — cancelled or deleted while its conversion ran, when
+    `release_asset` had to leave it because the conversion still read its file. Once that has
+    stopped, free it as the release would have; left alone, its file would sit on disk counted by no
+    owner's footprint until the GC's upload pass. False when a job still holds it (paused)."""
+    with SessionLocal() as db:
+        asset = db.get(MediaAsset, asset_id)
+        if asset is None or asset.ref_count != 0:
+            return False
+        if asset.rel_path:
+            storage.delete_asset_files(settings.download_root, asset.rel_path)
+        db.delete(asset)
+        db.commit()
+    return True
+
+
+def _convert_upload(job_id: int, asset_id: int) -> None:
+    """Convert a queued upload into the phone-safe shape (`playable`) and make it ready.
+
+    The original sits at the asset's `rel_path` (see `upload`); the converted file replaces it there
+    under the new extension, and only then is the original removed — so a failed or stopped
+    conversion leaves the original in place for Retry, and a delete still finds it by `rel_path`.
+    A converted file placed but never committed is removed again on the way out. When the commit's
+    own outcome is unknown neither file is removed: the row names one of them, and the GC's `.uploads`
+    sweep takes the one it does not — as it does an original whose unlink after the commit failed."""
+    with SessionLocal() as db:
+        asset = db.get(MediaAsset, asset_id)
+        old_rel = asset.rel_path if asset else None
+        old_ext = asset.container if asset else None
+        old_size = (asset.size_bytes or 0) if asset else 0
+        duration = asset.duration_s if asset else None
+        job = db.get(DownloadJob, job_id)
+        owner_id = job.user_id if job else None
+    src = storage.safe_abs_path(settings.download_root, old_rel) if old_rel else None
+    if old_rel is None or src is None or not src.exists():
+        _fail_edit(job_id, asset_id, dlerrors.UPLOAD_MISSING_MESSAGE, what="upload")
+        return
+    staging = _staging_dir(asset_id)
+    placed: Path | None = None  # the converted file while it is at its new rel, uncommitted
+    try:
+        shutil.rmtree(staging, ignore_errors=True)
+        staging.mkdir(parents=True, exist_ok=True)
+        safe = _ensure_phone_safe(job_id, src, staging, CONVERTING, duration, keep_source=True)
+        ext = safe.suffix.lstrip(".").lower()
+        new_rel = storage.with_ext(old_rel, old_ext, ext)
+        if safe != src:
+            _check_converted_fits(owner_id, old_size, safe.stat().st_size)
+            storage.place_file(safe, settings.download_root, new_rel)
+            if new_rel != old_rel:
+                placed = storage.abs_path(settings.download_root, new_rel)
+        final = storage.abs_path(settings.download_root, new_rel)
+        vcodec, acodec, width, height, dur = probe.probe_media(final)
+        with SessionLocal() as db:
+            asset = db.get(MediaAsset, asset_id)
+            if asset:
+                asset.status = "ready"
+                asset.rel_path = new_rel
+                asset.container = ext
+                asset.size_bytes = final.stat().st_size
+                asset.vcodec, asset.acodec = vcodec, acodec
+                asset.width, asset.height = width or asset.width, height or asset.height
+                asset.duration_s = dur or asset.duration_s
+                asset.error = None
+                asset.last_access_at = datetime.now(timezone.utc)
+            title = asset.title if asset else None
+            finalized = service.finalize_if_running(
+                db, job_id, status=JobStatus.DONE, progress=100, phase=None
+            )
+            try:
+                db.commit()
+            except Exception:
+                landed = _conversion_landed(asset_id, new_rel)
+                if landed is False:
+                    raise  # rolled back: the original is still the asset's file — fail as any step
+                # Landed, or cannot tell: the converted file may be the asset's file now, so it is
+                # never removed blind (the sweep takes whichever file no row names).
+                placed = None
+                if landed is None:
+                    raise
+                db.rollback()
+            placed = None
+            if finalized:
+                _notify_download_done(db, owner_id, title, job_id, kind="upload")
+        if new_rel != old_rel:
+            _drop_original(job_id, src)
+        if not finalized:
+            _free_unheld_upload(asset_id)  # finished after its job was cancelled or deleted
+        if finalized:
+            log.info(
+                "upload job %s converted: asset %s",
+                job_id,
+                asset_id,
+                extra=logs.event("upload.job.done", job_id=job_id, asset_id=asset_id, ext=ext),
+            )
+    except editmod.EditAborted:
+        # Paused/canceled mid-encode: the original is untouched. A paused job keeps it for its
+        # resume; one cancelled or deleted holds it no more, so it is freed now.
+        _unplace(placed)
+        if not _free_unheld_upload(asset_id):
+            _reset_asset_pending(asset_id)
+        log.info(
+            "upload job %s conversion aborted (pause/cancel)",
+            job_id,
+            extra=logs.event("upload.job.aborted", job_id=job_id),
+        )
+    except Exception as exc:  # noqa: BLE001 — surface the failure, keep the worker alive
+        # silent-ok: not silent — `_fail_edit` is the report, and it takes `exc` for its stack.
+        _unplace(placed)
+        _fail_edit(job_id, asset_id, str(exc)[:500], exc, what="upload")
+
+    finally:
+        shutil.rmtree(staging, ignore_errors=True)

backend/app/downloads/playable.py

@@ -0,0 +1,103 @@
+"""The phone-safe rule: what a library file must be so it plays wherever a share link lands.
+
+A shared `/watch` link is opened by whoever got it, on whatever they hold: an iPhone's WebKit, an
+Android Chrome, a desktop Firefox. The one shape every one of them plays:
+
+  * video → H.264, 8-bit 4:2:0 (`yuv420p`, or its full-range twin `yuvj420p`), AAC audio, in .mp4
+  * audio only → MP3 in .mp3, or AAC in .m4a
+
+Measured on the owner's iPhone + iPad (2026-10-05): AV1, Matroska, HEVC tagged `hev1` and MP3
+inside an .mp4 all fail there. VP9/WebM, HEVC `hvc1` and 10-bit H.264 played on those devices, but
+Chrome and Firefox do not decode HEVC and browsers decode 10-bit / 4:4:4 H.264 only patchily, so the
+rule is the shape every browser plays, not the shape one device plays.
+
+Every file that becomes a ready library asset passes through here — a download, an upload (the
+worker converts it, see `upload`), and an editor clip — so a link can only ever serve this shape.
+`plan` says what (if anything) must change; `build_cmd` is the one ffmpeg command that changes it:
+re-encode only the stream that fails, copy the rest, or just remux when only the container is wrong.
+"""
+
+from dataclasses import dataclass
+from pathlib import Path
+
+from app.downloads import edit as editmod
+from app.downloads import probe
+
+_SAFE_VCODEC = "h264"
+_SAFE_PIX_FMTS = frozenset({"yuv420p", "yuvj420p"})
+_VIDEO_EXT = "mp4"
+# An audio-only file: the codec, and the extension it plays in. Anything else becomes AAC in .m4a.
+_AUDIO_HOME = {"mp3": "mp3", "aac": "m4a"}
+_AUDIO_FALLBACK_EXT = "m4a"
+
+
+@dataclass(frozen=True)
+class Plan:
+    """What makes one file phone-safe. `ext` is the extension the safe file has (no dot)."""
+
+    ext: str
+    has_video: bool
+    reencode_video: bool
+    reencode_audio: bool
+    needs_change: bool
+
+
+def plan(data: dict | None, ext: str) -> Plan | None:
+    """The phone-safe plan for a file whose `probe.probe_json` is `data` and whose extension is
+    `ext`; None when it is not media at all (an unreadable probe, a still image, no audio or video
+    stream) — such a file is a download-only 'file' and never plays inline, so there is nothing to
+    make safe."""
+    if data is None:
+        return None
+    video, audio = probe.primary_streams(data)
+    ext = ext.lower().lstrip(".")
+    if video is not None:
+        rv = (video.get("codec_name") or "").lower() != _SAFE_VCODEC or (
+            video.get("pix_fmt") or ""
+        ).lower() not in _SAFE_PIX_FMTS
+        ra = audio is not None and (audio.get("codec_name") or "").lower() != "aac"
+        return Plan(_VIDEO_EXT, True, rv, ra, rv or ra or ext != _VIDEO_EXT)
+    if audio is not None:
+        home = _AUDIO_HOME.get((audio.get("codec_name") or "").lower())
+        target = home or _AUDIO_FALLBACK_EXT
+        return Plan(target, False, False, home is None, home is None or ext != target)
+    return None
+
+
+def build_cmd(src: Path, dest: Path, p: Plan, threads: int | None = None) -> list[str]:
+    """The ffmpeg command that writes `src` to `dest` (whose suffix is `p.ext`) in the phone-safe
+    shape: the failing stream(s) re-encoded with the package's shared encoder settings, the rest
+    stream-copied. An audio-only result drops any cover-art stream. `-progress pipe:1`
+    streams machine-readable progress; `threads` caps the encoder so a re-encode cannot claim every
+    core on the shared box (see the worker-child de-prioritization)."""
+    cmd = [
+        "ffmpeg",
+        "-y",
+        "-hide_banner",
+        "-loglevel",
+        "error",
+        "-nostdin",
+        "-progress",
+        "pipe:1",
+        "-nostats",
+        *editmod.INPUT_GUARD,
+        "-i",
+        str(src),
+    ]
+    # Exactly the two streams `plan` judged: `V` is a video stream that is not cover art. Nothing
+    # else rides along — a subtitle or data track the .mp4 muxer cannot take would fail the run.
+    if p.has_video:
+        cmd += ["-map", "0:V:0", "-map", "0:a:0?"]
+        if p.reencode_video:
+            cmd += ["-vf", editmod.EVEN_FRAME, *editmod.X264_ARGS]
+        else:
+            cmd += ["-c:v", "copy"]
+    else:
+        cmd += ["-map", "0:a:0"]
+    cmd += editmod.AAC_ARGS if p.reencode_audio else ["-c:a", "copy"]
+    if threads and threads > 0:
+        cmd += ["-threads", str(threads)]
+    if p.ext in ("mp4", "m4a"):
+        cmd += ["-movflags", "+faststart"]
+    cmd += [str(dest)]
+    return cmd

backend/app/downloads/probe.py

@@ -34,6 +34,28 @@ _IMAGE_FORMATS = {
 }
 
 
+def primary_streams(data: dict) -> tuple[dict | None, dict | None]:
+    """(video stream, audio stream) — the first of each kind — from a NON-None ffprobe JSON dict, by
+    the same reading the stored facts use: a still image / image-sequence container has neither (it
+    is a generic 'file'), and a video stream flagged `attached_pic` (an mp3/m4a's cover art) is not a
+    video track. The phone-safe rule (`playable`) judges exactly these two streams."""
+    fmt_names = {
+        n.strip() for n in ((data.get("format") or {}).get("format_name") or "").split(",")
+    }
+    if fmt_names & _IMAGE_FORMATS:
+        return None, None
+    video = audio = None
+    for stream in data.get("streams", []):
+        kind = stream.get("codec_type")
+        if kind == "video" and video is None:
+            if (stream.get("disposition") or {}).get("attached_pic"):
+                continue  # cover art, not a video track
+            video = stream
+        elif kind == "audio" and audio is None:
+            audio = stream
+    return video, audio
+
+
 def _facts_from_probe(
     data: dict,
 ) -> tuple[str | None, str | None, int | None, int | None, int | None]:
@@ -45,19 +67,11 @@ def _facts_from_probe(
     fmt_names = {n.strip() for n in (fmt.get("format_name") or "").split(",")}
     if fmt_names & _IMAGE_FORMATS:
         return None, None, None, None, None
-
-    vcodec = acodec = None
-    width = height = None
-    for stream in data.get("streams", []):
-        kind = stream.get("codec_type")
-        if kind == "video" and vcodec is None:
-            if (stream.get("disposition") or {}).get("attached_pic"):
-                continue  # cover art, not a video track
-            vcodec = stream.get("codec_name")
-            width = stream.get("width")
-            height = stream.get("height")
-        elif kind == "audio" and acodec is None:
-            acodec = stream.get("codec_name")
+    video, audio = primary_streams(data)
+    vcodec = video.get("codec_name") if video else None
+    width = video.get("width") if video else None
+    height = video.get("height") if video else None
+    acodec = audio.get("codec_name") if audio else None
 
     duration = None
     try:
@@ -71,16 +85,29 @@ def _facts_from_probe(
     return vcodec, acodec, width, height, duration
 
 
-def probe_media(path: Path) -> tuple[str | None, str | None, int | None, int | None, int | None]:
-    """(vcodec, acodec, width, height, duration_s) from one ffprobe call; all None if ffprobe fails
-    or the file isn't playable media (an unprobeable file — or a still image — is treated as a generic
-    non-media file, so it becomes a download-only 'file' rather than a broken 'video')."""
-    data = editmod.run_ffprobe_json(path, show_streams=True, show_format=True)
+def probe_json(path: Path) -> dict | None:
+    """The full ffprobe JSON (streams + format) of `path`, or None when ffprobe could not read it —
+    the one call both the stored facts (`facts`) and the phone-safe rule read from."""
+    return editmod.run_ffprobe_json(path, show_streams=True, show_format=True)
+
+
+def facts(
+    data: dict | None,
+) -> tuple[str | None, str | None, int | None, int | None, int | None]:
+    """(vcodec, acodec, width, height, duration_s) of a `probe_json` result; all None when it is None
+    (an unprobeable file is treated as a generic non-media file, so it becomes a download-only 'file'
+    rather than a broken 'video')."""
     if data is None:
         return None, None, None, None, None
     return _facts_from_probe(data)
 
 
+def probe_media(path: Path) -> tuple[str | None, str | None, int | None, int | None, int | None]:
+    """(vcodec, acodec, width, height, duration_s) from one ffprobe call; all None if ffprobe fails
+    or the file isn't playable media (a still image is a generic 'file', see `primary_streams`)."""
+    return facts(probe_json(path))
+
+
 def resolve_media_facts(
     path: Path, info: dict
 ) -> tuple[str | None, str | None, int | None, int | None, int | None]:
@@ -94,18 +121,18 @@ def resolve_media_facts(
     is authoritative — INCLUDING a deliberate None (cover-art-only audio, a still image). A per-field
     `probe or info` fallback would let `info`'s stale vcodec resurrect a video track the file doesn't
     have (yt-dlp reports a progressive `vcodec` for an audio-mode download whose video it then strips),
-    mislabelling an audio download as a broken in-app 'video'. Only when ffprobe could NOT read the
-    file at all do we fall back to `info` (the pre-probe behaviour, a safety net); the cosmetic
-    dimensions and the duration still gap-fill from `info` in either case."""
+    mislabelling an audio download as a broken in-app 'video'. When ffprobe could NOT read the file
+    at all, no codec is reported: the probe runs with the media demuxer whitelist, so a file it
+    cannot open is one the player and the editor cannot open either — and taking `info`'s codecs
+    for it would make it an editable 'video' whose bytes are whatever the remote site served. The
+    cosmetic dimensions and the duration still gap-fill from `info` in either case."""
     data = editmod.run_ffprobe_json(path, show_streams=True, show_format=True)
-    info_v = info.get("vcodec") if info.get("vcodec") not in (None, "none") else None
-    info_a = info.get("acodec") if info.get("acodec") not in (None, "none") else None
     info_w = int(info["width"]) if info.get("width") else None
     info_h = int(info["height"]) if info.get("height") else None
     info_d = int(info["duration"]) if info.get("duration") else None
     if data is None:
-        # ffprobe could not read the file — fall back to the extractor's info entirely.
-        return info_v, info_a, info_w, info_h, info_d
+        # Unreadable to the guarded probe: a download-only 'file' (no codecs → media_kind FILE).
+        return None, None, info_w, info_h, info_d
     pv, pa, pw, ph, pd = _facts_from_probe(data)
     # Probe ran → its codecs win outright (a None means "no such track", not "unknown"); info only
     # gap-fills the width/height/duration the probe couldn't read.

backend/app/downloads/quota.py

@@ -1,7 +1,9 @@
 """Per-user download limits + footprint accounting.
 
 A user's *footprint* is the total size of the ready assets their library references (cache hits
-count — it's "how much disk you're responsible for", not "how much you personally downloaded").
+count — it's "how much disk you're responsible for", not "how much you personally downloaded"). An
+upload counts from the moment it lands, ready or not: its original is on disk while its conversion
+waits, and stays there when the conversion fails (for Retry) — bytes a ready-only sum never saw.
 Limits come from a per-user DownloadQuota row if the admin set one, else the sysconfig defaults.
 `unlimited` (e.g. the admin's own account) bypasses the byte cap.
 
@@ -18,16 +20,24 @@ is unknown until it finishes. Edit jobs do not consult it at all (tracked as SEC
 
 from dataclasses import dataclass
 
-from sqlalchemy import func, select
+from sqlalchemy import and_, func, or_, select
 from sqlalchemy.orm import Session
 
 from app import sysconfig
+from app.downloads.status import JobStatus
 from app.models import DownloadJob, DownloadQuota, MediaAsset
 
 # Jobs a user is considered to be "holding" (occupy a job slot / footprint). Terminal failures
 # (error, canceled) don't count.
-_HOLDING = ("queued", "running", "paused", "done")
-_RUNNING = ("queued", "running")
+_HOLDING = (JobStatus.QUEUED, JobStatus.RUNNING, JobStatus.PAUSED, JobStatus.DONE)
+_RUNNING = (JobStatus.QUEUED, JobStatus.RUNNING)
+# An asset whose file is on disk: a ready one, or an upload in any state (its original lands at
+# ingest, see the module docstring). Every byte sum — a user's footprint, the GC's total-cap pass,
+# the admin's storage view — reads it, so they all count the same disk.
+ON_DISK = or_(
+    MediaAsset.status == "ready",
+    and_(MediaAsset.source_kind == "upload", MediaAsset.rel_path.is_not(None)),
+)
 
 
 @dataclass
@@ -67,12 +77,17 @@ def resolve(db: Session, user_id: int) -> QuotaLimits:
 
 
 def footprint(db: Session, user_id: int) -> int:
-    """Total bytes of the distinct ready assets this user's active jobs reference."""
+    """Total bytes of the distinct assets this user's jobs hold on disk: the ready ones their active
+    jobs reference, and every upload of theirs — including one whose conversion failed, which keeps
+    its original (see the module docstring)."""
     held_assets = (
         select(DownloadJob.asset_id)
         .where(
             DownloadJob.user_id == user_id,
-            DownloadJob.status.in_(_HOLDING),
+            or_(
+                DownloadJob.status.in_(_HOLDING),
+                and_(DownloadJob.job_kind == "upload", DownloadJob.status == JobStatus.ERROR),
+            ),
             DownloadJob.asset_id.is_not(None),
         )
         .distinct()
@@ -80,7 +95,7 @@ def footprint(db: Session, user_id: int) -> int:
     return int(
         db.execute(
             select(func.coalesce(func.sum(MediaAsset.size_bytes), 0)).where(
-                MediaAsset.id.in_(held_assets), MediaAsset.status == "ready"
+                MediaAsset.id.in_(held_assets), ON_DISK
             )
         ).scalar()
         or 0
@@ -111,7 +126,7 @@ def at_concurrency_limit(db: Session, user_id: int) -> bool:
     lim = resolve(db, user_id)
     if lim.unlimited:
         return False
-    running = _count(db, user_id, ("running",))
+    running = _count(db, user_id, (JobStatus.RUNNING,))
     return running >= lim.max_concurrent
 
 
@@ -160,6 +175,18 @@ def check_upload(db: Session, user_id: int, add_bytes: int) -> None:
         raise QuotaExceeded("max_bytes", lim.max_bytes, projected)
 
 
+def fits_after_conversion(db: Session, user_id: int, old_bytes: int, new_bytes: int) -> bool:
+    """Whether an upload converted from `old_bytes` (what it is charged at now) to `new_bytes` keeps
+    its owner within the cap. A re-encode can come out larger than the original, and nothing else
+    re-checks it once the upload was let in."""
+    if new_bytes <= old_bytes:  # a smaller file never breaks a cap the larger one was let in under
+        return True
+    lim = resolve(db, user_id)
+    if lim.unlimited:
+        return True
+    return footprint(db, user_id) - old_bytes + new_bytes <= lim.max_bytes
+
+
 def usage(db: Session, user_id: int) -> dict:
     """Compact usage snapshot for the UI (footprint + counts vs limits)."""
     lim = resolve(db, user_id)

backend/app/downloads/runner.py

@@ -57,7 +57,7 @@ def main() -> None:
     logs.configure_logging("runner")
     _deprioritize()
     if len(sys.argv) != 2:
-        # noqa T201: a usage line is for the person who typed the command. Logging is already
+        # T201: a usage line is for the person who typed the command. Logging is already
         # configured here, and in a deployment it writes JSON records — the one reader of this
         # line wants the bare sentence, not a record with a request id they do not have.
         print(  # noqa: T201
@@ -95,16 +95,18 @@ def _run(job_id, asset_id, job_kind, source_kind, source_ref, spec, user_id) ->
     (`cause=child_exit`)."""
     # Imported here (not at module top) so the child pays the yt-dlp import cost only when it has
     # real work, and to keep the import graph obvious: runner is a thin entry over the pipeline.
-    from app.downloads.pipeline import _download, _process_edit
+    from app.downloads.pipeline import _convert_upload, _download, _process_edit
 
     with formats.hiding(source_ref):
         try:
             if job_kind == "edit":
                 _process_edit(job_id, asset_id)
+            elif job_kind == "upload":
+                _convert_upload(job_id, asset_id)
             else:
                 _download(job_id, asset_id, source_kind, source_ref, spec, user_id)
         except Exception as exc:  # noqa: BLE001 — whatever it is, it is reported here or raw
-            # noqa TRY400: the stack goes with it scrubbed (`scrubbed_stack`); `exc_info` is raw.
+            # TRY400: the stack goes with it scrubbed (`scrubbed_stack`); `exc_info` is raw.
             log.error(  # noqa: TRY400
                 "the download child for job %s failed",
                 job_id,

backend/app/downloads/service.py

@@ -318,14 +318,22 @@ def release_asset(db: Session, job: DownloadJob) -> None:
     re-enqueuing the same (source,format) — get_or_create_asset would reset that row to `pending` and
     +1 it, and deleting it under that would FK-null the new job's asset (a lost update on ref_count).
     A ready asset is still freed at ref==0 (its file is the point); an orphaned errored row is cheap
-    (no file) and gets reused+reset by the next enqueue, or reclaimed by a later GC pass."""
+    (no file) and gets reused+reset by the next enqueue, or reclaimed by a later GC pass.
+
+    An UPLOAD is the exception: it is never shared or re-enqueued (its ref is unique), and while it
+    waits for or failed its phone-safe conversion the original file already sits at `rel_path` — so
+    it is freed at ref==0 whatever its state, except mid-conversion (`downloading`: the worker child
+    still reads it; the GC upload pass takes it once the child has stopped)."""
     if not job.asset_id or job.status not in HOLDING:
         return
     asset = db.get(MediaAsset, job.asset_id)
     if asset is None:
         return
     new_count = decr_ref_count(db, asset.id)
-    if new_count == 0 and asset.status == "ready":
+    freeable = asset.status == "ready" or (
+        asset.source_kind == "upload" and asset.status != "downloading"
+    )
+    if new_count == 0 and freeable:
         if asset.rel_path:
             storage.delete_asset_files(settings.download_root, asset.rel_path)
         db.delete(asset)

backend/app/downloads/storage.py

@@ -168,13 +168,25 @@ def upload_rel_path(user_id: int, asset_id: int, filename: str) -> str:
     episode sidecars). The name is system-decided and bound to the asset id (`_a{id}` keeps two
     same-named uploads apart); the original filename is display metadata only. The extension is
     preserved (shell-safe, bounded) so the served file + its Content-Disposition carry a correct
-    suffix; an extensionless upload stays extensionless."""
-    stem = sanitize(Path(filename).stem, 80)
+    suffix; an extensionless upload stays extensionless.
+
+    The stem holds no dot, so the last dot of the name is always the extension's: every sibling
+    derived by swapping it (`with_suffix` — the poster, the converted file) keeps `_a{id}`. A dotted
+    stem with no extension ('Holiday.2024.ä') once lost it, and two such uploads shared one file."""
+    stem = sanitize(Path(filename).stem, 80).replace(".", "_")
     ext = _clean_upload_ext(filename)
     suffix = f".{ext}" if ext else ""
     return f".uploads/{user_id}/{stem}_a{asset_id}{suffix}"
 
 
+def with_ext(rel: str, old_ext: str | None, ext: str) -> str:
+    """`rel` with its extension — `old_ext`, the stored `container`, None for a bare name — swapped
+    for `ext`. Not `Path.with_suffix`: on a bare name whose stem holds a dot (an upload placed before
+    the stems lost their dots) that would take the stem's tail for the extension, and `_a{id}` with it."""
+    base = rel[: -len(old_ext) - 1] if old_ext and rel.endswith(f".{old_ext}") else rel
+    return f"{base}.{ext}"
+
+
 def upload_ext(filename: str) -> str | None:
     """The stored `container` for an uploaded file: its shell-safe lowercased extension, or None when
     the upload has no extension (a bare name). Shares `_clean_upload_ext` with `upload_rel_path` so the

backend/app/downloads/upload.py

@@ -1,15 +1,24 @@
-"""Local file ingest for the download center (E7 S5) — the non-yt-dlp path into the media library.
+"""Local file ingest for the download center — the non-yt-dlp path into the media library.
 
 A user-uploaded file joins the SAME library as a downloaded one, so it inherits the whole function
 set already built (metadata chips, share links, and — for media — in-app play/edit). Unlike a
-download it bypasses yt-dlp AND the worker entirely: the file is already on the host, so ingest
-probes it, places it on disk, and creates an already-`ready` MediaAsset + `done` DownloadJob in one
-step — the same "the file exists, mark it done now" shortcut `service.enqueue` takes on a cache hit.
+download it bypasses yt-dlp: the file is already on the host, so ingest probes it and places it on
+disk. A file that already is the phone-safe shape (`playable`) — or is not media at all — becomes an
+already-`ready` MediaAsset + `done` DownloadJob in one step, the same "the file exists, mark it done
+now" shortcut `service.enqueue` takes on a cache hit.
+
+Media that is NOT phone-safe (an iPhone's HEVC .mov, a VP9 .webm, an .mkv) is placed as it came and
+queued: the asset stays `pending` and the job `queued`, and the download worker converts it
+(`pipeline._convert_upload`) — so a share link never hands a friend a file their phone cannot play.
+Only the converted file is kept (the owner's choice: one file per upload). A 10-bit HDR original
+becomes 8-bit H.264 with its colour tags kept — still tagged HDR (HLG/PQ, BT.2020), not tone-mapped
+to SDR (the owner's choice too). Until then the original sits at the asset's `rel_path`, which is
+what a failed conversion's Retry starts from again, and what a delete removes
+(`service.release_asset`).
 
 Kinds (the invariant lives in `mediakind`): a probe with a real video stream -> video, audio-only
 -> audio, neither -> a generic 'file' (pdf/zip/docx/…) that is download-only and never rendered
-inline (see the force-download serving in the file/watch routes). This module is the ingest seam a
-future sub-function (e.g. convert) plugs a new adapter into, mirroring how it sidesteps the worker.
+inline (see the force-download serving in the file/watch routes).
 
 An upload does NOT get an `expires_at`: it's user-provided and may be irreplaceable (there's no
 source to re-download), so it doesn't age out on the retention clock the way a re-fetchable download
@@ -24,7 +33,8 @@ from sqlalchemy.orm import Session
 
 from app.config import settings
 from app.downloads import edit as editmod
-from app.downloads import mediakind, probe, quota, service, storage
+from app.downloads import mediakind, playable, probe, quota, service, storage
+from app.downloads.status import JobStatus
 from app.models import DownloadJob, MediaAsset, display_text
 
 
@@ -44,15 +54,22 @@ def ingest_upload(
     size: int,
     display_name: str | None = None,
 ) -> DownloadJob:
-    """Ingest an already-uploaded file at `src_path` into `user_id`'s library and return the `done`
-    job. `src_path` is MOVED into place on success (the caller stages it to a temp file first).
+    """Ingest an already-uploaded file at `src_path` into `user_id`'s library and return its job —
+    `done`, or `queued` for the worker's phone-safe conversion (see the module docstring). `src_path`
+    is MOVED into place on success (the caller stages it to a temp file first).
 
     Raises quota.QuotaExceeded (holdings/footprint cap) — checked BEFORE any disk write, so a
     rejected upload leaves nothing behind and the caller just cleans up its temp file."""
     quota.check_upload(db, user_id, size)  # the file's size is known → charge it up front
 
-    vcodec, acodec, width, height, duration = probe.probe_media(src_path)
+    data = probe.probe_json(src_path)
+    vcodec, acodec, width, height, duration = probe.facts(data)
     kind = mediakind.derive(vcodec, acodec)
+    # Judged by the extension the file is PLACED under (and the worker re-judges), not the raw one:
+    # 'clip.mp4~' is stored as .mp4, and judging 'mp4~' queued a conversion with nothing to convert.
+    ext = storage.upload_ext(filename) or ""
+    safe = playable.plan(data, ext) if kind != mediakind.FILE else None
+    convert = safe is not None and safe.needs_change
 
     # An upload never dedups (each file is its own asset): a random ref makes the cache key unique,
     # so `format_sig` is a constant marker rather than a format hash.
@@ -66,17 +83,18 @@ def ingest_upload(
     rel = storage.upload_rel_path(user_id, asset.id, filename)
     storage.place_file(src_path, settings.download_root, rel)
     # From here the file is on disk under `.uploads/`. If the transaction fails before it commits, the
-    # asset/job rows roll back but the placed file would be ORPHANED — nothing references it, the
-    # per-user delete path keys off asset.rel_path (which is gone), and the GC orphan sweep skips the
-    # dot-prefixed `.uploads/` tree. So roll back the on-disk side too on any failure.
+    # asset/job rows roll back and nothing names the placed file: the per-user delete path keys off
+    # asset.rel_path (which is gone), and the GC's `.uploads` sweep reaches it only an hour later. So
+    # roll back the on-disk side too — unless the commit itself raised after it landed (see below).
+    asset_id = asset.id
     try:
         final = storage.abs_path(settings.download_root, rel)
         actual_size = final.stat().st_size if final.exists() else size
 
-        asset.status = "ready"
+        asset.status = "pending" if convert else "ready"
         asset.rel_path = rel
         asset.size_bytes = actual_size
-        asset.container = storage.upload_ext(filename)
+        asset.container = ext or None
         asset.vcodec = vcodec
         asset.acodec = acodec
         asset.width = width
@@ -100,8 +118,8 @@ def ingest_upload(
             source_ref=ref,
             profile_snapshot={},
             display_name=display_text(display_name),
-            status="done",
-            progress=100,
+            status=JobStatus.QUEUED if convert else JobStatus.DONE,
+            progress=0 if convert else 100,
             queue_pos=service.next_queue_pos(db),
         )
         db.add(job)
@@ -109,6 +127,11 @@ def ingest_upload(
         db.commit()
     except Exception:
         db.rollback()
+        if _row_landed(db, asset_id) is not False:
+            # The commit reached the server and lost its answer, or the row cannot be read: the file
+            # may be the asset's now, so it is never removed blind — the sweep takes it if no row
+            # names it.
+            raise
         # The rows are gone; remove the now-unreferenced file so no bytes leak. Clean the poster by its
         # DETERMINISTIC path (rel with a .jpg suffix, the same path ensure_poster writes to) rather than
         # ensure_poster's return value: it writes the frame directly to that path and returns None on an
@@ -120,3 +143,13 @@ def ingest_upload(
         raise
     db.refresh(job)
     return job
+
+
+def _row_landed(db: Session, asset_id: int) -> bool | None:
+    """After the ingest transaction raised: is the asset row there anyway? A commit can reach the
+    server and lose its answer (the connection drops after COMMIT). None when it cannot be read."""
+    try:
+        return db.get(MediaAsset, asset_id) is not None
+    except Exception:  # noqa: BLE001 — "cannot tell" is the answer, the caller acts on it
+        # silent-ok: the caller re-raises the ingest's own error.
+        return None

backend/app/email.py

@@ -197,7 +197,7 @@ def _next_step(verify_link: str | None, has_account: bool, hours: int, lang: str
 
     Three states, because a mail that guesses wrong sends someone looking for a link that isn't
     coming — or worse, tells them to sign in when their address is still unconfirmed and sign-in will
-    refuse them (E9.5 S4 named that state on screen; the mail must not contradict it).
+    refuse them (the sign-in screen names that state; the mail must not contradict it).
 
     The link branch is the point of the C option: in the common ordering — approve first, the person
     signs up later, or signs up and waits — they would otherwise have to find and open TWO separate
@@ -253,7 +253,7 @@ def send_access_invitation(
     the closing line offers no deadline — nudging is the one thing this mail must not do.
 
     Its language is the one place the app has no signal to read: the admin typed this address in, so
-    nobody has ever told us how to address its owner. That is why the send button asks (E9.5 S7).
+    nobody has ever told us how to address its owner. That is why the send button asks.
     """
     admin, contact = _reply_contact(lang)
     body = mailtext.text(

backend/app/feedquery.py

@@ -12,9 +12,10 @@ import binascii
 import json
 from datetime import date, datetime, timedelta, timezone
 
-from sqlalchemy import BigInteger, Select, and_, cast, false, func, or_, select
+from sqlalchemy import BigInteger, Select, and_, case, cast, false, func, or_, select
 from sqlalchemy.orm import Session, aliased
 
+from app.collation import collated
 from app.models import (
     LIVE_OR_UPCOMING,
     BlockedChannel,
@@ -105,15 +106,30 @@ def _tag_visible(user: User):
     return or_(ChannelTag.user_id.is_(None), ChannelTag.user_id == user.id)
 
 
-def tags_by_category(db: Session, tag_ids: list[int]) -> dict[str, list[int]]:
+# The pseudo-category of a selected tag id no Tag row answers (a saved view or the feed state outlives
+# a deleted tag). It filters like a category whose tags no channel carries — to nothing, in the SQL and
+# in feed_summary's mirror alike — instead of the id being dropped, which showed the WHOLE feed under
+# the view's old name. No real category is called this (language/topic/other).
+MISSING_TAG_CATEGORY = "missing"
+
+
+def tags_by_category(db: Session, user: User, tag_ids: list[int]) -> dict[str, list[int]]:
     """Group tag ids by their category (language/topic/other). The single source for the category split
     shared by the feed's tag WHERE (filtered_query) and the facet/count summary (feed_summary), so the
-    two can't disagree on which selected tag belongs to which category."""
+    two can't disagree on which selected tag belongs to which category. An unknown id lands in
+    MISSING_TAG_CATEGORY — and so does another user's private tag: resolving it to its real category
+    told the caller which foreign ids exist."""
     if not tag_ids:
         return {}  # skip the pointless `WHERE id IN ()` round-trip on the no-tag-filter hot path
     out: dict[str, list[int]] = {}
-    for tid, cat in db.execute(select(Tag.id, Tag.category).where(Tag.id.in_(tag_ids))).all():
+    visible = or_(Tag.user_id.is_(None), Tag.user_id == user.id)
+    for tid, cat in db.execute(
+        select(Tag.id, Tag.category).where(Tag.id.in_(tag_ids), visible)
+    ).all():
         out.setdefault(cat, []).append(tid)
+    known = {tid for ids in out.values() for tid in ids}
+    if missing := [tid for tid in tag_ids if tid not in known]:
+        out[MISSING_TAG_CATEGORY] = missing
     return out
 
 
@@ -264,7 +280,8 @@ def filtered_query(
     # the queries that surfaced the video > description; see Video.search_vector). YouTube-like
     # "fuzzy" matching: word-order-independent, multi-word AND, prefix on the word being typed,
     # accent-insensitive (unaccent_simple). The channel name still matches as a substring so
-    # channel searches work. `rank_expr` (weight-aware ts_rank) drives the "relevance" sort.
+    # channel searches work. `rank_expr` (weight-aware ts_rank) drives the "relevance" sort, over the
+    # newest RANK_WINDOW matches only (see there).
     rank_expr = None
     if q:
         # Resolve the channels whose NAME matches up front (there are only a few hundred channels, so
@@ -302,7 +319,7 @@ def filtered_query(
         # ALL its selected tags. ⚠️ feed_summary.passes() reimplements this exact rule in Python for the
         # sidebar count/facets — keep the two in lockstep (equivalence-tested by
         # test_feed_summary_matches_count_plus_facets).
-        by_category = tags_by_category(db, tags)
+        by_category = tags_by_category(db, user, tags)
         # Facet counting drops the category being counted so its own chips don't zero each
         # other out (standard drill-down faceting: a category's count ignores its own
         # selections but still honours the other categories' filters).
@@ -393,7 +410,7 @@ def feed_summary(db: Session, user: User, **params) -> dict:
         return {"count": 0, "facets": {}}
 
     # channel -> its visible tag ids, plus each tag's category (for grouping the facet output).
-    visible = or_(ChannelTag.user_id.is_(None), ChannelTag.user_id == user.id)
+    visible = _tag_visible(user)
     channel_tags: dict[str, set[int]] = {}
     tag_category: dict[int, str] = {}
     for cid, tid, cat in db.execute(
@@ -408,7 +425,7 @@ def feed_summary(db: Session, user: User, **params) -> dict:
     # base channel carries still filters everything out for its category — matching the SQL `IN (empty)`
     # behaviour).
     selected_by_cat: dict[str, set[int]] = {
-        cat: set(ids) for cat, ids in tags_by_category(db, params.get("tags") or []).items()
+        cat: set(ids) for cat, ids in tags_by_category(db, user, params.get("tags") or []).items()
     }
 
     def passes(ch_tags: set[int], exclude_cat: str | None) -> bool:
@@ -436,7 +453,7 @@ def feed_summary(db: Session, user: User, **params) -> dict:
         # Drop this category's own selections so its chips keep independent counts; a conjunctive
         # topic-AND keeps them (each remaining chip narrows to channels that ALSO have the selected).
         exclude = None if tag_modes.get(category, "or") == "and" else category
-        for cid, ch_tags in channel_tags.items():
+        for ch_tags in channel_tags.values():
             if not passes(ch_tags, exclude):
                 continue
             for tid in ch_tags:
@@ -446,7 +463,7 @@ def feed_summary(db: Session, user: User, **params) -> dict:
 
 
 def feed_years(db: Session, user: User, **params) -> list[dict]:
-    """Per-year video counts for the current filter context (item 8 / E8·S2): powers the year control's
+    """Per-year video counts for the current filter context: powers the year control's
     quick chips + the "more…" per-year-counts dialog. Every OTHER filter (tags, channels, content type,
     search, watch state, scope) still applies, but the publish-date narrowing itself is DROPPED —
     published_after/before and max_age_days — so the year list is the full set of years available and
@@ -478,8 +495,10 @@ def feed_years(db: Session, user: User, **params) -> list[dict]:
 # replaces OFFSET/LIMIT: deep scrolling no longer scans-and-discards skipped rows,
 # and the page boundary is stable even as the scheduler ingests new videos mid-scroll
 # (OFFSET would shift and duplicate/skip rows).
-def sort_keys(sort: str, seed: int, rank_expr=None) -> list[dict]:
+def sort_keys(sort: str, seed: int, rank_expr=None, lang: str | None = None) -> list[dict]:
+    """`lang` orders the title sorts in the reader's alphabet (`collation`); the other keys ignore it."""
     pub = {"expr": Video.published_at, "asc": False, "nullable": False, "kind": "dt"}
+    title = collated(Video.title, lang)
     table: dict[str, list[dict]] = {
         "newest": [pub],
         "oldest": [{**pub, "asc": True}],
@@ -491,10 +510,8 @@ def sort_keys(sort: str, seed: int, rank_expr=None) -> list[dict]:
         "duration_asc": [
             {"expr": Video.duration_seconds, "asc": True, "nullable": True, "kind": "num"}
         ],
-        "title": [{"expr": func.lower(Video.title), "asc": True, "nullable": True, "kind": "str"}],
-        "title_desc": [
-            {"expr": func.lower(Video.title), "asc": False, "nullable": True, "kind": "str"}
-        ],
+        "title": [{"expr": title, "asc": True, "nullable": True, "kind": "str"}],
+        "title_desc": [{"expr": title, "asc": False, "nullable": True, "kind": "str"}],
         "subscribers": [
             {"expr": Channel.subscriber_count, "asc": False, "nullable": True, "kind": "num"}
         ],
@@ -542,6 +559,54 @@ def sort_keys(sort: str, seed: int, rank_expr=None) -> list[dict]:
     return table.get(sort) or table["newest"]
 
 
+# Relevance ranks only the newest RANK_WINDOW matches: ts_rank reads every match's tsvector, and
+# a broad word matches most of the library ("a" ranked 193k rows in 6.8 s on prod, against 1.1 s to just
+# find them). The rows past the window keep rank 0, so they follow the ranked ones newest first; a word
+# that broad has no meaningful relevance order anyway, and a specific one fits the window whole.
+RANK_WINDOW = 2000
+
+
+def rank_window_start(db: Session, query: Select) -> datetime | None:
+    """The published_at the ranked window starts at: the RANK_WINDOW-th newest match, None when fewer
+    match (rank them all). The newest-first walk stops after RANK_WINDOW matches (40 ms for "a")."""
+    return db.scalar(
+        query.with_only_columns(Video.published_at)
+        .order_by(Video.published_at.desc())
+        .offset(RANK_WINDOW - 1)
+        .limit(1)
+    )
+
+
+def windowed_rank(rank_expr, since: datetime | None):
+    """`rank_expr` inside the window, 0 past it. CASE evaluates ts_rank only for the rows inside."""
+    if since is None:
+        return rank_expr
+    return case((Video.published_at >= since, rank_expr), else_=0)
+
+
+def windowed_relevance_rows(
+    db: Session, query: Select, keys: list[dict], window: datetime, cursor_id: str | None, n: int
+) -> list:
+    """Up to `n` rows of the windowed relevance order, read as its two halves: the window ranked, then
+    the rest newest first (every window row sorts before every row past it). As one ORDER BY the
+    planner computes the CASE for every match and sorts them all (2.7 s for "a" on prod); split, each
+    half rides an index (80 ms + 5 ms). With a cursor, `keys` carry its values (keyset_predicate)."""
+    pub_key = keys[1]  # relevance: [rank, published_at]
+    rows: list = []
+    if cursor_id is None or pub_key["value"] >= window:
+        inside = query.where(Video.published_at >= window).order_by(*order_by(keys))
+        if cursor_id is not None:
+            inside = inside.where(keyset_predicate(keys, cursor_id))
+        rows = list(db.execute(inside.limit(n)).all())
+    if len(rows) < n:
+        newest = [pub_key]  # past the window every rank is 0: the order is newest first
+        rest = query.where(Video.published_at < window).order_by(*order_by(newest))
+        if cursor_id is not None:
+            rest = rest.where(keyset_predicate(newest, cursor_id))
+        rows += db.execute(rest.limit(n - len(rows))).all()
+    return rows
+
+
 def order_by(keys: list[dict]):
     cols = []
     for k in keys:
@@ -588,26 +653,45 @@ def keyset_predicate(keys: list[dict], cursor_id: str):
     return rec(0)
 
 
-def encode_cursor(keys: list[dict], row) -> str:
+def encode_cursor(keys: list[dict], row, window: datetime | None = None) -> str:
+    """`window` is the relevance window's start: the cursor carries it, so every page ranks against the
+    window page one did. Recomputed per page, a video ingested mid-scroll would push the window's oldest
+    row to rank 0, behind the cursor, and the reader would see it twice."""
     vals = []
     for i, k in enumerate(keys):
         v = getattr(row, f"_k{i}")
         vals.append(v.isoformat() if (v is not None and k["kind"] == "dt") else v)
-    payload = json.dumps({"k": vals, "id": row.id})
+    body: dict = {"k": vals, "id": row.id}
+    if window is not None:
+        body["w"] = window.isoformat()
+    payload = json.dumps(body)
     return base64.urlsafe_b64encode(payload.encode()).decode()
 
 
-def decode_cursor(keys: list[dict], cursor: str) -> tuple[list, str]:
+def _cursor_dt(v):
+    """A cursor's timestamp: aware or refused. The server writes only aware ones (timestamptz), and
+    windowed_relevance_rows compares them in Python, where a naive one beside an aware one raises — a
+    500 for a hand-made cursor instead of the 400."""
+    if v is None:
+        return None
+    dt = datetime.fromisoformat(v)
+    if dt.tzinfo is None:
+        raise ValueError("naive cursor timestamp")
+    return dt
+
+
+def decode_cursor(keys: list[dict], cursor: str) -> tuple[list, str, datetime | None]:
     try:
         payload = json.loads(base64.urlsafe_b64decode(cursor.encode()))
         raw = payload["k"]
         if len(raw) != len(keys):
             raise ValueError("key arity mismatch")
-        vals = [
-            datetime.fromisoformat(v) if (v is not None and k["kind"] == "dt") else v
-            for k, v in zip(keys, raw)
-        ]
-        return vals, payload["id"]
+        vals = [_cursor_dt(v) if k["kind"] == "dt" else v for k, v in zip(keys, raw, strict=True)]
+        for k, v in zip(keys, vals, strict=True):
+            if v is None and not k["nullable"]:
+                raise ValueError("null value for a non-nullable key")
+        window = payload.get("w")
+        return vals, payload["id"], _cursor_dt(window) if window else None
     except (binascii.Error, json.JSONDecodeError, KeyError, ValueError, TypeError) as exc:
         # Framework-free: the route handler translates this into an HTTP 400 so the
         # read-model stays web-layer-agnostic (see routes/feed.get_feed).

backend/app/issues/__init__.py

@@ -1,3 +1,3 @@
-"""Bug reports built from debug sessions (E16): the files an issue carries, and where they go —
-downloaded as a zip, or (S4) sent to a Gitea issue tracker. The session itself — what is recorded
+"""Bug reports built from debug sessions: the files an issue carries, and where they go —
+downloaded as a zip, or sent to a Gitea issue tracker. The session itself — what is recorded
 and tagged — is the core module `app/debugsession.py`."""

backend/app/issues/bundle.py

@@ -5,7 +5,7 @@ Gitea upload, so the two cannot drift apart.
     in their language; the server does not rewrite it);
   * `client.log` / `server.log` — the session's evidence (`debugsession.bundle`) minus the rows the
     user took out, one JSON object per line, browser rows and server rows apart. `.log`, not
-    `.jsonl`: Gitea refuses that extension (measured, plans/e16 S0);
+    `.jsonl`: Gitea refuses that extension (measured);
   * the screenshots, in the order they were taken.
 
 Every row is already redacted twice (at capture, and again when stored), and nothing here adds a

backend/app/issues/gitea.py

@@ -1,5 +1,5 @@
-"""The Gitea issue tracker a debug session's issue goes to (E16 S4) — measured against Gitea 1.27.1
-before a line of this was written (plans/e16-debug-issue.md, S0).
+"""The Gitea issue tracker a debug session's issue goes to — measured against Gitea 1.27.1
+before a line of this was written.
 
 Configured by an admin (sysconfig `issues_gitea_url` / `_repo` / `_token`, the token encrypted at
 rest). The repository must be PRIVATE: an issue carries its reporter's log — page titles, filter
@@ -79,7 +79,7 @@ def target(db: Session) -> dict | None:
 @contextmanager
 def _client(settings: tuple[str, str, str]) -> Iterator[tuple[httpx.Client, str]]:
     # Takes the settings already read, not the session: reading them HERE opened a transaction one
-    # line after `deliver` had ended one, and held it across the calls below (deep-review round 4).
+    # line after `deliver` had ended one, and held it across the calls below.
     url, repo, token = settings
     if not (url and repo and token):
         raise GiteaError("not_configured")
@@ -203,7 +203,7 @@ def deliver(
     # call, so nothing holds a snapshot (and an idle connection) across minutes of tracker I/O.
     # The settings are read HERE and handed to `_client`, not read inside it: reading them there
     # was a SELECT one line after this commit, which reopened the transaction and held it — idle,
-    # with its snapshot pinned — across the repository check and the create (deep-review round 4
+    # with its snapshot pinned — across the repository check and the create (a review
     # measured it in pg_stat_activity). A comment saying otherwise sat right here.
     checkpoint()
     with _client(settings) as (client, repo):
@@ -223,7 +223,7 @@ def deliver(
             )
             # Built ON the state read at the start so nothing already in it is dropped. The claim
             # is NOT among those things, whatever this comment used to say: `set_delivery` strips
-            # any `sending` passed to it and stamps the row's own from `token` (round 4 — removing
+            # any `sending` passed to it and stamps the row's own from `token` (removing
             # the spread left the whole suite green, because the claim never travelled here).
             state = {
                 **state,
@@ -239,7 +239,7 @@ def deliver(
                 # already open on the tracker and this is what would have recorded its link. We
                 # still must not write — another send owns the row and is opening its own issue —
                 # so say which issue was left behind, because nothing else will.
-                # noqa TRY400: the traceback is one SendClaimLost carrying the session id and
+                # TRY400: the traceback is one SendClaimLost carrying the session id and
                 # nothing else; the issue number below is the whole news.
                 log.error(  # noqa: TRY400
                     "an issue was created by a send that had already lost its claim",

backend/app/logs.py

@@ -10,7 +10,7 @@ the logs were both unjoinable and leaky. `configure_logging()` replaces all four
 Event shape — the OpenTelemetry log data model, without the SDK; one JSON object per line:
     ts · level · sev · source · logger · event · msg · attrs
 plus whatever correlation the context carries (`request_id`, `user_id`, `debug_session_id`,
-`job_id`) and the app version. That shape is what E22.6's admin console filters on and what a Gitea
+`job_id`) and the app version. That shape is what the admin log console filters on and what a Gitea
 issue attachment will carry, so it is deliberately flat and stable.
 
 Call sites stay ordinary stdlib logging:
@@ -106,7 +106,7 @@ class _RequestScope:
         self.user_id: int | None = None
         # The user's consented diagnostics level as a logging number; None at the base level.
         self.level: int | None = None
-        # The user's own open debug session this request named (E16), else None.
+        # The user's own open debug session this request named, else None.
         self.debug_session_id: str | None = None
 
 
@@ -122,11 +122,11 @@ def set_user(
     debug_session: str | None = None,
 ) -> None:
     """Attribute the rest of this request's log lines to `user_id`, and log them at the user's
-    consented diagnostics `level` until `until` (E22 S5; "INFO" = no elevation). Called from the
+    consented diagnostics `level` until `until` ("INFO" = no elevation). Called from the
     auth dependency — the one place the session is actually resolved into an account — rather than
     from the request middleware, which runs OUTSIDE the session middleware.
 
-    `debug_session` tags every line with the user's open debug session (E16). The caller has
+    `debug_session` tags every line with the user's open debug session. The caller has
     already checked it is theirs and open; a session's level arrives here as `level`/`until`.
 
     The root logger's floor is tracked per elevation: a request without the session's header (an
@@ -166,7 +166,7 @@ def bind(**fields: Any):
     try:
         yield
     finally:
-        for key, token in zip(keys, tokens):
+        for key, token in zip(keys, tokens, strict=True):
             _CONTEXT_VARS[key].reset(token)
 
 
@@ -334,7 +334,7 @@ _RECORD_STANDARD = frozenset(
 class ContextFilter(logging.Filter):
     """Copies the correlation contextvars onto the record. A Filter (not the formatter) so the
     values are captured at emit time, before a QueueHandler can move the record to another
-    thread — the DB sink in S3 depends on that."""
+    thread — the DB sink depends on that."""
 
     def filter(self, record: logging.LogRecord) -> bool:
         for name, var in _CONTEXT_VARS.items():
@@ -355,7 +355,7 @@ class ContextFilter(logging.Filter):
 
 class LevelGate(logging.Filter):
     """The configured level, enforced at the HANDLER, with one exception: the lines of a request
-    whose user raised their diagnostics level (E22 S5).
+    whose user raised their diagnostics level.
 
     The root logger's level cannot do this alone: it is one number for the whole process, and the
     point is to log ONE user's requests in detail while everyone else's stay at the base level. So
@@ -457,7 +457,7 @@ class TextFormatter(logging.Formatter):
 # --- configuration -------------------------------------------------------------------------
 # Third-party loggers that are noisy at our own level. Pinned explicitly so the diagnostics knob can
 # lower the ROOT level to DEBUG/TRACE without drowning the output in SQL statements and HTTP chatter
-# — raising THESE is a deliberate, separate act (E22.6's console).
+# — raising THESE is a deliberate, separate act (the admin log console).
 _THIRD_PARTY_LEVELS = {
     "sqlalchemy": "WARNING",
     "sqlalchemy.engine": "WARNING",
@@ -481,7 +481,7 @@ def get_logger(name: str) -> logging.Logger:
     `siftlode.plex.sync`.
 
     The names were literals before, and 25 of them covered 34 modules — six different Plex modules
-    all called themselves `siftlode.plex`, so neither a level pinned per logger nor E22.6's console
+    all called themselves `siftlode.plex`, so neither a level pinned per logger nor the console's
     filter could tell a sync line from a stream line. Derived from `__name__` there is exactly one
     name per module, it cannot drift from the file it sits in, and `test_logger_names.py` fails if a
     module hand-rolls one again.
@@ -562,12 +562,12 @@ def current_level() -> str:
     return logging.getLevelName(_base_level or logging.getLogger().getEffectiveLevel())
 
 
-# --- per-user elevation (E22 S5) -----------------------------------------------------------
+# --- per-user elevation -----------------------------------------------------------
 # The base level (`LOG_LEVEL`, then the admin knob). NOTSET until configure_logging runs — a unit
 # test, a script — so the gate lets everything through there, as it did before the gate existed.
 _base_level = logging.NOTSET
 # → (level number, expiry) for every elevation this process has seen and not yet expired: a user's
-# own, by user id, and a debug session's (E16), by (user id, session id) — its requests alone carry
+# own, by user id, and a debug session's, by (user id, session id) — its requests alone carry
 # its level, so it is tracked apart from the user's own requests.
 _ElevationKey = int | tuple[int, str]
 _elevations: dict[_ElevationKey, tuple[int, datetime]] = {}
@@ -667,7 +667,7 @@ def current_source() -> str:
 
 
 # --- request logging -----------------------------------------------------------------------
-# A client may supply its own request id so a browser log line and a server line join up (E22 S4).
+# A client may supply its own request id so a browser log line and a server line join up.
 # Accepted only in this shape: it ends up in a log field, and an unbounded header would be a way to
 # write arbitrary text into the log.
 _REQUEST_ID_RE = re.compile(r"^[A-Za-z0-9._-]{8,64}$")
@@ -688,7 +688,6 @@ _POLLED_ROUTES = frozenset(
     {
         "/api/me",
         "/api/me/notifications/unread_count",
-        "/api/messages/unread_count",
         "/api/downloads",
         "/api/downloads/index",
         "/api/downloads/usage",

backend/app/logsink.py

@@ -1,4 +1,4 @@
-"""The `log_events` sink (E22 S3): the stored copy of the log the diagnostics console reads.
+"""The `log_events` sink: the stored copy of the log the diagnostics console reads.
 
 stdout stays the complete log (docker keeps 3×50 MB of it); this table keeps the part worth querying
 for `log_retention_days`:
@@ -6,7 +6,7 @@ for `log_retention_days`:
   * every WARNING and above,
   * the `http.request` line of every request at INFO or above (not the DEBUG-level health checks and
     static files),
-  * every line of a request whose user raised their diagnostics level (E22 S5 — `logs.LevelGate`
+  * every line of a request whose user raised their diagnostics level (`logs.LevelGate`
     marks them), which is the detail they consented to have kept,
   * nothing a caller marked `extra={logs.PERSIST: False}` — how the request log caps anonymous
     traffic per client.
@@ -347,8 +347,8 @@ def gc(db: Session, retention_days: int) -> int:
 
 
 def serialize(row: LogEvent) -> dict:
-    """A stored row as the API sends it — one shape for every reader: the admin console (E22.6) and
-    a debug session's evidence list (E16)."""
+    """A stored row as the API sends it — one shape for every reader: the admin console and
+    a debug session's evidence list."""
     return {
         "id": row.id,
         "ts": iso(row.ts),

backend/app/mailtext.py

@@ -1,5 +1,5 @@
 """Bilingual text for every outbound mail, and the one resolver that decides which language a
-recipient gets (E9.5 S7).
+recipient gets.
 
 Why a catalog instead of literals in `email.py`: HU/EN parity is a house rule for every user-facing
 string, and a mail is the one user-facing surface the reader CANNOT re-render by switching the UI
@@ -53,7 +53,7 @@ def lang_for_user(user: User | None) -> str:
 
 def lang_for_email(db: Session, email: str) -> str:
     """The language for an address that may not have an account yet: their own account preference,
-    else the language they filled our access-request form in, else English.
+    else the language their access request came in (the locale Google reported), else English.
 
     Matched CANONICALLY like every other identity lookup in this epic, and it walks past rows that
     carry no usable code instead of stopping at the first one — prod really does hold two accounts
@@ -82,8 +82,8 @@ def lang_for_email(db: Session, email: str) -> str:
 # a name, not a translated string, and giving them keys would invite someone to "translate" them.
 MAIL_TEXT: dict[str, dict[str, str]] = {
     # ── How long a link lasts ───────────────────────────────────────────────────────────────
-    # Rendered, not interpolated as a raw hour count: the approval mail's token lives seven days
-    # (E9.5 S6), and "expires in 168 hours" is a unit no reader converts.
+    # Rendered, not interpolated as a raw hour count: the approval mail's token lives seven days,
+    # and "expires in 168 hours" is a unit no reader converts.
     "duration.hours": {"en": "{n} hours", "hu": "{n} óra"},
     "duration.days": {"en": "{n} days", "hu": "{n} nap"},
     # ── Shared closing fragments ────────────────────────────────────────────────────────────────

backend/app/main.py

@@ -1,37 +1,36 @@
 import asyncio
 import math
+import mimetypes
 from contextlib import asynccontextmanager, suppress
 from pathlib import Path
 
 from fastapi import FastAPI, HTTPException, Request
 from fastapi.encoders import jsonable_encoder
 from fastapi.exceptions import RequestValidationError
-
-# Logging is configured HERE, before the route modules are imported, so every logger they create at
-# import time is already covered. ONE format for the api, the worker, the download child and alembic
-# alike (app/logs.py) — this replaces the old uvicorn --log-config + the hand-rolled fallback that
-# used to live here, and uvicorn's access log is replaced by the `http.request` middleware below.
-from app import logs
-from app.logs import configure_logging, current_level, log_requests, set_level
-
-configure_logging("api")
-
-log = logs.get_logger(__name__)
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import FileResponse, HTMLResponse, JSONResponse
 from fastapi.staticfiles import StaticFiles
 from sqlalchemy.exc import DataError
 from starlette.middleware.sessions import SessionMiddleware
 
-from app import bodylimit
-from app import logsink
-from app import state
-from app import sysconfig
-from app.config import settings
-from app.db import SessionLocal
-from app.plex import stream as plex_stream
-from app.plex import subtitles as plex_subtitles
-from app.routes import (
+# Logging is configured HERE, before the app's own modules are imported, so a line one of them logs
+# WHILE importing already has the format (a logger is configured whenever it is created —
+# app/logs.py — but a line emitted before `configure_logging` goes out raw). ONE format for the api,
+# the worker, the download child and alembic alike — this replaces the old uvicorn --log-config +
+# the hand-rolled fallback that used to live here, and uvicorn's access log is replaced by the
+# `http.request` middleware below. The imports after it are E402 on purpose, each one marked.
+from app import logs
+from app.logs import configure_logging, current_level, log_requests, set_level
+
+configure_logging("api")
+
+log = logs.get_logger(__name__)
+from app import bodylimit, collation, logsink, state, sysconfig  # noqa: E402 — logging is configured first
+from app.config import settings  # noqa: E402 — logging is configured first
+from app.db import SessionLocal, engine  # noqa: E402 — logging is configured first
+from app.plex import stream as plex_stream  # noqa: E402 — logging is configured first
+from app.plex import subtitles as plex_subtitles  # noqa: E402 — logging is configured first
+from app.routes import (  # noqa: E402 — logging is configured first
     admin,
     admin_logs as admin_logs_routes,
     audit as audit_routes,
@@ -45,7 +44,6 @@ from app.routes import (
     feed,
     health,
     me,
-    messages,
     notifications,
     playlists,
     plex as plex_routes,
@@ -59,7 +57,7 @@ from app.routes import (
     tags,
     version,
 )
-from app.scheduler import shutdown_scheduler, start_scheduler
+from app.scheduler import shutdown_scheduler, start_scheduler  # noqa: E402 — logging is configured first
 
 
 @asynccontextmanager
@@ -76,7 +74,7 @@ async def lifespan(app: FastAPI):
         # booting — an unreachable or not-yet-migrated database has to start into the env level.
         try:
             set_level(sysconfig.get_str(db, "log_level"))
-        except Exception:  # noqa: BLE001 — any DB/config failure keeps the bootstrap level
+        except Exception:  # any DB/config failure keeps the bootstrap level
             # With the traceback: this catch spans a connection, a query and a coercion at boot,
             # and `%s` of the exception does not say which of them raised.
             log.exception("could not read the stored log level; staying at %s", current_level())
@@ -97,6 +95,9 @@ async def lifespan(app: FastAPI):
     # owns, so racing it against early playback is safe.
     sweep = asyncio.create_task(_sweep_hls_orphans())
     reaper = asyncio.create_task(_reap_hls_sessions())
+    # Fire and forget as well: a rebuild after an ICU upgrade takes seconds, and the title sorts serve
+    # meanwhile (CONCURRENTLY) in the old index's order, as they did before the restart.
+    heal = asyncio.create_task(_heal_collation_drift())
     start_scheduler()
     try:
         yield
@@ -106,7 +107,7 @@ async def lifespan(app: FastAPI):
         # pending ("Task was destroyed but it is pending!" — noise that reads like a bug during
         # triage). Cancelling does NOT interrupt the rmtree thread itself; this only makes our own
         # bookkeeping honest.
-        for task in (sweep, reaper):
+        for task in (sweep, reaper, heal):
             task.cancel()
             with suppress(asyncio.CancelledError):
                 await task
@@ -130,7 +131,7 @@ async def _reap_hls_sessions() -> None:
     A FINISHED session is reaped on the short rule here, not kept for the full idle window: an open
     player checkpoints every 10 s and that route touches the session, so a paused viewer is no longer
     invisible to this pass (see `reap_idle`). Keeping every finished session for ten minutes instead
-    held one whole segmented copy of a film per item on the download volume (round 5)."""
+    held one whole segmented copy of a film per item on the download volume."""
     while True:
         await asyncio.sleep(_HLS_REAP_EVERY_S)
         try:
@@ -139,6 +140,18 @@ async def _reap_hls_sessions() -> None:
             log.exception("HLS session reap failed")
 
 
+async def _heal_collation_drift() -> None:
+    """Background startup check of the ICU collation the title-sort indexes are built in (see
+    `collation.heal_drift`). A failure is logged, never fatal: the sorts still answer, only possibly in
+    the old order."""
+    try:
+        await asyncio.to_thread(collation.heal_drift, engine)
+    except asyncio.CancelledError:
+        raise
+    except Exception:
+        log.exception("ICU collation drift check failed")
+
+
 async def _sweep_hls_orphans() -> None:
     """Background half of the startup HLS sweep (see lifespan). Owns its own errors: a failure here
     must never take the app down, and a cancelled shutdown is not one."""
@@ -174,8 +187,8 @@ async def _nul_in_text(request: Request, exc: DataError) -> JSONResponse:
     NUL in text, so the driver refuses to even send the query — and a `%00` in a query parameter or
     an escaped NUL in a JSON body (a playlist's `contains=`, a reorder's key list, a collection
     search) was a 500 with a traceback wherever it reached a comparison. Guarding each site did not
-    hold: round 3 guarded the by-rating_key lookup and missed the rest. Any other DataError is left
-    alone — re-raised, it stays the 500 it always was."""
+    hold: an earlier fix guarded the by-rating_key lookup and missed the rest. Any other DataError
+    is left alone — re-raised, it stays the 500 it always was."""
     if "NUL (0x00)" not in str(exc.orig):
         raise exc
     return JSONResponse(
@@ -297,7 +310,6 @@ app.include_router(me.router)
 app.include_router(diagnostics_routes.router)
 app.include_router(debug_sessions_routes.router)
 app.include_router(notifications.router)
-app.include_router(messages.router)
 app.include_router(channels.router)
 app.include_router(playlists.router)
 app.include_router(saved_views.router)
@@ -316,8 +328,6 @@ app.include_router(setup_routes.router)
 
 # Ensure modern image types resolve to the right Content-Type when FileResponse guesses from the
 # filename (the runtime's mimetypes db doesn't always know .webp → it'd fall back to octet-stream).
-import mimetypes
-
 mimetypes.add_type("image/webp", ".webp")
 mimetypes.add_type("image/avif", ".avif")
 
@@ -329,6 +339,8 @@ STATIC_DIR = Path(__file__).parent / "static_spa"
 # stay cached but forces revalidation (cheap 304 via the FileResponse ETag) on every load. The
 # hashed bundles under /assets can cache forever (a new build changes their filename).
 INDEX_HTML = STATIC_DIR / "index.html"
+# A public share link's own page: the player alone, without the app's bundle (frontend/watch.html).
+WATCH_HTML = STATIC_DIR / "watch.html"
 INDEX_HEADERS = {"Cache-Control": "no-cache"}
 app.mount(
     "/assets",
@@ -344,16 +356,16 @@ async def index() -> FileResponse:
 
 @app.get("/watch/{token}")
 async def watch_page(token: str):
-    """Serve the SPA for a public share link, but with per-video Open Graph tags injected so the
+    """Serve the public player page for a share link, with per-video Open Graph tags injected so the
     link unfurls richly (title/channel/thumbnail) in Messenger/social crawlers. A real browser
-    ignores the extra tags and hydrates WatchPage as usual; invalid/expired/password links fall
-    back to the generic card."""
+    ignores the extra tags and boots WatchPage; invalid/expired/password links fall back to the
+    generic card."""
     from app.downloads import og
 
     with SessionLocal() as db:
-        rendered = og.render_watch_html(token, INDEX_HTML, db)
+        rendered = og.render_watch_html(token, WATCH_HTML, db)
     if rendered is None:
-        return FileResponse(INDEX_HTML, headers=INDEX_HEADERS)
+        return FileResponse(WATCH_HTML, headers=INDEX_HEADERS)
     return HTMLResponse(rendered, headers=INDEX_HEADERS)
 
 
@@ -371,4 +383,8 @@ async def spa_fallback(full_path: str) -> FileResponse:
             # Real SPA-root assets (welcome images, favicon, robots.txt) rarely change and have
             # stable names, so a moderate cache is safe and satisfies the cache-lifetime audit.
             return FileResponse(candidate, headers={"Cache-Control": "public, max-age=604800"})
+    # A /watch/ path the route above does not match (a trailing segment, no token) is still the
+    # player's: WatchPage reads the token off the path and says when a link is gone.
+    if full_path.startswith("watch/"):
+        return FileResponse(WATCH_HTML, headers=INDEX_HEADERS)
     return FileResponse(INDEX_HTML, headers=INDEX_HEADERS)

backend/app/models.py

@@ -23,6 +23,7 @@ from sqlalchemy.dialects.postgresql import JSONB, TSVECTOR
 from sqlalchemy.orm import Mapped, mapped_column, relationship, validates
 
 from app.db import Base
+from app.utils import canonical_email
 
 # Every user-supplied display string in this schema shares one width, and it lives HERE so the
 # columns below are declared with it rather than with a literal: a name a person typed goes into
@@ -51,9 +52,6 @@ def display_text(raw: str | None) -> str | None:
     return (raw or "").strip()[:DISPLAY_TEXT_MAX] or None
 
 
-from app.utils import canonical_email
-
-
 # The resume-position columns (VideoState / PlexState.position_seconds) are int4. The checkpoint
 # routes bound their bodies by this, so an out-of-range position is a 422 rather than a 500.
 POSITION_MAX_S = 2**31 - 1
@@ -61,7 +59,7 @@ POSITION_MAX_S = 2**31 - 1
 
 class TimestampMixin:
     """Adds a server-defaulted `created_at`. Models that index created_at or order it
-    differently (QuotaEvent, Notification, Message) keep their own column on purpose."""
+    differently (QuotaEvent, Notification) keep their own column on purpose."""
 
     created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())
 
@@ -103,15 +101,15 @@ class User(Base, TimestampMixin, CanonicalEmailMixin):
     # Google sign-in that failed to find it), and a migration must not merge accounts on its own.
     email_canonical: Mapped[str] = mapped_column(String(320), index=True, server_default="")
     # The address Google asserted for this account, recorded at sign-in / link time. NOT the same
-    # thing as `email`: since E9.5 S3 a Google sign-in no longer overwrites the address the user
+    # thing as `email`: a Google sign-in no longer overwrites the address the user
     # registered with, so a deliberate `me+siftlode@gmail.com` survives — and this is where the other
     # half of the truth lives, for the UI to show when the two differ. NULL until the next Google
     # sign-in for accounts linked before this column existed.
     google_email: Mapped[str | None] = mapped_column(String(320))
-    # argon2id hash for email+password login; NULL for Google-only / demo accounts.
+    # argon2id hash for email+password login; NULL for Google-only accounts.
     password_hash: Mapped[str | None] = mapped_column(String(255))
     # When SEC-10 dropped a preset password on Google adoption (see the single caller of
-    # auth.revoke_preset_password_if_unverified; admin approval no longer revokes — E9.5 S1). Without
+    # auth.revoke_untrusted_preset_password; admin approval no longer revokes). Without
     # this trace a revoked account was indistinguishable from a Google-only one, so it failed the
     # password-reset eligibility check and could never be recovered — it could neither sign in nor
     # reset. Set = "had a password, lost it to the guard" -> reset-eligible. Cleared whenever a
@@ -125,7 +123,7 @@ class User(Base, TimestampMixin, CanonicalEmailMixin):
     is_active: Mapped[bool] = mapped_column(Boolean, default=True, server_default="true")
     # Admin-imposed access block, distinct from is_active (approval lifecycle). A suspended
     # account can't sign in by any method and its live sessions are rejected; the user is told
-    # why and pointed at the operator's contact. The demo account is never suspendable.
+    # why and pointed at the operator's contact.
     is_suspended: Mapped[bool] = mapped_column(Boolean, default=False, server_default="false")
     # Server-side session revocation (SA4): a monotonic counter bumped on password reset, password
     # change, or "log out everywhere". A signed session cookie records the epoch it was minted under;
@@ -134,9 +132,9 @@ class User(Base, TimestampMixin, CanonicalEmailMixin):
     session_epoch: Mapped[int] = mapped_column(Integer, default=0, server_default="0")
     # When this account last signed in (Google or password), stamped by the one sign-in path,
     # routes/auth._establish_session. NULL = not since the column existed (0085; no backfill is
-    # possible). The shared demo account is never stamped — its "last login" would mean nobody.
+    # possible).
     last_login_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
-    # The diagnostics level this user consented to (E22 S5) and when it lapses back to the base
+    # The diagnostics level this user consented to and when it lapses back to the base
     # level. Both NULL = base. Written only by `PUT /api/me/diagnostics`, which computes the expiry
     # itself; read through `app.diagnostics.effective_level`, never directly (an expired level is
     # not cleared eagerly — it simply stops counting).
@@ -148,12 +146,6 @@ class User(Base, TimestampMixin, CanonicalEmailMixin):
     # The user's own YouTube channel id, resolved lazily (channels.list?mine=true) and
     # cached. Used to exclude their own channel from playlist discovery. NULL = unresolved.
     yt_channel_id: Mapped[str | None] = mapped_column(String(32))
-    # The single shared "demo" account: signed into without Google OAuth (via a whitelisted
-    # email on the login page), has no OAuth token / YouTube scope, and its per-user state is
-    # shared by everyone who enters this way. Exactly one such row is expected.
-    is_demo: Mapped[bool] = mapped_column(
-        Boolean, default=False, server_default="false", index=True
-    )
     # Free-form UI preferences (theme, color scheme, font scale, default filters…).
     preferences: Mapped[dict | None] = mapped_column(JSON)
 
@@ -220,25 +212,12 @@ class Invite(Base, CanonicalEmailMixin):
     # The language this person asked us in ("en"/"hu"), captured when the request was made: the page
     # their form was on, or the locale Google reported for a refused sign-in. NULL for a row the
     # admin whitelisted by hand — nobody has told us anything about that mailbox, which is exactly
-    # why the invitation send asks (E9.5 S7). Captured rather than derived for the same reason
+    # why the invitation send asks. Captured rather than derived for the same reason
     # `preferences.language` and `google_email` are: the signal exists only in that one moment and
     # cannot be backfilled afterwards.
     lang: Mapped[str | None] = mapped_column(String(8))
 
 
-class DemoWhitelist(Base, TimestampMixin):
-    """Emails that may enter the shared demo account from the login page (no Google OAuth).
-    A hidden, admin-curated list of "keys to the same door" — every entry logs into the one
-    shared demo user. Distinct from Invite (which gates real Google sign-in)."""
-
-    __tablename__ = "demo_whitelist"
-
-    id: Mapped[int] = mapped_column(primary_key=True)
-    email: Mapped[str] = mapped_column(String(320), unique=True, index=True)
-    note: Mapped[str | None] = mapped_column(Text)
-    added_by: Mapped[str | None] = mapped_column(String(320))  # admin email
-
-
 class Channel(Base, TimestampMixin):
     """A YouTube channel. Shared across all users (one channel's videos are the same
     for everyone), so its expensive metadata is fetched and stored only once."""
@@ -330,10 +309,12 @@ class Video(Base, TimestampMixin):
     __tablename__ = "videos"
 
     id: Mapped[str] = mapped_column(String(16), primary_key=True)  # video id
-    # No standalone channel_id index: the composite `ix_videos_channel_pub_desc`
-    # (channel_id, published_at DESC NULLS LAST, id) from migration 0067 leads with channel_id, so it
-    # serves every channel_id lookup (FK cascade, sync upsert, channel-detail) AND the per-channel feed
-    # sort. Like the search_vector GIN, this composite is migration-defined, not an inline `index=True`.
+    # No standalone channel_id index: the composite `ix_videos_channel_pub_cover`
+    # (channel_id, published_at DESC NULLS LAST, id) from migration 0095 leads with channel_id, so it
+    # serves every channel_id lookup (FK cascade, sync upsert, channel-detail) and the per-channel feed
+    # sort. Its INCLUDEd columns make the channel manager's sums and the feed sidebar's counts Index Only
+    # Scans (no heap): a column those queries start reading belongs in it (test_channel_list_index_only).
+    # Like the search_vector GIN, this composite is migration-defined, not an inline `index=True`.
     channel_id: Mapped[str] = mapped_column(ForeignKey("channels.id", ondelete="CASCADE"))
     title: Mapped[str | None] = mapped_column(Text)  # normalized for display (see app.titles)
     original_title: Mapped[str | None] = mapped_column(Text)  # raw YouTube title, preserved
@@ -347,6 +328,9 @@ class Video(Base, TimestampMixin):
         DateTime(timezone=True), index=True, nullable=False, server_default=func.now()
     )
     thumbnail_url: Mapped[str | None] = mapped_column(String(1024))
+    # The feed's views / duration / title sorts each ride a migration-defined index in their exact
+    # ORDER BY, one per direction (and per ICU collation for title): migration 0096, held to
+    # `feedquery.order_by` by test_feed_sort_indexes.
     duration_seconds: Mapped[int | None] = mapped_column(Integer)
     view_count: Mapped[int | None] = mapped_column(BigInteger)
     like_count: Mapped[int | None] = mapped_column(BigInteger)
@@ -583,7 +567,7 @@ class AuditLog(Base):
 
 
 class LogEvent(Base):
-    """One stored log event (E22 S3) — the `log_events` table the diagnostics console reads. Rows are
+    """One stored log event — the `log_events` table the diagnostics console reads. Rows are
     written by app/logsink.py's background writer, never inside a request's transaction, already
     redacted (the same `logs.record_fields` the stdout line is built from).
 
@@ -596,7 +580,7 @@ class LogEvent(Base):
     __tablename__ = "log_events"
     __table_args__ = (
         Index("ix_log_events_user_ts", "user_id", "ts"),
-        # E16: an issue's bundle is `WHERE debug_session_id = …`; nearly every row has none.
+        # An issue's bundle is `WHERE debug_session_id = …`; nearly every row has none.
         Index(
             "ix_log_events_debug_session",
             "debug_session_id",
@@ -622,7 +606,7 @@ class LogEvent(Base):
 
 
 class DebugSession(Base):
-    """One recording: Start → reproduce → Stop (E16, app/debugsession.py). While it is open, every
+    """One recording: Start → reproduce → Stop (app/debugsession.py). While it is open, every
     request of its user that names it (`X-Debug-Session`) is logged at `level` and tagged with `id`,
     and the issue it ends in is built from the rows carrying that tag. `until` is the hard cap the
     server set at Start; an open session past it is over whether or not the sweep has closed it yet.
@@ -647,12 +631,12 @@ class DebugSession(Base):
     # app/debugsession.py `EndReason`: stopped by the user, or past `until`.
     ended_reason: Mapped[str | None] = mapped_column(String(16))
     issue_url: Mapped[str | None] = mapped_column(String(512))
-    # The issue being written from it (E16 S3; app/debugsession.py `save_draft` holds its shape).
+    # The issue being written from it (app/debugsession.py `save_draft` holds its shape).
     draft: Mapped[dict | None] = mapped_column(JSONB)
 
 
 class DebugScreenshot(Base):
-    """A screenshot attached to a debug session (E16) — pasted, dropped, or captured from the pill.
+    """A screenshot attached to a debug session — pasted, dropped, or captured from the pill.
     Stored in the row: bounded per session (app/debugsession.py), and gone with it by cascade."""
 
     __tablename__ = "debug_screenshots"
@@ -811,44 +795,6 @@ class Notification(Base):
     )
 
 
-class Message(Base):
-    """A message in the notification module (phase 2).
-
-    Two kinds share this table:
-      - `kind="user"`: a private, END-TO-END ENCRYPTED direct message between two users. The
-        server only ever stores `ciphertext` + `iv` (base64); it never holds the plaintext or
-        the keys, so not even an admin can read it. A conversation is just the set of messages
-        between two users — no separate thread entity; queries group by the {sender, recipient}
-        pair.
-      - `kind="system"`: a server-authored plaintext message (e.g. the welcome, or future admin
-        announcements). `sender_id` is NULL and `body` holds the text. These are non-private
-        boilerplate, readable without any E2EE key setup, and surface as a "Siftlode" conversation.
-
-    `read_at` is stamped when the recipient opens the thread (NULL = unread, feeds the nav
-    badge). The recipient FK cascades (and sender, for user messages) so a GDPR account erasure
-    removes that user's messages outright."""
-
-    __tablename__ = "messages"
-
-    id: Mapped[int] = mapped_column(primary_key=True)
-    kind: Mapped[str] = mapped_column(String(16), default="user", server_default="user")
-    # NULL for system messages (no human sender).
-    sender_id: Mapped[int | None] = mapped_column(
-        ForeignKey("users.id", ondelete="CASCADE"), index=True
-    )
-    recipient_id: Mapped[int] = mapped_column(
-        ForeignKey("users.id", ondelete="CASCADE"), index=True
-    )
-    # System messages carry plaintext `body`; user messages carry `ciphertext`+`iv` (base64).
-    body: Mapped[str | None] = mapped_column(Text)
-    ciphertext: Mapped[str | None] = mapped_column(Text)
-    iv: Mapped[str | None] = mapped_column(String(32))
-    read_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
-    created_at: Mapped[datetime] = mapped_column(
-        DateTime(timezone=True), server_default=func.now(), index=True
-    )
-
-
 class MediaAsset(Base, TimestampMixin):
     """A physical downloaded media file — the SHARED cache layer of the download center.
 
@@ -943,8 +889,8 @@ class DownloadJob(Base, TimestampMixin, UpdatedAtMixin):
     """A user's request to download one source — the per-user layer over `MediaAsset`.
 
     The worker claims `queued` jobs with ``FOR UPDATE SKIP LOCKED`` and writes live
-    `progress`/`speed_bps`/`eta_s`/`phase` here (the frontend polls via useLiveQuery — the
-    worker is a separate process and can't reach the in-process WS registry). `profile_snapshot`
+    `progress`/`speed_bps`/`eta_s`/`phase` here (the frontend polls via useLiveQuery: there is
+    no server push channel, and the worker is a separate process anyway). `profile_snapshot`
     freezes the spec at enqueue so later edits/deletion of the profile don't change this job.
     `display_name` is the user's own name for the file (display + the Content-Disposition on
     local download); the on-disk asset is never renamed. `asset_id` links to the shared file
@@ -966,7 +912,8 @@ class DownloadJob(Base, TimestampMixin, UpdatedAtMixin):
     )
     source_kind: Mapped[str] = mapped_column(String(16))  # "youtube" | "url"
     source_ref: Mapped[str] = mapped_column(String(512))
-    # "download" (default) or "edit" (a per-user trim/crop derivative of another job).
+    # "download" (default), "edit" (a per-user trim/crop derivative of another job) or "upload" (a
+    # user's own file, converted by the worker when it is not phone-safe — see `downloads.upload`).
     job_kind: Mapped[str] = mapped_column(String(16), default="download", server_default="download")
     # For edit jobs: the parent download this clip is cut from (SET NULL if the parent is deleted).
     source_job_id: Mapped[int | None] = mapped_column(
@@ -1091,29 +1038,6 @@ class DownloadCredential(Base, TimestampMixin, UpdatedAtMixin):
     cookies_enc: Mapped[str] = mapped_column(Text)  # Fernet ciphertext — never returned/logged
 
 
-class MessageKey(Base, TimestampMixin):
-    """A user's end-to-end-encryption key material for direct messaging (phase 2).
-
-    One row per user. `public_key` (base64 SPKI, ECDH P-256) is shared with others so they can
-    derive the pairwise conversation key. `wrapped_private_key` is the user's private key
-    encrypted IN THE BROWSER with a key derived (PBKDF2) from a message passphrase the server
-    never sees — so the server stores only an opaque blob it cannot unwrap. `salt`/`wrap_iv`
-    parameterise that derivation; `key_check` is a small ciphertext the client decrypts to
-    verify a passphrase on unlock. The server is purely a key directory + blob store; it can
-    never read any message."""
-
-    __tablename__ = "message_keys"
-
-    user_id: Mapped[int] = mapped_column(
-        ForeignKey("users.id", ondelete="CASCADE"), primary_key=True
-    )
-    public_key: Mapped[str] = mapped_column(Text)
-    wrapped_private_key: Mapped[str] = mapped_column(Text)
-    salt: Mapped[str] = mapped_column(String(64))
-    wrap_iv: Mapped[str] = mapped_column(String(32))
-    key_check: Mapped[str] = mapped_column(Text)
-
-
 # --- Plex integration (optional) -------------------------------------------------------------
 # A dedicated, self-contained catalog mirrored from a locally-reachable Plex server. Playback is
 # from the LOCAL physical file (Plex is metadata only); per-user watch state lives in plex_states
@@ -1134,7 +1058,7 @@ class PlexLibrary(Base, TimestampMixin, UpdatedAtMixin):
     kind: Mapped[str] = mapped_column(String(16))  # "movie" | "show"
     enabled: Mapped[bool] = mapped_column(Boolean, default=True, server_default="true")
     synced_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
-    # HU-overlay support (E5 S5). `metadata_lang` tags the section's metadata language. An OVERLAY
+    # HU-overlay support. `metadata_lang` tags the section's metadata language. An OVERLAY
     # library (overlay_of set) is the Hungarian-metadata twin of another section: it is never browsable
     # on its own — its titles' HU fields are written onto the matched EN rows (paired by `guid`).
     metadata_lang: Mapped[str] = mapped_column(String(8), default="en", server_default="en")
@@ -1163,7 +1087,7 @@ class PlexShow(Base, TimestampMixin, UpdatedAtMixin):
         ForeignKey("plex_libraries.id", ondelete="CASCADE"), index=True
     )
     # Canonical external id (e.g. "tmdb:12345") — the key that pairs this EN row with its HU-overlay
-    # twin during sync (E5 S5). Indexed for the guid->row overlay match.
+    # twin during sync. Indexed for the guid->row overlay match.
     guid: Mapped[str | None] = mapped_column(String(64), index=True)
     title: Mapped[str] = mapped_column(String(512))
     summary: Mapped[str | None] = mapped_column(Text)
@@ -1193,7 +1117,7 @@ class PlexShow(Base, TimestampMixin, UpdatedAtMixin):
     )  # cast+directors, folded into search_vector
     # Set once the show's FULL cast has been rich-fetched by the backfill; NULL = still the listing top-3.
     people_enriched_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
-    # Rich-fetch-only detail fields, persisted by the same backfill that fills the full cast (E5 S5b) so
+    # Rich-fetch-only detail fields, persisted by the same backfill that fills the full cast so
     # the info page serves from this row instead of a live plex.metadata() call. NULL until enriched.
     imdb_rating: Mapped[float | None] = mapped_column(
         Float
@@ -1203,7 +1127,7 @@ class PlexShow(Base, TimestampMixin, UpdatedAtMixin):
     full_cast: Mapped[list | None] = mapped_column(
         JSONB(none_as_null=True)
     )  # ordered [{name, role, thumb}]
-    # --- TMDB enrichment (E5 S2): see the identical block on plex_items for the full rationale. Additive;
+    # --- TMDB enrichment: see the identical block on plex_items for the full rationale. Additive;
     #     served from the row; NULL until enriched. (`budget`/`revenue` stay NULL for shows — TV has none.) ---
     tmdb_id: Mapped[int | None] = mapped_column(Integer)
     tmdb_enriched_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
@@ -1221,9 +1145,10 @@ class PlexShow(Base, TimestampMixin, UpdatedAtMixin):
     production_countries: Mapped[list | None] = mapped_column(JSONB(none_as_null=True))  # [str]
     recommendations: Mapped[list | None] = mapped_column(JSONB(none_as_null=True))
     backdrops: Mapped[list | None] = mapped_column(JSONB(none_as_null=True))
-    # Hungarian-metadata twin fields, written by the overlay sync (E5 S5). NULL = no HU twin matched,
+    # Hungarian-metadata twin fields, written by the overlay sync. NULL = no HU twin matched,
     # so serving falls back to the EN columns. `people_text` is language-neutral, reused for both vectors.
     title_hu: Mapped[str | None] = mapped_column(String(512))
+    tagline_hu: Mapped[str | None] = mapped_column(Text)  # the HU overlay's tagline
     summary_hu: Mapped[str | None] = mapped_column(Text)
     genres_hu: Mapped[list | None] = mapped_column(JSONB(none_as_null=True))
     search_vector: Mapped[object | None] = mapped_column(
@@ -1285,7 +1210,7 @@ class PlexItem(Base, TimestampMixin, UpdatedAtMixin):
     )
     kind: Mapped[str] = mapped_column(String(16), index=True)  # "movie" | "episode"
     # Canonical external id (e.g. "tmdb:12345") — pairs a MOVIE row with its HU-overlay twin during
-    # sync (E5 S5); indexed for the guid->row match. Episodes carry it too but it's unused for them.
+    # sync; indexed for the guid->row match. Episodes carry it too but it's unused for them.
     guid: Mapped[str | None] = mapped_column(String(64), index=True)
     show_id: Mapped[int | None] = mapped_column(
         ForeignKey("plex_shows.id", ondelete="CASCADE"), index=True
@@ -1312,7 +1237,7 @@ class PlexItem(Base, TimestampMixin, UpdatedAtMixin):
     # Browser-playability class from ffprobe: "direct" | "remux" | "transcode" (None = unprobed).
     playable: Mapped[str | None] = mapped_column(String(12), index=True)
     probed_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
-    # Transcode HW-capability facet (E7 playback-class filter). NULL for non-transcode items; for a
+    # Transcode HW-capability facet (the playback-class filter). NULL for non-transcode items; for a
     # `playable="transcode"` item: True = the prod iGPU can hardware-transcode it (codec is HW-decodable
     # AND the source is 8-bit SDR), False = software libx264 only (AV1/XviD, or a 10-bit/HDR source that
     # breaks the VAAPI nv12 pipeline — see plex/stream.py). Derived by `plex.sync.transcode_hw_capable`:
@@ -1348,7 +1273,7 @@ class PlexItem(Base, TimestampMixin, UpdatedAtMixin):
     people_text: Mapped[str | None] = mapped_column(Text)
     # Set once the movie's FULL cast has been rich-fetched by the backfill; NULL = still the listing top-3.
     people_enriched_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True))
-    # Rich-fetch-only detail fields, persisted by the same backfill that fills the full cast (E5 S5b) so
+    # Rich-fetch-only detail fields, persisted by the same backfill that fills the full cast so
     # the info page serves from this row instead of a live plex.metadata() call. NULL until enriched.
     imdb_rating: Mapped[float | None] = mapped_column(
         Float
@@ -1358,7 +1283,7 @@ class PlexItem(Base, TimestampMixin, UpdatedAtMixin):
     full_cast: Mapped[list | None] = mapped_column(
         JSONB(none_as_null=True)
     )  # ordered [{name, role, thumb}]
-    # --- TMDB enrichment (E5 S2): matched by the canonical `guid` ("tmdb:12345"), filled by the TMDB
+    # --- TMDB enrichment: matched by the canonical `guid` ("tmdb:12345"), filled by the TMDB
     #     backfill job, SERVED FROM THIS ROW (never a live TMDB call). Additive — imdb_rating/rating
     #     above stay Plex-sourced; `tmdb_rating` is the distinct TMDb community score. NULL until enriched;
     #     `tmdb_enriched_at` is the "tried" watermark (stamped on a definitive no-match too, so an
@@ -1381,10 +1306,11 @@ class PlexItem(Base, TimestampMixin, UpdatedAtMixin):
     recommendations: Mapped[list | None] = mapped_column(JSONB(none_as_null=True))
     backdrops: Mapped[list | None] = mapped_column(
         JSONB(none_as_null=True)
-    )  # TMDB backdrop file_paths [str] (crossfade pool, S8)
-    # Hungarian-metadata twin fields, written by the overlay sync (E5 S5). NULL = no HU twin matched
+    )  # TMDB backdrop file_paths [str] (crossfade pool)
+    # Hungarian-metadata twin fields, written by the overlay sync. NULL = no HU twin matched
     # (serving falls back to the EN columns). `people_text` is language-neutral, reused for both vectors.
     title_hu: Mapped[str | None] = mapped_column(String(512))
+    tagline_hu: Mapped[str | None] = mapped_column(Text)  # the HU overlay's tagline
     summary_hu: Mapped[str | None] = mapped_column(Text)
     genres_hu: Mapped[list | None] = mapped_column(JSONB(none_as_null=True))
     search_vector: Mapped[object | None] = mapped_column(
@@ -1415,21 +1341,26 @@ class PlexPerson(Base, TimestampMixin, UpdatedAtMixin):
     top few cast. Rows whose name no longer appears in ANY item's cast_names/directors (a library
     removal orphaned them) are pruned by the maintenance sweep, mirroring how Plex drops orphaned
     metadata. The filterable cast lives in plex_items/plex_shows.cast_names (self-cleaning with the
-    row); this table is ONLY the headshot cache."""
+    row); this table is the people's own facts: the headshot link, the TMDB link and the TMDB bio
+    (the headshot FILE lives in the image cache, filled by the people_enrich job)."""
 
     __tablename__ = "plex_people"
 
     id: Mapped[int] = mapped_column(primary_key=True)
     name: Mapped[str] = mapped_column(String(255), unique=True, index=True)
     photo_thumb: Mapped[str | None] = mapped_column(Text)  # proxied /person-image URL, or None
-    # TMDB person id, linked by name from a title's TMDB credits (E5 S2 lays the foundation; the /person
-    # biography fetch + person-page render land in S6). NULL until a TMDB-enriched title names this person.
+    # TMDB person id, linked by name from a title's TMDB credits (the /person
+    # biography fetch builds on it). NULL until a TMDB-enriched title names this person.
     tmdb_person_id: Mapped[int | None] = mapped_column(Integer)
-    # --- TMDB person enrichment (E5 S6): fetched from /person/{tmdb_person_id} the first time the person
-    #     page is opened, then SERVED FROM THE ROW. `tmdb_enriched_at` is the watermark (set even on a
+    # --- TMDB person enrichment: fetched from /person/{tmdb_person_id} by the people_enrich job,
+    #     or by the first person-page open the job hasn't reached, then SERVED FROM THE ROW. `tmdb_enriched_at` is the watermark (set even on a
     #     no-op fetch so it isn't retried every open). `known_for` = top combined-credits [{tmdb_id,
     #     media_type, title}], resolved to library rows by guid at serve time (like recommendations). ---
     biography: Mapped[str | None] = mapped_column(Text)
+    # TMDB's Hungarian biography, from the same request's `translations`; a Hungarian reader is
+    # served it, and the instance-language `biography` when TMDB has none. Birthplace, aliases and
+    # keywords have no Hungarian on TMDB, so they stay in the instance language.
+    biography_hu: Mapped[str | None] = mapped_column(Text)
     birthday: Mapped[str | None] = mapped_column(String(10))  # "YYYY-MM-DD", served as-is
     deathday: Mapped[str | None] = mapped_column(String(10))
     place_of_birth: Mapped[str | None] = mapped_column(Text)
@@ -1467,7 +1398,7 @@ class PlexCollection(Base, TimestampMixin, UpdatedAtMixin):
 class PlexState(Base, UpdatedAtMixin):
     """Per-user watch state for a Plex item — mirrors VideoState. Lives in Siftlode's own DB so
     it works for users WITHOUT a Plex account (the whole point of the access-layer design).
-    `synced_to_plex` marks whether this state was pushed back to a linked Plex account (P5)."""
+    `synced_to_plex` marks whether this state was pushed back to a linked Plex account."""
 
     __tablename__ = "plex_states"
     __table_args__ = (UniqueConstraint("user_id", "item_id", name="uq_plex_user_item"),)
@@ -1546,7 +1477,7 @@ class PlexPlaylistItem(Base):
 
 
 class ComfortPin(Base, TimestampMixin):
-    """One entry on a user's comfort shelf — the anti-paralysis fallback: "just put on X" (E24).
+    """One entry on a user's comfort shelf — the anti-paralysis fallback: "just put on X".
 
     Deliberately NOT a boolean on `VideoState`/`PlexState`, and NOT a polymorphic `(kind, ref)` string
     pair. A flag cannot ORDER a shelf and cannot hold a channel or a show — and comfort media is

backend/app/plex/cards.py

@@ -15,20 +15,35 @@ def _art_url(rating_key: str, art_key: str | None) -> str | None:
 
 
 # Card hover-overview: a short synopsis for the reveal panel, trimmed so a strip/related row of cards
-# stays a few hundred bytes each rather than carrying full synopses. (These non-grid card paths serve
-# the EN summary/rating; the main grid coalesces the HU twin in browse.unified_library.)
+# stays a few hundred bytes each rather than carrying full synopses.
 def _card_summary(summary: str | None) -> str | None:
     return (summary or "")[:300] or None
 
 
-def movie_card(it: PlexItem, st: PlexState | None) -> dict:
+def norm_lang(lang: str | None) -> str:
+    """The overlay has an EN base + an optional HU twin; anything other than `hu` serves EN."""
+    return "hu" if lang == "hu" else "en"
+
+
+# Every card speaks the reader's language: `lang="hu"` serves the HU overlay twin's title and
+# summary, falling back to EN where a row has none — the same rule as browse.unified_library's
+# `title_col`/`summary_col`. These rails once served English to a Hungarian reader.
+def display_title(row: PlexItem | PlexShow, lang: str) -> str:
+    return (row.title_hu or row.title) if lang == "hu" else row.title
+
+
+def display_summary(row: PlexItem | PlexShow, lang: str) -> str | None:
+    return (row.summary_hu or row.summary) if lang == "hu" else row.summary
+
+
+def movie_card(it: PlexItem, st: PlexState | None, *, lang: str) -> dict:
     return {
         "id": it.rating_key,
         "type": "movie",
-        "title": it.title,
+        "title": display_title(it, lang),
         "year": it.year,
         "rating": it.rating,
-        "summary": _card_summary(it.summary),
+        "summary": _card_summary(display_summary(it, lang)),
         "duration_seconds": it.duration_s,
         "thumb": f"/api/plex/image/{it.rating_key}",
         "art": _art_url(it.rating_key, it.art_key),
@@ -39,14 +54,14 @@ def movie_card(it: PlexItem, st: PlexState | None) -> dict:
     }
 
 
-def show_card(sh: PlexShow, status: str = "new") -> dict:
+def show_card(sh: PlexShow, status: str = "new", *, lang: str) -> dict:
     return {
         "id": sh.rating_key,
         "type": "show",
-        "title": sh.title,
+        "title": display_title(sh, lang),
         "year": sh.year,
         "rating": sh.rating,
-        "summary": _card_summary(sh.summary),
+        "summary": _card_summary(display_summary(sh, lang)),
         "thumb": f"/api/plex/image/{sh.rating_key}",
         "art": _art_url(sh.rating_key, sh.art_key),
         "season_count": sh.child_count,
@@ -55,12 +70,12 @@ def show_card(sh: PlexShow, status: str = "new") -> dict:
     }
 
 
-def episode_card(e: PlexItem, st: PlexState | None) -> dict:
+def episode_card(e: PlexItem, st: PlexState | None, *, lang: str) -> dict:
     return {
         "id": e.rating_key,
         "type": "episode",
-        "title": e.title,
-        "summary": e.summary,
+        "title": display_title(e, lang),
+        "summary": display_summary(e, lang),
         "season_number": e.season_number,
         "episode_number": e.episode_number,
         "duration_seconds": e.duration_s,
@@ -77,12 +92,15 @@ def leaf_card(
     st: PlexState | None,
     show_title: str | None = None,
     show_rk: str | None = None,
+    *,
+    lang: str,
 ) -> dict:
     """A playable-leaf card (movie or episode). Used by playlists, which can mix both kinds.
-    `show_rk` (the show's rating_key) lets the client group a playlist's episodes by show/season."""
+    `show_rk` (the show's rating_key) lets the client group a playlist's episodes by show/season.
+    `show_title` is the caller's, already in `lang` (see `display_title`)."""
     if it.kind == "episode":
-        card = episode_card(it, st)
+        card = episode_card(it, st, lang=lang)
         card["show_title"] = show_title
         card["show_id"] = show_rk
         return card
-    return movie_card(it, st)
+    return movie_card(it, st, lang=lang)

backend/app/plex/client.py

@@ -2,7 +2,7 @@
 
 Server-side ONLY: uses the admin ``X-Plex-Token`` from sysconfig. Playback never goes through
 this client — Siftlode plays the LOCAL physical file (see app.plex.paths). Plex is used purely for
-metadata, image proxying, and (P5) optional watch-state write-back. Requests JSON via the
+metadata, image proxying, and optional watch-state write-back. Requests JSON via the
 ``Accept: application/json`` header (Plex serves XML by default).
 """
 
@@ -65,7 +65,7 @@ class PlexClient:
         except ValueError as e:
             raise PlexError("Plex returned a non-JSON response") from e
 
-    # --- Connectivity + sections (P0) ---
+    # --- Connectivity + sections ---
     def server_info(self) -> dict:
         """Root identity (friendlyName / version) — used by the admin 'Test connection'."""
         return self._get("/")
@@ -75,25 +75,29 @@ class PlexClient:
         mc = self._get("/library/sections")
         return mc.get("Directory", []) or []
 
-    # --- Catalog mirror (P1) ---
+    # --- Catalog mirror ---
     def section_items(
         self, key: str, item_type: int, start: int = 0, size: int = 200
-    ) -> tuple[list[dict], int]:
+    ) -> tuple[list[dict], int | None]:
         """A page of a section's top-level items. type 1 = movie, 2 = show. Returns
-        (items, total_size)."""
+        (items, total_size); the total is None when the answer carries none, so a container that
+        says "0" (a genuinely empty section) stays distinguishable from one that says nothing."""
         mc = self._get(
             f"/library/sections/{key}/all",
             params={
                 "type": item_type,
                 # includeGuids=1 carries each item's Guid array (tmdb/imdb/tvdb ids) in the cheap
                 # listing, so the sync can capture the canonical guid without a per-item rich fetch —
-                # the pairing key for the HU overlay (E5 S5).
+                # the pairing key for the HU overlay.
                 "includeGuids": 1,
                 "X-Plex-Container-Start": start,
                 "X-Plex-Container-Size": size,
             },
         )
-        return mc.get("Metadata", []) or [], int(mc.get("totalSize", mc.get("size", 0)) or 0)
+        # `totalSize` only: `size` is THIS page's count, and read as the total it made a single
+        # page look like the whole section — "complete" to a prune or a reconcile.
+        total = mc.get("totalSize")
+        return mc.get("Metadata", []) or [], None if total is None else int(total)
 
     def children(self, rating_key: str) -> list[dict]:
         """Direct children (show → seasons, season → episodes)."""
@@ -133,12 +137,17 @@ class PlexClient:
         mc = self._get(f"/library/collections/{rating_key}/children")
         return mc.get("Metadata", []) or []
 
-    # --- Collection editing (P2 — write-back to Plex; all shared/global, admin-gated in the routes) ---
+    # --- Collection editing (write-back to Plex; all shared/global, admin-gated in the routes) ---
     @property
     def machine_id(self) -> str:
-        """Server machineIdentifier — needed in the `uri` of collection add/create calls. Cached."""
+        """Server machineIdentifier — needed in the `uri` of collection add/create calls. Cached.
+        Without one the uri would read `server:///…`, a malformed write Plex answers in its own way;
+        it is refused here, by name."""
         if not getattr(self, "_machine_id", None):
-            self._machine_id = self._get("/identity").get("machineIdentifier") or ""
+            mid = self._get("/identity").get("machineIdentifier")
+            if not mid:
+                raise PlexError("Plex did not report its machine identifier")
+            self._machine_id = mid
         return self._machine_id
 
     def _write(self, method: str, path: str) -> dict:
@@ -197,7 +206,7 @@ class PlexClient:
         items = mc.get("Metadata", []) or []
         return items[0] if items else None
 
-    # --- Watch-state write-back (P5 Phase B — Siftlode→Plex push; acts on the token's account) ---
+    # --- Watch-state write-back (Siftlode→Plex push; acts on the token's account) ---
     _LIBRARY_IDENTIFIER = "com.plexapp.plugins.library"
 
     def scrobble(self, rating_key: str) -> None:
@@ -234,7 +243,7 @@ class PlexClient:
             f"&state={state}&time={int(time_ms)}&duration={int(duration_ms)}",
         )
 
-    # --- Watch-state read for incremental sync (P5 Phase C) ---
+    # --- Watch-state read for incremental sync ---
     def accounts(self) -> list[dict]:
         """Server accounts (the owner + any managed/home users): id, name. Used to resolve the
         owner's accountID so the shared watch-history feed can be filtered to just their views."""

backend/app/plex/detail.py

@@ -1,5 +1,5 @@
 """Parse the "info page" rich metadata out of a full Plex metadata dict, and persist the subset that
-the detail endpoints serve from the DB row (E5 S5b).
+the detail endpoints serve from the DB row.
 
 Shared by the live info-page path (routes.plex.browse) and the background rich re-sync (plex.sync) so
 the two never diverge — the same one-parser rule that keeps `people.py`'s cast handling consistent.

backend/app/plex/genres.py

@@ -0,0 +1,100 @@
+"""One Hungarian word per Plex genre.
+
+Plex's Hungarian agent returns a MIXED vocabulary: some genres come back translated ("Dráma",
+"Vígjáték"), others stay English ("Action", "Adventure", "Biography"), and a title with no HU twin has
+only its English genres. The HU genre list therefore offered "Drama" beside "Dráma" and a dozen English
+words. This map sends every genre word we have met — English (Plex EN agent, TMDB, IMDb) or Plex's
+Hungarian — to the ONE word a Hungarian reader sees. The overlay sync writes it into `genres_hu`, and a
+genre filter is read through it too, so a saved filter still holding "Action" keeps matching "Akció".
+
+A word not in the map passes through unchanged: an unknown genre stays visible (in whatever language
+Plex gave it) rather than disappearing.
+"""
+
+from collections.abc import Iterable
+
+# Keys are matched case-insensitively (see `_BY_KEY`). The Hungarian values double as keys mapping to
+# themselves, so running a list through twice changes nothing.
+_HU: dict[str, str] = {
+    "Action": "Akció",
+    "Action & Adventure": "Akció és kaland",
+    "Adventure": "Kaland",
+    "Animation": "Animációs",
+    "Anime": "Anime",
+    "Biography": "Életrajzi",
+    "Children": "Gyerek",
+    "Comedy": "Vígjáték",
+    "Crime": "Bűnügyi",
+    "Documentary": "Dokumentum",
+    "Drama": "Dráma",
+    "Family": "Családi",
+    "Fantasy": "Fantasy",
+    "Film-Noir": "Film noir",
+    "Food": "Gasztronómia",
+    "Game Show": "Vetélkedő",
+    "Game-Show": "Vetélkedő",
+    "History": "Történelmi",
+    "Home and Garden": "Otthon és kert",
+    "Horror": "Horror",
+    "Indie": "Független",
+    "Kids": "Gyerek",
+    "Martial Arts": "Harcművészet",
+    "Mini-Series": "Minisorozat",
+    "Music": "Zenei",
+    "Musical": "Musical",
+    "Mystery": "Rejtély",
+    "News": "Hírek",
+    "Reality": "Valóságshow",
+    "Reality-TV": "Valóságshow",
+    "Romance": "Romantikus",
+    "Sci-Fi": "Sci-Fi",
+    "Sci-Fi & Fantasy": "Sci-Fi és fantasy",
+    "Science Fiction": "Sci-Fi",
+    "Short": "Rövidfilm",
+    "Soap": "Szappanopera",
+    "Sport": "Sport",
+    "Suspense": "Feszültség",
+    "Talk": "Talkshow",
+    "Talk Show": "Talkshow",
+    "Thriller": "Thriller",
+    "Travel": "Utazás",
+    "TV Movie": "TV film",
+    "War": "Háborús",
+    "War & Politics": "Háború és politika",
+    "Western": "Western",
+    # Plex's own Hungarian words, pinned to themselves (and to the spelling above where they differ).
+    "Akció": "Akció",
+    "Animációs": "Animációs",
+    "Bűnügyi": "Bűnügyi",
+    "Családi": "Családi",
+    "Dokumentum": "Dokumentum",
+    "Dokumentumfilm": "Dokumentum",
+    "Dráma": "Dráma",
+    "Életrajzi": "Életrajzi",
+    "Háborús": "Háborús",
+    "Kaland": "Kaland",
+    "Rejtély": "Rejtély",
+    "Romantikus": "Romantikus",
+    "Történelmi": "Történelmi",
+    "TV film": "TV film",
+    "Vígjáték": "Vígjáték",
+    "Zene": "Zenei",
+    "Zenei": "Zenei",
+}
+_BY_KEY = {k.casefold(): v for k, v in _HU.items()}
+
+
+def hu_genre(word: str) -> str:
+    """The Hungarian word for one genre; an unknown word comes back unchanged."""
+    return _BY_KEY.get(word.strip().casefold(), word)
+
+
+def hu_genres(words: Iterable[str] | None) -> list[str] | None:
+    """A genre list in Hungarian, in its first-seen order, each word once ("Drama" + "Dráma" → one
+    "Dráma"). None or empty stays None, like the listing's own `_tags`."""
+    out: list[str] = []
+    for w in words or ():
+        hu = hu_genre(w)
+        if hu not in out:
+            out.append(hu)
+    return out or None

backend/app/plex/imgcache.py

@@ -8,6 +8,9 @@ few hundred MB at most for a big library — acceptable, matching the download d
 The cache directory (``cache.CACHE_DIR``) is shared with the subtitle cache (see ``subtitles.py``).
 """
 
+import time
+from datetime import timedelta
+
 from fastapi import Response
 from sqlalchemy.orm import Session
 
@@ -20,9 +23,23 @@ log = logs.get_logger(__name__)
 
 _EXT_CT = [("jpg", "image/jpeg"), ("png", "image/png"), ("webp", "image/webp")]
 _CT_EXT = {ct: ext for ext, ct in _EXT_CT}
-# Target box per variant — Plex resizes to these server-side. Posters fill ≤176px cells (grid/info/
-# strips), so 400px covers 2× DPR; the faint backdrop art only needs ~1280px. Keeps images ~tens of KB.
-SIZES = {"thumb": (400, 600), "art": (1280, 720)}
+# Target box per variant — Plex resizes to these server-side. The full poster (400px) covers a detail
+# page's ≤224px poster at 2× DPR, the full art (1280px) a hero backdrop. A card is much smaller: a grid
+# poster is ~144px and a shelf's landscape art ~212px, so the `_sm` pair serves those at 1.33× DPR
+# (cards used to fetch the 400 poster, 42 KB, and the 1280 art for a 212px picture). The browser
+# picks between a pair itself (`srcset`, built by the frontend's `lib/plexImage` cardImage, whose
+# widths plexImage.contract.test holds to these).
+SIZES = {
+    "thumb": (400, 600),
+    "thumb_sm": (240, 360),
+    "art": (1280, 720),
+    "art_sm": (480, 270),
+}
+
+
+def picture(variant: str) -> str:
+    """Which stored picture a size variant is cut from: the poster (`thumb`) or the backdrop (`art`)."""
+    return "art" if variant.startswith("art") else "thumb"
 
 
 def key_for(db: Session, rating_key: str, variant: str) -> str | None:
@@ -56,6 +73,49 @@ def read(key: str) -> tuple[bytes, str] | None:
     return None
 
 
+def has(key: str) -> bool:
+    """A cached, non-empty file exists for `key` (a prefetch's cheap check — no read)."""
+    for ext, _ in _EXT_CT:
+        try:
+            if (CACHE_DIR / f"{key}.{ext}").stat().st_size > 0:
+                return True
+        # silent-ok: a miss, as in read().
+        except OSError:
+            pass
+    return False
+
+
+def drop(key: str) -> None:
+    """Remove `key`'s file, whatever its extension (a superseded variant). Best-effort."""
+    for ext, _ in _EXT_CT:
+        try:
+            (CACHE_DIR / f"{key}.{ext}").unlink(missing_ok=True)
+        # silent-ok: a leftover original only costs disk; the small twin is already written.
+        except OSError:
+            pass
+
+
+def mark_miss(key: str) -> None:
+    """Remember that `key`'s source could not be fetched (an empty `.miss` file; its mtime is when), so
+    a prefetch can leave a dead source alone for a while. Best-effort."""
+    try:
+        CACHE_DIR.mkdir(parents=True, exist_ok=True)
+        (CACHE_DIR / f"{key}.miss").touch()
+    # silent-ok: without the mark the dead source is only asked for again on the next run.
+    except OSError:
+        pass
+
+
+def missed_within(key: str, age: timedelta) -> bool:
+    """`key` was marked as a miss less than `age` ago."""
+    try:
+        mtime = (CACHE_DIR / f"{key}.miss").stat().st_mtime
+    # silent-ok: no mark (the usual case).
+    except OSError:
+        return False
+    return time.time() - mtime < age.total_seconds()
+
+
 def write(key: str, data: bytes, ct: str) -> None:
     ext = _CT_EXT.get((ct or "").split(";")[0].strip(), "jpg")
     try:

backend/app/plex/overlay.py

@@ -0,0 +1,150 @@
+"""The HU overlay: a Hungarian Plex section written onto its English twin's rows, split out of
+`plex/sync.py`, which runs it once per paired section during the full sync.
+
+The pairing is admin-configured (`plex_overlay_map`, ``huKey:enKey,...``). An overlay section creates
+no browsable rows of its own; it only fills the `*_hu` columns of the EN rows it matches by guid.
+`plex/sync.py` and this module import each other as MODULES (attribute access at call time), which is
+what lets either be imported first.
+"""
+
+from collections.abc import Iterable
+
+from sqlalchemy import select
+from sqlalchemy.orm import Session
+
+from app import logs, sysconfig
+from app.models import PlexItem, PlexLibrary, PlexShow
+from app.plex import sync as plex_sync
+from app.plex.client import PlexClient
+from app.plex.genres import hu_genres
+
+log = logs.get_logger(__name__)
+
+
+def overlay_map(db: Session) -> dict[str, str]:
+    """The configured HU-overlay pairing: {huSectionKey: enSectionKey} from `plex_overlay_map`
+    ('huKey:enKey,...'). Each HU section is synced as a non-browsable overlay onto its EN twin.
+    Malformed pairs are skipped."""
+    raw = sysconfig.get_str(db, "plex_overlay_map")
+    out: dict[str, str] = {}
+    for pair in raw.split(","):
+        hu, sep, en = pair.partition(":")
+        hu, en = hu.strip(), en.strip()
+        if sep and hu and en:
+            out[hu] = en
+    return out
+
+
+def sync_overlay(
+    db: Session, plex: PlexClient, lib: PlexLibrary, en_lib_id: int, stats: dict
+) -> None:
+    """Write an overlay section's Hungarian metadata onto its EN twin's rows, matched by guid. Creates
+    NO browsable rows — an overlay library is never listed on its own. Only title/summary/tagline/genres
+    differ by language; the language-neutral people_text feeds both search vectors. A show section's
+    episodes are overlaid too: the HU section lists every episode with its Hungarian title and
+    summary, and the episode rows live in the EN show library.
+
+    This is a full RECONCILE: after a COMPLETE HU listing every guid-bearing EN row's HU fields are
+    cleared, then the matches are repopulated — so an EN row whose HU twin vanished (or was re-guid'd)
+    falls back to EN instead of keeping stale Hungarian metadata. A listing cut short only adds. HU
+    items with no EN match are skipped."""
+    if lib.kind == "movie":
+        passes = [
+            (
+                plex_sync._TYPE_MOVIE,
+                select(PlexItem).where(
+                    PlexItem.library_id == en_lib_id,
+                    PlexItem.kind == "movie",
+                    PlexItem.guid.is_not(None),
+                ),
+            )
+        ]
+    else:
+        passes = [
+            (
+                plex_sync._TYPE_SHOW,
+                select(PlexShow).where(
+                    PlexShow.library_id == en_lib_id, PlexShow.guid.is_not(None)
+                ),
+            ),
+            (
+                plex_sync._TYPE_EPISODE,
+                select(PlexItem).where(
+                    PlexItem.library_id == en_lib_id,
+                    PlexItem.kind == "episode",
+                    PlexItem.guid.is_not(None),
+                ),
+            ),
+        ]
+    for item_type, rows_q in passes:
+        _overlay_pass(db, plex, lib, item_type, db.execute(rows_q).scalars(), stats)
+    db.flush()
+
+
+def _overlay_pass(
+    db: Session,
+    plex: PlexClient,
+    lib: PlexLibrary,
+    item_type: int,
+    rows: Iterable[PlexItem] | Iterable[PlexShow],
+    stats: dict,
+) -> None:
+    """One kind of the overlay (movies, shows or episodes) — see `sync_overlay`."""
+    # guid -> the EN rows carrying it (a list: two editions of one film can share a guid; both overlay).
+    by_guid: dict[str, list[PlexItem | PlexShow]] = {}
+    for r in rows:
+        if r.guid:  # the passes select guid-bearing rows only
+            by_guid.setdefault(r.guid, []).append(r)
+    # The HU listing is read first and applied after, so the clear can wait for its verdict.
+    hu_by_guid, complete = _read_hu_listing(plex, lib, item_type, by_guid)
+    episodes = item_type == plex_sync._TYPE_EPISODE
+    if complete:
+        _clear_hu(by_guid, episodes)
+    else:
+        stats["overlay_incomplete"] = stats.get("overlay_incomplete", 0) + 1
+        log.warning(
+            "the HU listing ended before the server's total; existing HU text is kept",
+            extra=logs.event("plex.sync.overlay_incomplete", library=lib.plex_key),
+        )
+    for g, meta in hu_by_guid.items():
+        targets = by_guid[g]
+        title_hu = (meta.get("title") or "").strip() or None
+        summary_hu = meta.get("summary")
+        tagline_hu = (meta.get("tagline") or "").strip() or None
+        genres_hu = hu_genres(plex_sync._tags(meta, "Genre"))
+        for row in targets:
+            row.title_hu = title_hu
+            row.summary_hu = summary_hu
+            if not episodes:
+                row.tagline_hu = tagline_hu
+                row.genres_hu = genres_hu or row.genres_hu
+        stats["overlaid"] = stats.get("overlaid", 0) + 1
+
+
+def _read_hu_listing(
+    plex: PlexClient, lib: PlexLibrary, item_type: int, by_guid: dict
+) -> tuple[dict[str, dict], bool]:
+    """The HU section's entries for the guids the EN side carries, and whether the walk reached the
+    server's total. The first HU entry for a guid wins; a dup HU item can't overwrite or double-count."""
+    progress: dict = {}
+    hu_by_guid: dict[str, dict] = {}
+    for meta in plex_sync._paginate(plex, lib.plex_key, item_type, progress):
+        g = plex_sync._canonical_guid(meta)
+        if g and g in by_guid and g not in hu_by_guid:
+            hu_by_guid[g] = meta
+    return hu_by_guid, bool(progress.get("complete"))
+
+
+def _clear_hu(by_guid: dict, episodes: bool) -> None:
+    """Reconcile step: clear HU fields on all guid-bearing rows; the matches are re-set after. Only
+    rows that actually held HU data become dirty (SQLAlchemy skips a no-net-change UPDATE), so this is
+    cheap. The genres are the exception: every row gets Hungarian ones, from its HU twin or, lacking
+    one, from its own English list — otherwise the HU genre list mixed "Drama" (no twin) with "Dráma".
+    Only after a COMPLETE listing: one a Plex hiccup cut short would clear the HU text of every title
+    it failed to reach, until the next full sync (the watch reconcile's twin)."""
+    for group in by_guid.values():
+        for row in group:
+            row.title_hu = row.summary_hu = None
+            if not episodes:
+                row.tagline_hu = None
+                row.genres_hu = hu_genres(row.genres)

backend/app/plex/people.py

@@ -3,13 +3,23 @@ or show (cast_names + the people_text search blob + the plex_people headshot cac
 orphaned people. Used by BOTH the info-page lazy enrichment (routes.plex) and the background rich
 re-sync (plex.sync), so the two paths never diverge."""
 
-from urllib.parse import quote
+import hashlib
+import re
+from datetime import timedelta
+from typing import Literal
+from urllib.parse import quote, unquote
 
+import httpx
 from sqlalchemy import select, text
 from sqlalchemy.dialects.postgresql import insert as pg_insert
 from sqlalchemy.orm import Session
 
+from app import logs, sysconfig
 from app.models import PlexPerson
+from app.plex import imgcache
+from app.plex.client import PlexClient, PlexError, PlexNotConfigured
+
+log = logs.get_logger(__name__)
 
 # Cast/crew photos come from Plex's PUBLIC metadata CDN (no token). We proxy them (host-whitelisted)
 # so the user's browser makes no third-party request (privacy + self-host CSP friendliness).
@@ -19,9 +29,7 @@ PERSON_IMG_HOST = "https://metadata-static.plex.tv/"
 def person_thumb_url(thumb: str | None) -> str | None:
     """Proxied /person-image URL for a Plex CDN headshot, or None when it's missing / off-CDN."""
     t = str(thumb or "")
-    return (
-        f"/api/plex/person-image?u={quote(t, safe='')}" if t.startswith(PERSON_IMG_HOST) else None
-    )
+    return f"/api/plex/person-image?u={quote(t, safe='')}" if is_portrait_url(t) else None
 
 
 def cast_from_meta(meta: dict, limit: int = 24) -> list[dict]:
@@ -89,14 +97,23 @@ def enrich_row_people(db: Session, row, cast: list[dict]) -> bool:
     return dirty
 
 
+# Every credit reference: one person name per (movie/show, credit column). The ONE list of credit
+# columns, so the orphan prune and the enrichment order never disagree about who is credited.
+_CREDIT_COLUMNS = [
+    (table, col) for table in ("plex_items", "plex_shows") for col in ("cast_names", "directors")
+]
+
+
+def _credit_refs_sql(union: Literal["UNION", "UNION ALL"]) -> str:
+    return f" {union} ".join(
+        f"SELECT jsonb_array_elements_text({col}) FROM {table} WHERE jsonb_typeof({col}) = 'array'"
+        for table, col in _CREDIT_COLUMNS
+    )
+
+
 # The person names still referenced by ANY movie/show (cast OR directors). plex_people rows outside
 # this set are orphans left by a library removal.
-_REFERENCED_PEOPLE_SQL = """
-    SELECT jsonb_array_elements_text(cast_names) FROM plex_items WHERE jsonb_typeof(cast_names) = 'array'
-    UNION SELECT jsonb_array_elements_text(directors) FROM plex_items WHERE jsonb_typeof(directors) = 'array'
-    UNION SELECT jsonb_array_elements_text(cast_names) FROM plex_shows WHERE jsonb_typeof(cast_names) = 'array'
-    UNION SELECT jsonb_array_elements_text(directors) FROM plex_shows WHERE jsonb_typeof(directors) = 'array'
-"""
+_REFERENCED_PEOPLE_SQL = _credit_refs_sql("UNION")
 
 
 def prune_orphan_people(db: Session) -> int:
@@ -113,3 +130,152 @@ def prune_orphan_people(db: Session) -> int:
     )
     db.commit()
     return res.rowcount or 0
+
+
+# How often each person is credited across movies, shows and episodes (cast OR director) — the order the
+# people_enrich job works in, so the faces met most often are local first.
+_CREDIT_COUNTS_SQL = (
+    f"SELECT ref, count(*) AS n FROM ({_credit_refs_sql('UNION ALL')}) AS refs(ref) GROUP BY ref"
+)
+_NEEDS = {
+    # linked to TMDB, bio never fetched
+    "bio": "p.tmdb_person_id IS NOT NULL AND p.tmdb_enriched_at IS NULL",
+    # has a Plex headshot (whether it is cached is the disk's answer, not the row's)
+    "portrait": "p.photo_thumb IS NOT NULL",
+}
+
+
+def people_by_credits(
+    db: Session, need: Literal["bio", "portrait"], limit: int | None
+) -> list[int]:
+    """plex_people ids that `need` something, the most-credited first (ties by id)."""
+    rows = db.execute(
+        text(
+            f"SELECT p.id FROM plex_people p LEFT JOIN ({_CREDIT_COUNTS_SQL}) c ON c.ref = p.name "
+            f"WHERE {_NEEDS[need]} ORDER BY coalesce(c.n, 0) DESC, p.id"
+            + (" LIMIT :limit" if limit is not None else "")
+        ),
+        {"limit": limit} if limit is not None else {},
+    )
+    return [r[0] for r in rows]
+
+
+# --- Portraits ---------------------------------------------------------------------------
+# A Plex CDN headshot is a full-size original (~170 KB on average, measured). Plex's photo transcoder
+# fits it into this box (~13 KB), which covers the largest place a portrait is drawn (the person page's
+# 6rem avatar) at 2x density and a raised text size. The box is part of the cache key, like a poster's:
+# a new box re-caches instead of serving a stale size. Only a resized image is ever written under it —
+# an original fetched without Plex keeps its own key, and goes once its small twin lands.
+PORTRAIT_BOX = 320
+# A headshot Plex answered "not found" for (a dead CDN file) is not asked for again this long, so a
+# few dead faces never hold up the rest of the prefetch.
+PORTRAIT_RETRY = timedelta(days=7)
+# The one shape a Plex CDN headshot has (…/7/people/<hash>.jpg): plain path segments, no query, no
+# fragment, no dot segment or empty segment (an HTTP client folds `a/../b` and the CDN `a//b` onto one
+# file). The person-image proxy takes nothing else, so it cannot be made to cache endless spellings of
+# one file (the image cache has no eviction). Used where a stored headshot is made, served and warmed.
+_SEGMENT = r"[A-Za-z0-9_-]+(?:\.[A-Za-z0-9_-]+)*"  # a name, dots only between its parts
+_PORTRAIT_URL = re.compile(re.escape(PERSON_IMG_HOST) + _SEGMENT + rf"(?:/{_SEGMENT})*")
+
+
+def is_portrait_url(url: str) -> bool:
+    return _PORTRAIT_URL.fullmatch(url) is not None
+
+
+def _portrait_keys(url: str) -> tuple[str, str]:
+    """(the resized portrait's key, the full-size original's key)."""
+    base = "person_" + hashlib.sha1(url.encode()).hexdigest()
+    return f"{base}_{PORTRAIT_BOX}", base
+
+
+def portrait_cached(url: str) -> tuple[bytes, str] | None:
+    """The cached portrait: the resized one, else an original kept from a time without Plex."""
+    small, original = _portrait_keys(url)
+    return imgcache.read(small) or imgcache.read(original)
+
+
+def cache_portrait(plex: PlexClient | None, url: str) -> tuple[bytes, str]:
+    """Fetch one CDN headshot and cache it. With a Plex client it is resized (the box key; the original
+    is removed); without one the original comes straight from the public CDN and is cached under its own
+    key, so the job still resizes it once Plex is there. Raises PlexError / httpx.HTTPError."""
+    small, original = _portrait_keys(url)
+    if plex is None:
+        with httpx.Client(timeout=10.0, follow_redirects=True) as c:
+            r = c.get(url)
+            r.raise_for_status()
+        data, ct = r.content, r.headers.get("content-type", "image/jpeg")
+        imgcache.write(original, data, ct)
+        return data, ct
+    data, ct = plex.image_bytes(url, PORTRAIT_BOX, PORTRAIT_BOX)
+    imgcache.write(small, data, ct)
+    imgcache.drop(original)
+    return data, ct
+
+
+def cdn_url(photo_thumb: str) -> str | None:
+    """The CDN url inside a stored proxied `/person-image?u=` value (see person_thumb_url)."""
+    _, _, q = photo_thumb.partition("?u=")
+    url = unquote(q)
+    return url if is_portrait_url(url) else None
+
+
+# Plex failing this many headshots in a row ends a run: it is Plex that is failing, not the files.
+_PLEX_DOWN_AFTER = 3
+# The statuses that mean the one file is gone.
+_DEAD = frozenset({404, 410})
+
+
+def warm_portraits(db: Session, limit: int) -> dict:
+    """Cache up to `limit` not-yet-cached headshots, the most-credited people first, so a
+    person's face comes from local disk. Needs Plex (the resize) and the Plex module on; without them
+    nothing is prefetched — full-size originals for every person would cost gigabytes. A headshot Plex
+    answers "not found" for is marked and left for PORTRAIT_RETRY; Plex failing in a row ends the run."""
+    if not sysconfig.get_bool(db, "plex_enabled"):
+        return {"cached": 0, "skipped": "disabled"}
+    try:
+        plex = PlexClient(db)
+    # silent-ok: no Plex server set up is a setting, not a failure; the run's result says "disabled".
+    except PlexNotConfigured:
+        return {"cached": 0, "skipped": "disabled"}
+    photos: dict[int, str] = {
+        pid: thumb
+        for pid, thumb in db.execute(
+            select(PlexPerson.id, PlexPerson.photo_thumb).where(PlexPerson.photo_thumb.is_not(None))
+        ).tuples()
+        if thumb
+    }
+    # `down` counts Plex failing in a row (any answer from Plex breaks it), `plex_failed` in all
+    done = dead = plex_failed = down = 0
+    with plex:
+        for pid in people_by_credits(db, "portrait", None):
+            if done + dead >= limit or down >= _PLEX_DOWN_AFTER:
+                break
+            url = cdn_url(photos.get(pid) or "")
+            if url is None:
+                continue
+            key = _portrait_keys(url)[0]
+            if imgcache.has(key) or imgcache.missed_within(key, PORTRAIT_RETRY):
+                continue
+            try:
+                cache_portrait(plex, url)
+                done, down = done + 1, 0
+            except PlexError as exc:
+                # "Not found" is Plex's answer about this one file (a dead CDN headshot: measured 404,
+                # the CDN itself says 403). Anything else — no answer, a refused token, a 5xx — is
+                # Plex failing, which must neither mark the file nor keep the run going.
+                cause = exc.__cause__
+                if isinstance(cause, httpx.HTTPStatusError) and cause.response.status_code in _DEAD:
+                    dead, down = dead + 1, 0  # Plex answered: it is up, the streak is broken
+                    imgcache.mark_miss(key)
+                else:
+                    plex_failed, down = plex_failed + 1, down + 1
+                log.warning(
+                    "a portrait could not be cached",
+                    extra=logs.event(
+                        "plex.people.portrait_failed",
+                        person_id=pid,
+                        reason=type(exc.__cause__).__name__,
+                    ),
+                )
+    failed = dead + plex_failed
+    return {"cached": done, **({"failed": failed} if failed else {})}

backend/app/plex/query.py

@@ -11,6 +11,7 @@ from sqlalchemy.dialects.postgresql import JSONB
 from sqlalchemy.orm import Session
 
 from app.models import PlexItem, PlexLibrary
+from app.plex.genres import hu_genre
 
 TS_CONFIG = "public.unaccent_simple"
 
@@ -30,9 +31,9 @@ PLEX_SORT_KEYS: tuple[str, ...] = (
     "title",
 )
 
-# --- Language-aware column pickers (E5 S6) ----------------------------------
+# --- Language-aware column pickers ----------------------------------
 #
-# The HU-overlay twin (E5 S5) stores Hungarian metadata in `*_hu` columns on the SAME EN row. When
+# The HU-overlay twin stores Hungarian metadata in `*_hu` columns on the SAME EN row. When
 # the app language is `hu` we serve those, falling back to the EN column where a row has no HU twin.
 # Two flavours of picker, on purpose:
 #   * DISPLAY / FILTER (`title_col`/`genres_col`): `coalesce(*_hu, *)` — an EN fallback so a row with
@@ -40,8 +41,8 @@ PLEX_SORT_KEYS: tuple[str, ...] = (
 #     HU summary directly off the loaded row, so no `summary_col` SQL helper is needed.)
 #   * SEARCH (`vector_col` + `field_recheck`): the raw `*_hu` column coalesced to '' — MIRRORING the
 #     generated `search_vector_hu` definition (which uses `coalesce(title_hu, '')`, NOT an EN
-#     fallback), so the field recheck stays a strict subset of the GIN vector prefilter (the S2
-#     invariant). `people_text` is language-neutral and shared by both vectors.
+#     fallback), so the field recheck stays a strict subset of the GIN vector prefilter (the
+#     field-search invariant). `people_text` is language-neutral and shared by both vectors.
 
 
 def title_col(model, lang: str):
@@ -183,7 +184,13 @@ def meta_filter_conds(model, p: dict, lang: str = "en") -> list:
     gsel = split_csv(p.get("genres"))
     if gsel:
         gcol = genres_col(model, lang)
-        gc = [gcol.contains([g]) for g in gsel]
+        if lang == "hu":
+            # The HU column holds one Hungarian word per genre (plex/genres.py), so a filter saved while
+            # it offered "Action" is read as "Akció" too. The word as picked still matches as well: a row
+            # the sync has not reached since keeps its English (or Plex's mixed) list until it does.
+            gc = [or_(*(gcol.contains([w]) for w in dict.fromkeys((hu_genre(g), g)))) for g in gsel]
+        else:
+            gc = [gcol.contains([g]) for g in gsel]
         conds.append(and_(*gc) if p.get("genre_mode") == "all" else or_(*gc))
     crs = split_csv(p.get("content_ratings"))
     if crs:
@@ -209,7 +216,7 @@ def meta_filter_conds(model, p: dict, lang: str = "en") -> list:
     stds = split_csv(p.get("studios"))
     if stds:  # OR
         conds.append(model.studio.in_(stds))
-    for kw in split_csv(p.get("keywords")):  # AND — TMDB keyword chips from the detail page (E5 S5)
+    for kw in split_csv(p.get("keywords")):  # AND — TMDB keyword chips from the detail page
         conds.append(model.keywords.contains([kw]))
     csel = split_csv(p.get("collections"))
     if (
@@ -225,7 +232,7 @@ def apply_meta_filters(q, model, p: dict, lang: str = "en"):
     return q.where(*conds) if conds else q
 
 
-# --- Playback-class filter (E7) — MOVIE-ONLY (a show is an aggregate with no single class) ---------
+# --- Playback-class filter — MOVIE-ONLY (a show is an aggregate with no single class) ---------
 # The 4 classes the feed filter + facet expose, derived from the stored `playable` + `transcode_hw`.
 # ONE definition (this SQL CASE) so the filter WHERE and the facet counts can't drift: `playable_conds`
 # filters on it, `facets` groups by it. transcode_hw IS NOT True groups NULL (unprobed) with software.
@@ -271,7 +278,7 @@ class LibraryFilters:
     actors: str | None = None
     actor_mode: str = "any"  # how multiple actors combine: any (OR, default) | all (AND)
     studios: str | None = None
-    keywords: str | None = None  # TMDB keyword chips (E5 S5), CSV; language-neutral
+    keywords: str | None = None  # TMDB keyword chips, CSV; language-neutral
     collections: str | None = None  # CSV of collection rating_keys
     collection_mode: str = "any"  # how multiple collections combine: any (OR, default) | all (AND)
     playable: str | None = None  # CSV of playback classes: direct|remux|transcode_hw|transcode_sw
@@ -304,7 +311,7 @@ class LibraryFilters:
 # watch state. The SAME predicate has to be applied in two places — `unified_library` (the grid) and
 # `facets.wsq` (so the facet counts reflect exactly the visible set) — and they MUST stay identical:
 # a hidden movie kept out of the grid must not inflate the facets either. These helpers are the one
-# home for that predicate so the two call sites can't drift (E3 S1; pinned by test_plex_facets_grid_
+# home for that predicate so the two call sites can't drift (pinned by test_plex_facets_grid_
 # parity). Callers pass the already-joined state expressions, so this module stays model-agnostic:
 #   * movies — `st` is an aliased PlexState outer-joined on (item_id, user_id).
 #   * shows  — `tot`/`wat`/`inp` are the coalesced totals from `rollup.agg_subquery`.
@@ -360,9 +367,9 @@ def apply_show_state_filter(query, tot, wat, inp, show: str):
 def content_libs(db: Session) -> tuple[list[int], list[int]]:
     """The BROWSABLE content libraries, partitioned into (movie_lib_ids, show_lib_ids) — the single
     source of the browse scope. Excludes HU overlay libraries (`overlay_of` set): those are never
-    browsed on their own, they only lend their Hungarian metadata to the EN twin's rows (E5 S5).
+    browsed on their own, they only lend their Hungarian metadata to the EN twin's rows.
 
-    Lives here, in the query layer, rather than in `routes/plex/_shared.py` where it started: E24's
+    Lives here, in the query layer, rather than in `routes/plex/_shared.py` where it started: the
     Home surface needs the same scope, and a domain package importing from `routes/` would invert the
     one-way dependency CLAUDE.md draws. It never touched request state, so nothing was lost by the
     move."""

backend/app/plex/rollup.py

@@ -13,10 +13,10 @@ from app.models import PlexItem, PlexState
 # Sort sentinel for a rail row whose timestamp is missing, so `key=lambda r: r[0] or NO_TS` reads the
 # same at every rail that merges two type-homogeneous lists in Python.
 #
-# It is NOT a crash fix, whatever the E24 commit that introduced it says. The `(ts is not None, ts)`
+# It is NOT a crash fix, whatever the commit that introduced it says. The `(ts is not None, ts)`
 # key it replaced cannot raise: tuple comparison finds the first UNEQUAL element via `==`, so two
 # `(False, None)` keys compare equal and `None < None` is never reached — verified by execution
-# during the E24 epic-end review, over every permutation of null/non-null rows. What the sentinel
+# during a review, over every permutation of null/non-null rows. What the sentinel
 # buys is one idiom instead of two, and a key that says "missing sorts last" outright rather than
 # encoding it in a boolean flag. A BARE `r[0]` would genuinely raise, which is the mutation the
 # rails' tests do guard against.

backend/app/plex/stream.py

@@ -15,7 +15,7 @@ HLS muxer does the segmentation (the only reliable way to cut a stream-copy at k
 
 Sessions live in this (single) API process; a periodic reaper kills idle ones and frees the temp
 segments. The remux is CPU-light (video copy + a tiny audio transcode), so it runs fine on the
-CPU-only prod host; only P3's full transcode is CPU-heavy.
+CPU-only prod host; only the full transcode is CPU-heavy.
 """
 
 import os
@@ -120,8 +120,8 @@ class HlsSession:
         # Unguessable per-session id, carried in the URLs the client is handed. A seek-restart
         # replaces the viewer's registered session for the item — and the old player's next request
         # for `seg_12.ts`, resolved by (viewer, rating_key) alone, was answered 200 with the NEW session's twelfth segment, produced from a different offset
-        # under `-copyts`. Wrong bytes, not an error, so the client's reclaim never fired on it
-        # (deep-review round 5). A request naming a session that is no longer registered is a 404,
+        # under `-copyts`. Wrong bytes, not an error, so the client's reclaim never fired on it.
+        # A request naming a session that is no longer registered is a 404,
         # which the player recovers from by restarting at the playhead.
         self.id = secrets.token_urlsafe(9)
         self.key = key  # the item's rating_key (what the logs name)
@@ -245,6 +245,9 @@ def _probe_video_kind(src: Path) -> str:
             capture_output=True,
             text=True,
             timeout=10,
+            # A non-zero exit is a failed probe (CalledProcessError, a SubprocessError → the branch
+            # below), not an answer: whatever it printed before refusing is not the file's truth.
+            check=True,
         )
     except (subprocess.SubprocessError, OSError) as exc:
         # The safe kind is a real answer, but it is the answer for an 8-bit SDR source — a blind
@@ -410,7 +413,7 @@ def _venc_opts(
 def _render_node_present() -> bool:
     """Whether the VAAPI render node is available — i.e. `/dev/dri` was passed into the container.
     Present on prod after the iGPU enablement; absent on the CPU-only dev box and in the localdev
-    Docker. Its own function so the selection is monkeypatchable in tests and one place for S1c's knob."""
+    Docker. Its own function so the selection is monkeypatchable in tests and the one place to change it."""
     return Path(_VAAPI_DEVICE).exists()
 
 
@@ -502,6 +505,9 @@ def _probe_audio_langs(src: Path) -> list[str]:
             capture_output=True,
             text=True,
             timeout=10,
+            # A non-zero exit is a failed probe (CalledProcessError, a SubprocessError → the branch
+            # below), not an answer: whatever it printed before refusing is not the file's truth.
+            check=True,
         )
         return [ln.strip() for ln in out.stdout.splitlines()]
     except (subprocess.SubprocessError, OSError) as exc:
@@ -570,7 +576,7 @@ def _ffmpeg_cmd(
                 entry += ",default:yes"
             var.append(entry)
         # The name patterns come from the vocabulary functions, with ffmpeg's own placeholders passed
-        # through them — so the producer and the parser cannot spell the layout differently (round 7).
+        # through them — so the producer and the parser cannot spell the layout differently.
         args += _HLS_TAIL + [
             "-master_pl_name",
             _MASTER_PL,
@@ -615,6 +621,9 @@ def _probe_first_pts(seg: Path) -> float | None:
             capture_output=True,
             text=True,
             timeout=10,
+            # A non-zero exit is a failed probe (CalledProcessError, a SubprocessError → the branch
+            # below), not an answer: whatever it printed before refusing is not the file's truth.
+            check=True,
         )
         for line in out.stdout.splitlines():
             tok = line.strip().strip(",")
@@ -736,8 +745,7 @@ def start_session(
         # check-and-replace happens under ONE _lock, so a newer same-key session (a concurrent
         # seek-restart) installed between the fallback's decision and here is never clobbered. When the
         # registered session is no longer `replace_only`, we abort and return None (the caller yields to
-        # whoever is registered now). The INITIAL spawn passes replace_only=None and always supersedes —
-        # that is the seek-restart's intended last-writer-wins.
+        # whoever is registered now). The INITIAL spawn never gets None — see its call site.
         # Software-fallback encode tuning — read only on the SW transcode path (the VAAPI path and
         # remux ignore preset/crf/max_height, and remux must not touch the DB for them).
         preset, crf, max_height = "veryfast", 20, 0
@@ -814,6 +822,7 @@ def start_session(
         return s
 
     s = _spawn(*_select_encode(db, item, transcode, src))
+    assert s is not None  # replace_only=None always supersedes: a seek-restart's last-writer-wins
     # Wait for the first video segment outside the lock (this blocks on ffmpeg, and we must not hold up
     # other sessions). Pass the proc so a HW session that dies at VAAPI init fails fast (not after 25s).
     ready = wait_for(_seg(0), timeout=25.0, proc=s.proc)
@@ -878,7 +887,7 @@ def current_session(user_id: int, key: str) -> HlsSession | None:
 # The names ffmpeg writes into a session directory, in ONE place — and "one place" means the argv
 # BUILDS its patterns from these same functions (`segment_name("%d", "%v")`), not that a comment
 # says so. Getting this wrong is invisible in both directions: a name we failed to anticipate used
-# to fall back to "wait and see" (round 6's HIGH), and now that the rule is positive it falls back
+# to fall back to "wait and see", and now that the rule is positive it falls back
 # to a 404 instead — so the producer and the parser have to be literally the same spelling.
 #
 # MEASURED (ffmpeg 7.1.5 in the API container — the binary that actually writes these files — and
@@ -892,7 +901,7 @@ def current_session(user_id: int, key: str) -> HlsSession | None:
 # SESSION and not to HLS in general: a single-audio session can never write `master.m3u8`,
 # `stream_*.m3u8` or any `seg_<variant>_<n>.ts`, and a multi-audio one can never write `index.m3u8`
 # or `seg_<n>.ts`. Accepting the union meant a hopeless name of the other shape was waited out for
-# the full playlist/segment timeout — the threadpool parking this guard exists to end (round 7).
+# the full playlist/segment timeout — the threadpool parking this guard exists to end.
 _MASTER_PL = "master.m3u8"
 _INDEX_PL = "index.m3u8"
 _SEG_NAME_RE = re.compile(r"^seg_(?:([0-9a-z]{1,4})_)?(\d{1,9})\.ts$")
@@ -944,12 +953,12 @@ def artifact_may_arrive(s: "HlsSession", filename: str) -> bool:
     name belonging to the other shape is as hopeless here as a random string. Reading those as "an
     HLS name, so wait and see" left the whole class reachable — measured through the real route
     against a live single-audio session, `master.m3u8` and `stream_q7z.m3u8` were 404s after 30.0 s
-    and `seg_a0_0.ts` / `seg_zzzz_0.ts` after 20.1 s each (deep-review round 7).
+    and `seg_a0_0.ts` / `seg_zzzz_0.ts` after 20.1 s each.
 
     Saying no at once matters because the route that waits is a SYNC one: it holds a thread off the
     ~40-slot pool, and a pooled DB connection with it, for the whole timeout. An ordinary seek past
     the produced region was enough to park one for 20 s, and forty concurrent requests for a name
-    that will never exist parked the whole API (deep-review round 5). Round 5's guard read only
+    that will never exist parked the whole API. An earlier guard read only
     single-audio segment names, so both halves of that were still fully reachable: every multi-audio
     name fell through, and so did any `.m3u8`/`.vtt` name at all."""
     if filename in playlist_names(s.variants):
@@ -1084,15 +1093,15 @@ def reap_idle(finished_idle_s: float = _FINISHED_IDLE_S) -> int:
     A finished session is also what a PAUSED viewer has: a remux runs far faster than realtime, so
     ffmpeg is done a minute or two in and writes ENDLIST; hls.js then stops asking for the playlist,
     and with its forward buffer full a paused player fetches nothing at all. With the 30 s rule that
-    killed a one-minute pause, and playback died ~30 s after resuming (deep-review round 4).
+    killed a one-minute pause, and playback died ~30 s after resuming.
 
     What makes the short rule safe again is that fetching is no longer the only signal: a player
     parked MID-ITEM checkpoints its position every 10 s, and that route calls `touch_session`, so a
-    viewer who is merely paused says so far more often than this rule asks (round 5). The heartbeat
+    viewer who is merely paused says so far more often than this rule asks. The heartbeat
     stops once the item is finished or before a position is known (`usePlexProgress.flush`) — both
     states where reaping is correct, and where the player's 404 → reclaim recovers a viewer who
-    scrubs back in anyway. Do not restate this as "every 10 s whether or not it is playing": round 6
-    found that claim in two comments and not in the code. Round 4's blanket
+    scrubs back in anyway. Do not restate this as "every 10 s whether or not it is playing":
+    a review found that claim in two comments and not in the code. An earlier blanket
     `reap_abandoned` bought the same safety by keeping every finished session for ten minutes — each
     one a whole segmented copy of the film on a volume shared with the download center."""
     now = time.time()

backend/app/plex/subtitles.py

@@ -162,7 +162,8 @@ def ffmpeg_to_vtt(src: str, sub_ord: int | None) -> str:
         "webvtt",
         "pipe:1",
     ]
-    r = subprocess.run(cmd, capture_output=True, timeout=EXTRACT_TIMEOUT_S)
+    # check=False: the exit status is read on the next line, with ffmpeg's own stderr as the reason.
+    r = subprocess.run(cmd, capture_output=True, timeout=EXTRACT_TIMEOUT_S, check=False)
     if r.returncode != 0 or not r.stdout:
         raise PlexError(
             (r.stderr or b"").decode("utf-8", "replace")[:200] or "ffmpeg subtitle convert failed"

backend/app/plex/sync.py

@@ -12,7 +12,7 @@ on a NON-empty listing — a section that returns zero items is left untouched,
 glitch that yields an empty page can't wipe a library (the metadata is cheap to re-sync; an accidental
 mass-delete is the only real risk). Library-level reconcile (dropping a whole removed section, the
 dead `enabled` column) is a separate follow-up. Cast + intro/credit markers are fetched lazily on the
-item-detail/player path (P2), not here.
+item-detail/player path, not here.
 """
 
 from datetime import datetime, timezone
@@ -22,8 +22,9 @@ from sqlalchemy.orm import Session
 
 from app import logs, sysconfig
 from app.models import PlexCollection, PlexItem, PlexLibrary, PlexSeason, PlexShow
-from app.plex import detail, people
+from app.plex import detail, overlay, people
 from app.plex.client import PlexClient, PlexError, PlexNotConfigured
+from app.plex.genres import hu_genres
 from app.plex.lookups import by_rating_key
 
 log = logs.get_logger(__name__)
@@ -149,7 +150,7 @@ _GUID_PREFERENCE = ("tmdb", "imdb", "tvdb")
 
 
 def _canonical_guid(meta: dict) -> str | None:
-    """The canonical external id used to pair an EN title with its HU-overlay twin (E5 S5). Reads the
+    """The canonical external id used to pair an EN title with its HU-overlay twin. Reads the
     `Guid` array (present only with includeGuids=1 in the listing), preferring tmdb > imdb > tvdb,
     normalised to '<provider>:<id>' (e.g. 'tmdb:12345'). None when no known provider id is present."""
     found: dict[str, str] = {}
@@ -192,20 +193,6 @@ def _enabled_section_keys(db: Session) -> set[str] | None:
     return keys or None
 
 
-def _overlay_map(db: Session) -> dict[str, str]:
-    """The configured HU-overlay pairing: {huSectionKey: enSectionKey} from `plex_overlay_map`
-    ('huKey:enKey,...'). Each HU section is synced as a non-browsable overlay onto its EN twin.
-    Malformed pairs are skipped."""
-    raw = sysconfig.get_str(db, "plex_overlay_map")
-    out: dict[str, str] = {}
-    for pair in raw.split(","):
-        hu, sep, en = pair.partition(":")
-        hu, en = hu.strip(), en.strip()
-        if sep and hu and en:
-            out[hu] = en
-    return out
-
-
 def sync(db: Session) -> dict:
     """Full reconcile of the enabled movie/show sections. Returns a small stats dict."""
     if not sysconfig.get_bool(db, "plex_enabled"):
@@ -221,7 +208,7 @@ def sync(db: Session) -> dict:
         "pruned": 0,
     }
     wanted = _enabled_section_keys(db)
-    overlay_map = _overlay_map(db)  # {huKey: enKey}
+    overlay_map = overlay.overlay_map(db)  # {huKey: enKey}
     overlay_keys = set(overlay_map)
     try:
         with PlexClient(db) as plex:
@@ -265,13 +252,13 @@ def sync(db: Session) -> dict:
                 lib.metadata_lang = "hu"
                 lib.overlay_of = en_lib.id
                 db.flush()
-                _sync_overlay(db, plex, lib, en_lib.id, stats)
+                overlay.sync_overlay(db, plex, lib, en_lib.id, stats)
                 db.commit()
             # Rich-fetch a bounded batch of not-yet-enriched titles' FULL cast, so the whole library
             # becomes filterable over successive runs (the cheap listing above only stored the top ~3).
             try:
                 stats["people_enriched"] = _enrich_people_batch(db, plex)
-            except Exception:  # noqa: BLE001 — best-effort; never fail the sync over a backfill
+            except Exception:  # best-effort; never fail the sync over a backfill
                 # With the traceback: a blind catch over a batch that fetches AND writes can be
                 # a Plex shape change, a DB error or our own bug, and `%s` told them apart badly.
                 log.exception("the Plex people-enrich batch failed")
@@ -280,7 +267,7 @@ def sync(db: Session) -> dict:
             # successive runs. Only HW-decodable-codec transcodes are probed — the rest are SW by codec.
             try:
                 stats["media_probed"] = _probe_media_batch(db, plex)
-            except Exception:  # noqa: BLE001 — best-effort; never fail the sync over a backfill
+            except Exception:  # best-effort; never fail the sync over a backfill
                 log.exception("the Plex media-probe batch failed")
     # silent-ok: the return value IS the report. `{"skipped": …}` goes to the scheduler, which
     # logs the job result, and to the admin route, which answers it to the caller.
@@ -294,7 +281,7 @@ def sync(db: Session) -> dict:
     # appear in ANY item's cast/directors — drop those rows, mirroring how Plex drops orphaned metadata.
     try:
         stats["people_pruned"] = people.prune_orphan_people(db)
-    except Exception:  # noqa: BLE001 — best-effort housekeeping; never fail the sync over it
+    except Exception:  # best-effort housekeeping; never fail the sync over it
         log.exception("the Plex people prune failed")
     log.info("Plex sync done: %s", stats)
     return stats
@@ -431,7 +418,10 @@ def _paginate(plex: PlexClient, key: str, item_type: int, progress: dict | None
     if progress is not None:
         # A truthy `total` we reached = a trustworthy complete listing. `total` unknown/zero-with-rows
         # (a Plex that doesn't report totalSize) is treated as NOT complete, so prune stays conservative.
+        # `total` itself is kept for a caller to whom a server that SAID 0 is news it can trust (the
+        # watch reconcile); an empty mirror or overlay must not read it so.
         progress["complete"] = bool(total) and start >= total
+        progress["total"] = total
 
 
 # The section listing exposes at most the top ~20 cast (Plex's own cap); a stored cast longer than this
@@ -448,12 +438,16 @@ def _apply_facet_fields(row, meta: dict) -> None:
     row.studio = (meta.get("studio") or None) and str(meta.get("studio"))[:255]
     row.originally_available_at = _date(meta.get("originallyAvailableAt"))
     row.genres = _tags(meta, "Genre")
+    # Every row has Hungarian genres, not only one with an HU overlay twin (the overlay pass, which runs
+    # after this, puts the twin's own list over it): the HU genre facets and filter read this column,
+    # and an English word there could never match a filter read in Hungarian.
+    row.genres_hu = hu_genres(row.genres)
     row.directors = _tags(meta, "Director")
     # The listing carries only the top ~20 cast, so re-applying it would DOWNGRADE an already-enriched
     # full cast (up to 24, grown by the rich backfill/info-page enrich), silently breaking the
     # person-page cast filter for names past the top ~20. The watermark gate below (`people_enriched_at
     # is None`) normally protects this, but it is NOT enough on its own: a migration that resets the
-    # watermark re-opens the regression (0065 nulled it, which is exactly how the S5b fix got undone in
+    # watermark re-opens the regression (0065 nulled it, which is exactly how a fix got undone in
     # prod). So guard STRUCTURALLY too, by LENGTH against the listing cap: a stored cast LONGER than the
     # cap can only be an enriched fuller list, so keep it (robust to any watermark reset). A stored cast
     # at or below the cap is itself listing-sourced, so take the fresh listing — which tracks Plex cast
@@ -655,59 +649,6 @@ def _sync_shows(db: Session, plex: PlexClient, lib: PlexLibrary, stats: dict) ->
     _prune_stale(db, shows, shows_seen, stats, complete=shows_progress.get("complete", False))
 
 
-def _sync_overlay(
-    db: Session, plex: PlexClient, lib: PlexLibrary, en_lib_id: int, stats: dict
-) -> None:
-    """Write an overlay section's Hungarian metadata onto its EN twin's rows, matched by guid. Creates
-    NO browsable rows — an overlay library is never listed on its own. Only title/summary/genres differ
-    by language; the language-neutral people_text feeds both search vectors.
-
-    This is a full RECONCILE: every guid-bearing EN row's HU fields are cleared first, then the HU
-    section repopulates the matches — so an EN row whose HU twin vanished (or was re-guid'd) falls back
-    to EN instead of keeping stale Hungarian metadata. HU items with no EN match are skipped."""
-    if lib.kind == "movie":
-        rows = db.execute(
-            select(PlexItem).where(
-                PlexItem.library_id == en_lib_id,
-                PlexItem.kind == "movie",
-                PlexItem.guid.is_not(None),
-            )
-        ).scalars()
-        item_type = _TYPE_MOVIE
-    else:
-        rows = db.execute(
-            select(PlexShow).where(PlexShow.library_id == en_lib_id, PlexShow.guid.is_not(None))
-        ).scalars()
-        item_type = _TYPE_SHOW
-    # guid -> the EN rows carrying it (a list: two editions of one film can share a guid; both overlay).
-    by_guid: dict[str, list] = {}
-    for r in rows:
-        by_guid.setdefault(r.guid, []).append(r)
-    # Reconcile step: clear HU fields on all guid-bearing rows. Only rows that actually held HU data
-    # become dirty (SQLAlchemy skips a no-net-change UPDATE), so this is cheap; matches are re-set below.
-    for group in by_guid.values():
-        for row in group:
-            row.title_hu = row.summary_hu = row.genres_hu = None
-    seen: set[str] = (
-        set()
-    )  # first HU entry for a guid wins; a dup HU item can't overwrite or double-count
-    for meta in _paginate(plex, lib.plex_key, item_type):
-        g = _canonical_guid(meta)
-        targets = by_guid.get(g) if g else None
-        if not targets or g in seen:
-            continue
-        seen.add(g)
-        title_hu = (meta.get("title") or "").strip() or None
-        summary_hu = meta.get("summary")
-        genres_hu = _tags(meta, "Genre")
-        for row in targets:
-            row.title_hu = title_hu
-            row.summary_hu = summary_hu
-            row.genres_hu = genres_hu
-        stats["overlaid"] = stats.get("overlaid", 0) + 1
-    db.flush()
-
-
 def _upsert_collection(
     db: Session, existing: PlexCollection | None, rk: str, lib_id: int, meta: dict
 ) -> PlexCollection:
@@ -733,9 +674,11 @@ def resync_collection(db: Session, plex: PlexClient, lib: PlexLibrary, rk: str)
     meta = plex.metadata(rk) or {}
     col = _upsert_collection(db, by_rating_key(db, PlexCollection, rk), rk, lib.id, meta)
     Model = PlexItem if lib.kind == "movie" else PlexShow
-    new_members = {
-        str(c.get("ratingKey")) for c in plex.collection_children(rk) if c.get("ratingKey")
-    }
+    children = _believed_children(plex, rk, meta)
+    if children is None:
+        db.flush()
+        return col
+    new_members = {str(c.get("ratingKey")) for c in children if c.get("ratingKey")}
     current = db.execute(select(Model).where(Model.collection_keys.contains([rk]))).scalars().all()
     for row in current:  # drop rk from rows that are no longer members
         if row.rating_key not in new_members:
@@ -748,6 +691,22 @@ def resync_collection(db: Session, plex: PlexClient, lib: PlexLibrary, rk: str)
     return col
 
 
+def _believed_children(plex: PlexClient, rk: str, meta: dict) -> list[dict] | None:
+    """A collection's members, or None when the answer is not believed: no members from a collection
+    whose own listing counts some (`childCount`) is a glitch, not a collection emptied — the same
+    empty-answer rule as the collection listing's. Rebuilding on it showed the
+    collection empty until the next sync."""
+    children = plex.collection_children(rk)
+    count = str(meta.get("childCount") or "0")
+    if children or not count.isdigit() or int(count) == 0:
+        return children
+    log.warning(
+        "Plex listed no members for a non-empty collection; keeping its membership",
+        extra=logs.event("plex.sync.collection_children_empty", collection=rk),
+    )
+    return None
+
+
 def delete_collection_local(db: Session, lib: PlexLibrary, rk: str) -> None:
     """Local cleanup after a collection is deleted from Plex: strip rk from members + drop the row."""
     Model = PlexItem if lib.kind == "movie" else PlexShow
@@ -759,6 +718,19 @@ def delete_collection_local(db: Session, lib: PlexLibrary, rk: str) -> None:
     db.flush()
 
 
+def _rebuild_membership(
+    db: Session, lib: PlexLibrary, membership: dict[str, list[str]], kept: set[str]
+) -> None:
+    """Rebuild collection_keys on every member row of this library (clear + set), carrying over the
+    keys of a collection whose empty answer was not believed (`_believed_children`)."""
+    for Model in (PlexItem, PlexShow):
+        for row in db.execute(select(Model).where(Model.library_id == lib.id)).scalars():
+            held = [k for k in (row.collection_keys or []) if k in kept]
+            keys = [*membership.get(row.rating_key, []), *held]
+            row.collection_keys = list(dict.fromkeys(keys)) or None
+    db.flush()
+
+
 def _sync_collections(db: Session, plex: PlexClient, lib: PlexLibrary, stats: dict) -> None:
     """Mirror a library's collections + membership. Reads never touch Plex afterwards. Membership is
     written as `collection_keys` on member movies (plex_items) / shows (plex_shows). A full reconcile:
@@ -770,7 +742,17 @@ def _sync_collections(db: Session, plex: PlexClient, lib: PlexLibrary, stats: di
             select(PlexCollection).where(PlexCollection.library_id == lib.id)
         ).scalars()
     }
+    if not cols and existing:
+        # The catalog's own rule (see the module doc): an empty listing while we hold rows is a Plex
+        # glitch, not a library emptied of collections. Reconciling on it dropped every collection
+        # (its user-set `editable` flag with it) and every member's keys.
+        log.warning(
+            "Plex listed no collections for a library that has some; keeping them this run",
+            extra=logs.event("plex.sync.collections_empty", library=lib.plex_key),
+        )
+        return
     seen: set[str] = set()
+    kept: set[str] = set()  # collections whose empty children answer is not believed
     membership: dict[str, list[str]] = {}  # member rating_key -> [collection rating_key, ...]
     for meta in cols:
         rk = str(meta.get("ratingKey") or "")
@@ -778,7 +760,10 @@ def _sync_collections(db: Session, plex: PlexClient, lib: PlexLibrary, stats: di
             continue
         _upsert_collection(db, existing.get(rk), rk, lib.id, meta)
         seen.add(rk)
-        for child in plex.collection_children(rk):
+        children = _believed_children(plex, rk, meta)
+        if children is None:
+            kept.add(rk)
+        for child in children or []:
             crk = str(child.get("ratingKey") or "")
             if crk:
                 membership.setdefault(crk, []).append(rk)
@@ -787,9 +772,4 @@ def _sync_collections(db: Session, plex: PlexClient, lib: PlexLibrary, stats: di
     for rk, col in existing.items():
         if rk not in seen:
             db.delete(col)
-    # Rebuild collection_keys on every member row of this library (clear + set).
-    for row in db.execute(select(PlexItem).where(PlexItem.library_id == lib.id)).scalars():
-        row.collection_keys = membership.get(row.rating_key) or None
-    for row in db.execute(select(PlexShow).where(PlexShow.library_id == lib.id)).scalars():
-        row.collection_keys = membership.get(row.rating_key) or None
-    db.flush()
+    _rebuild_membership(db, lib, membership, kept)

backend/app/plex/tmdb.py

@@ -1,4 +1,4 @@
-"""TMDB enrichment for the mirrored Plex catalog (E5 S2).
+"""TMDB enrichment for the mirrored Plex catalog.
 
 A thin synchronous TMDB v3 client plus the parse/persist of the enrichment fields onto plex_items /
 plex_shows, matched by the stored canonical `guid` ("tmdb:12345", or imdb/tvdb resolved via /find).
@@ -10,8 +10,10 @@ One-parser rule (mirrors app/plex/detail.py): `parse_tmdb` is the pure parser, `
 the row. The enrich batch + scheduler entry live here too so the whole TMDB integration is one module.
 """
 
+import re
 import time
 from datetime import datetime, timezone
+from typing import cast
 
 import httpx
 from sqlalchemy import func, select
@@ -21,6 +23,8 @@ from sqlalchemy.orm import Session
 from app import logs, sysconfig
 from app.config import settings
 from app.models import PlexItem, PlexPerson, PlexShow
+from app.plex.cards import display_title
+from app.plex.people import people_by_credits, warm_portraits
 
 log = logs.get_logger(__name__)
 # Politeness throttle: TMDB tolerates ~50 req/s; ~33/s leaves comfortable headroom for a shared host.
@@ -124,9 +128,17 @@ class TmdbClient:
         """Resolve an external id (imdb_id / tvdb_id) to TMDB items via /find."""
         return self._get(f"/find/{external_id}", {"external_source": source})
 
+    def countries(self) -> list[dict] | None:
+        """TMDB's country list: `[{iso_3166_1, english_name, native_name}]` (a JSON array, which
+        `_get`'s dict annotation does not describe)."""
+        return cast("list[dict] | None", self._get("/configuration/countries"))
+
     def fetch_person(self, person_id: int) -> dict | None:
-        """One person's bio + combined credits (E5 S6), in a single append_to_response request."""
-        return self._get(f"/person/{person_id}", {"append_to_response": "combined_credits"})
+        """One person's bio + combined credits + the bio's translations, in a single
+        append_to_response request."""
+        return self._get(
+            f"/person/{person_id}", {"append_to_response": "combined_credits,translations"}
+        )
 
 
 def guid_to_tmdb_id(guid: str | None, media_type: str, tmdb: TmdbClient) -> int | None:
@@ -214,7 +226,7 @@ def parse_tmdb(payload: dict, media_type: str) -> dict:
     crew.sort(key=lambda c: priority.get(c["job"], 99))
     crew = crew[:12]
 
-    # name -> tmdb person id, for the plex_people foundation (S6 fetches bios off these ids).
+    # name -> tmdb person id, for the plex_people foundation (bios are fetched off these ids).
     people: list[dict] = []
     for c in (credits.get("cast") or [])[:30]:
         if c.get("name") and c.get("id"):
@@ -254,8 +266,12 @@ def parse_tmdb(payload: dict, media_type: str) -> dict:
             c.get("name") for c in (payload.get("production_companies") or []) if c.get("name")
         ][:8]
         or None,
+        # The ISO code beside TMDB's English name: the client names the country in the reader's
+        # language from the code; the name is the fallback for a code it cannot name.
         "production_countries": [
-            c.get("name") for c in (payload.get("production_countries") or []) if c.get("name")
+            {"code": c.get("iso_3166_1") or None, "name": c["name"]}
+            for c in (payload.get("production_countries") or [])
+            if c.get("name")
         ]
         or None,
         "recommendations": recs or None,
@@ -301,8 +317,8 @@ def _link_people(db: Session, links: dict[str, int]) -> None:
 
 
 def enrich_batch(db: Session, tmdb: TmdbClient, limit: int) -> int:
-    """Enrich up to `limit` not-yet-TMDB-enriched titles (movies first, shows fill the remainder —
-    mirrors _enrich_people_batch). A transient TmdbError leaves the watermark NULL for retry; a
+    """Enrich up to `limit` not-yet-TMDB-enriched titles (movies first, shows fill the
+    remainder). A transient TmdbError leaves the watermark NULL for retry; a
     definitive no-match (unresolvable guid / TMDB 404) still stamps the watermark so it isn't re-fetched
     every run. One commit at the end. Returns the number processed (matched OR definitively unmatched)."""
     movies = (
@@ -356,13 +372,62 @@ def enrich_batch(db: Session, tmdb: TmdbClient, limit: int) -> int:
     return processed
 
 
+def _country(c: dict | str) -> dict:
+    """One stored production country as served: `{code, name}`. An older row stored the
+    bare English name; until `backfill_country_codes` reaches it, it serves with no code (the client
+    shows the name)."""
+    return c if isinstance(c, dict) else {"code": None, "name": c}
+
+
+def backfill_country_codes(db: Session, tmdb: TmdbClient) -> int:
+    """Give the older countries (bare English names) their ISO codes, from TMDB's own
+    country list — the names came from that list, so they match it exactly. One request, and only
+    while such a row is left: re-fetching each title instead would take the enrichment batch days.
+    A name the list does not hold keeps no code. Returns the number of rows rewritten."""
+    stale = [
+        row
+        for model in (PlexItem, PlexShow)
+        for row in db.execute(
+            select(model).where(func.jsonb_typeof(model.production_countries[0]) == "string")
+        ).scalars()
+    ]
+    if not stale:
+        return 0
+    codes = {
+        c["english_name"]: c["iso_3166_1"]
+        for c in tmdb.countries() or []
+        if c.get("english_name") and c.get("iso_3166_1")
+    }
+    for row in stale:
+        row.production_countries = [
+            {"code": codes.get(c), "name": c} if isinstance(c, str) else c
+            for c in row.production_countries or []
+        ]
+    db.commit()
+    return len(stale)
+
+
 def run_enrich(db: Session) -> dict:
     """Scheduler entry point. No-op (never an error) when the TMDB key isn't configured."""
     if not sysconfig.get_str(db, "tmdb_api_key"):
         return {"enriched": 0, "skipped": "tmdb_api_key not set"}
     with TmdbClient(db) as tmdb:
+        coded = backfill_country_codes(db, tmdb)
         n = enrich_batch(db, tmdb, settings.tmdb_enrich_batch)
-    return {"enriched": n}
+    return {"enriched": n, **({"countries_coded": coded} if coded else {})}
+
+
+def run_people_enrich(db: Session) -> dict:
+    """Scheduler entry point of people_enrich: a batch of person bios (the most-credited
+    first) and a batch of portraits. Each half is a no-op — never an error — without its TMDB key /
+    without Plex."""
+    batch = settings.people_enrich_batch
+    if sysconfig.get_str(db, "tmdb_api_key"):
+        with TmdbClient(db) as tmdb:
+            bios = enrich_people_batch(db, tmdb, batch)
+    else:
+        bios = {"skipped": "tmdb_api_key not set"}
+    return {"bios": bios, "portraits": warm_portraits(db, batch)}
 
 
 def resolve_library_guids(
@@ -406,10 +471,11 @@ def resolve_library_guids(
     return movies, shows
 
 
-def resolve_recs(db: Session, recs: list[dict] | None) -> list[dict]:
+def resolve_recs(db: Session, recs: list[dict] | None, *, lang: str) -> list[dict]:
     """Tag each recommendation with whether it's in OUR library (matched by guid `tmdb:<id>`, bucketed
     by media_type — see resolve_library_guids) and, if so, its rating_key — so the client can deep-link
-    the ones we have and merely display the rest."""
+    the ones we have and merely display the rest. One we have is named as the library names it, in
+    `lang`; one we lack keeps the title TMDB gave."""
     if not recs:
         return []
     movies, shows = resolve_library_guids(db, recs)
@@ -417,13 +483,19 @@ def resolve_recs(db: Session, recs: list[dict] | None) -> list[dict]:
     for r in recs:
         g = f"tmdb:{r.get('tmdb_id')}"
         row = shows.get(g) if r.get("media_type") == "tv" else movies.get(g)
+        named = {"title": display_title(row, lang)} if row is not None else {}
         out.append(
-            {**r, "in_library": row is not None, "rating_key": row.rating_key if row else None}
+            {
+                **r,
+                **named,
+                "in_library": row is not None,
+                "rating_key": row.rating_key if row else None,
+            }
         )
     return out
 
 
-def serve_block(db: Session, row) -> dict | None:
+def serve_block(db: Session, row, *, lang: str) -> dict | None:
     """The TMDB extras for a movie/show detail response, served from the row (no live TMDB call).
     None until the row has been TMDB-enriched (the client then shows nothing extra)."""
     if row.tmdb_enriched_at is None:
@@ -437,30 +509,58 @@ def serve_block(db: Session, row) -> dict | None:
         "keywords": row.keywords or [],
         "crew": row.crew or [],
         "studios": row.studios or [],
-        "production_countries": row.production_countries or [],
-        "recommendations": resolve_recs(db, row.recommendations),
+        "production_countries": [_country(c) for c in row.production_countries or []],
+        "recommendations": resolve_recs(db, row.recommendations, lang=lang),
         "backdrops": row.backdrops or [],
     }
 
 
-# --- Person enrichment (E5 S6) -------------------------------------------------------------------
-# A person's bio + birth facts + a "Known for" credit pool, fetched once (the first person-page open)
-# and served from the plex_people row. `known_for` stores the top combined-credits ids; the person
+# --- Person enrichment -------------------------------------------------------------------
+# A person's bio + birth facts + a "Known for" credit pool, fetched once — by the people_enrich job, or
+# by the first person-page open if the job has not reached them yet — and served from the plex_people row. `known_for` stores the top combined-credits ids; the person
 # endpoint resolves them to library rows by guid at serve time (like recommendations).
 
-_KNOWN_FOR_MAX = 15
+# The candidate pool kept per person. Only the titles that are in the library are shown (at most
+# routes.plex.people._KNOWN_FOR_SHOWN), so the pool is deeper than the rail: a star's 15 most-voted
+# credits are often not the 15 a given library holds.
+_KNOWN_FOR_MAX = 40
+# TMDB genre ids of the shows a person visits as a guest (News, Reality, Talk). Ranked by popularity,
+# talk-show visits crowded out the films (Tom Cruise: 14 of 15 were talk shows).
+_GUEST_GENRES = {10763, 10764, 10767}
+# A credit in which the person plays themselves ("Self", "Himself - Guest", "Herself (archive footage)").
+_SELF_CREDIT = re.compile(r"^\s*(self|himself|herself|themselves)\b", re.IGNORECASE)
+# The crew jobs that make a title the person's own work (a director is known for what they directed).
+_KNOWN_FOR_CREW_JOBS = {"Director", "Creator"}
+
+
+def _known_for_credit(c: dict, crew: bool) -> bool:
+    if crew:
+        return c.get("job") in _KNOWN_FOR_CREW_JOBS
+    if _SELF_CREDIT.match(c.get("character") or ""):
+        return False
+    return not _GUEST_GENRES.intersection(c.get("genre_ids") or [])
 
 
 def parse_person(payload: dict) -> dict:
     """TMDB /person/{id} (+combined_credits) → the persisted bio fields. `known_for` = the person's
-    most-popular movie/tv credits (deduped by id), top `_KNOWN_FOR_MAX`, resolved to library later."""
-    credits = (payload.get("combined_credits") or {}).get("cast") or []
+    most-voted movie/tv credits — roles and directed works, not guest visits as themselves — deduped,
+    the top `_KNOWN_FOR_MAX`, resolved to library rows at serve time. Ranked by vote count, as TMDB's
+    own "Known for" is: popularity is this week's attention, which a talk-show visit wins."""
+    combined = payload.get("combined_credits") or {}
+    credits = [c for c in combined.get("cast") or [] if _known_for_credit(c, crew=False)] + [
+        c for c in combined.get("crew") or [] if _known_for_credit(c, crew=True)
+    ]
     # Dedup on (id, media_type), not the bare id: TMDB movie ids and tv ids are independent sequences
     # (id 603 can be both a movie and a show — see resolve_library_guids), so a same-id movie+show pair
     # is two distinct credits and both must survive.
     seen: set[tuple[int, str]] = set()
     known: list[dict] = []
-    for c in sorted(credits, key=lambda c: c.get("popularity") or 0, reverse=True):
+    ranked = sorted(
+        credits,
+        key=lambda c: (c.get("vote_count") or 0, c.get("popularity") or 0),
+        reverse=True,
+    )
+    for c in ranked:
         tid, mt = c.get("id"), c.get("media_type")
         if not tid or mt not in ("movie", "tv") or (tid, mt) in seen:
             continue
@@ -470,6 +570,7 @@ def parse_person(payload: dict) -> dict:
             break
     return {
         "biography": (payload.get("biography") or "").strip() or None,
+        "biography_hu": _translated_bio(payload, "hu"),
         "birthday": payload.get("birthday") or None,
         "deathday": payload.get("deathday") or None,
         "place_of_birth": payload.get("place_of_birth") or None,
@@ -479,8 +580,17 @@ def parse_person(payload: dict) -> dict:
     }
 
 
+def _translated_bio(payload: dict, lang: str) -> str | None:
+    """The `lang` biography from the appended `translations` (TMDB leaves an untranslated one blank)."""
+    for tr in (payload.get("translations") or {}).get("translations") or []:
+        if tr.get("iso_639_1") == lang:
+            return ((tr.get("data") or {}).get("biography") or "").strip() or None
+    return None
+
+
 def apply_person(row: PlexPerson, parsed: dict) -> None:
     row.biography = parsed["biography"]
+    row.biography_hu = parsed["biography_hu"]
     row.birthday = parsed["birthday"]
     row.deathday = parsed["deathday"]
     row.place_of_birth = parsed["place_of_birth"]
@@ -490,11 +600,22 @@ def apply_person(row: PlexPerson, parsed: dict) -> None:
     row.tmdb_enriched_at = datetime.now(timezone.utc)
 
 
+def _fetch_person_into(tmdb: TmdbClient, person: PlexPerson) -> None:
+    """Fetch one person from TMDB and write the row (not committed). A 404 (deleted person) stamps the
+    watermark with empty fields so it isn't fetched again. Raises TmdbError on a transient failure,
+    leaving the row untouched and `tmdb_enriched_at` NULL for a retry."""
+    payload = tmdb.fetch_person(cast(int, person.tmdb_person_id))
+    if payload is None:
+        person.tmdb_enriched_at = datetime.now(timezone.utc)
+    else:
+        apply_person(person, parse_person(payload))
+
+
 def enrich_person(db: Session, person: PlexPerson) -> bool:
-    """Lazily fetch + persist one person's TMDB bio (first person-page open), committing the write so it
-    survives regardless of the read path. No-op (False) without a key or a `tmdb_person_id`; a transient
-    error leaves `tmdb_enriched_at` NULL so the next open retries. A 404 (deleted person) stamps the
-    watermark with empty fields so we don't refetch it every open."""
+    """Lazily fetch + persist one person's TMDB bio (a person page opened before the background job
+    reached them), committing the write so it survives regardless of the read path. No-op (False)
+    without a key or a `tmdb_person_id`; a transient error leaves `tmdb_enriched_at` NULL so the next
+    open retries."""
     if not person.tmdb_person_id or not sysconfig.get_str(db, "tmdb_api_key"):
         return False
     try:
@@ -502,7 +623,7 @@ def enrich_person(db: Session, person: PlexPerson) -> bool:
         # longer blocks a serve; still cap the outbound TMDB call with a short timeout so a hanging TMDB
         # doesn't tie up a threadpool worker for the default 30s.
         with TmdbClient(db, timeout=8.0) as tmdb:
-            payload = tmdb.fetch_person(person.tmdb_person_id)
+            _fetch_person_into(tmdb, person)
     except TmdbError as exc:
         # False leaves the person un-enriched and unstamped, so it is retried — but the page in
         # front of the user right now shows an empty biography with nothing to say why.
@@ -515,9 +636,41 @@ def enrich_person(db: Session, person: PlexPerson) -> bool:
             ),
         )
         return False
-    if payload is None:
-        person.tmdb_enriched_at = datetime.now(timezone.utc)
-    else:
-        apply_person(person, parse_person(payload))
     db.commit()
     return True
+
+
+# Lookups failing this many people in a row end a people batch: TMDB itself is degraded, and hammering
+# it helps nobody. One person failing alone is skipped, so an id TMDB keeps refusing never stalls the rest.
+_PEOPLE_FAILS_IN_A_ROW = 3
+
+
+def enrich_people_batch(db: Session, tmdb: TmdbClient, limit: int) -> dict:
+    """Fetch up to `limit` linked, not-yet-enriched people, the most-credited first, so a
+    person page reads its bio from the row instead of waiting on TMDB. Each person is committed as it
+    lands: a page open during the run already sees it. A failed lookup is skipped (left unstamped for the
+    next run); _PEOPLE_FAILS_IN_A_ROW of them in a row end the batch."""
+    ids = people_by_credits(db, "bio", limit)
+    done = failed = in_a_row = 0
+    for pid in ids:
+        if in_a_row >= _PEOPLE_FAILS_IN_A_ROW:
+            break
+        person = db.get(PlexPerson, pid)
+        # A page open may have enriched this person since the select: skip, don't fetch twice.
+        if person is None or person.tmdb_enriched_at is not None:
+            continue
+        try:
+            _fetch_person_into(tmdb, person)
+        except TmdbError as exc:
+            db.rollback()
+            failed, in_a_row = failed + 1, in_a_row + 1
+            log.warning(
+                "a TMDB person lookup in the batch failed",
+                extra=logs.event(
+                    "plex.tmdb.people_batch_failed", person_id=pid, reason=type(exc).__name__
+                ),
+            )
+            continue
+        db.commit()
+        done, in_a_row = done + 1, 0
+    return {"enriched": done, **({"failed": failed} if failed else {})}

backend/app/plex/watch_sync.py

@@ -11,7 +11,7 @@ already pages) and upsert them into the owner's `plex_states`. Plex wins on over
 exist only in Siftlode are left untouched (union, Plex-authoritative on the intersection).
 
 Phases B (Siftlode→Plex push via scrobble/timeline) and C (incremental Plex→Siftlode via the watch
-history + onDeck) build on this and land in later ships. As of S2 Plex is the source of truth: the
+history + onDeck) build on this and land in later ships. Plex is now the source of truth: the
 player pushes the resume position back to Plex on the periodic checkpoint (not just the final flush),
 and the incremental pull resolves a genuine divergence to Plex's state (see `_pull_apply`) rather than
 running a last-write-wins comparison.
@@ -29,7 +29,7 @@ from sqlalchemy.orm import Session
 
 from app import logs, sysconfig
 from app.db import SessionLocal
-from app.models import PlexItem, PlexLink, PlexState
+from app.models import PlexItem, PlexLibrary, PlexLink, PlexState
 from app.plex.client import PlexClient, PlexError, PlexNotConfigured
 from app.plex.sync import (
     _TYPE_EPISODE,
@@ -72,13 +72,27 @@ def _rk_to_id(db: Session) -> dict[str, int]:
 
 def _scan_plex_states(
     db: Session, plex: PlexClient
-) -> tuple[dict[int, tuple[str, int, datetime | None, datetime | None]], int]:
+) -> tuple[dict[int, tuple[str, int, datetime | None, datetime | None]], int, set[str], set[str]]:
     """Scan the enabled movie/show sections and return ``{item_id: (status, pos, watched_at,
-    prog_at)}`` for every mirrored leaf Plex holds a watch signal for, plus the scanned-leaf count.
-    Shared by the one-time import and the full reconcile (both need Plex's whole current picture)."""
+    prog_at)}`` for every mirrored leaf Plex holds a watch signal for, the scanned-leaf count, the
+    keys of the sections SETTLED — walked to the server's total (or said empty by a server we mirror
+    nothing from) — and the keys of every section walked.
+    Shared by the one-time import and the full reconcile (both need Plex's whole current picture).
+    The settled keys matter to the reconcile alone: it reads "absent from the scan" as "Plex
+    un-watched it", which a walk cut short by a Plex hiccup does not mean (see `_paginate`). Settled
+    is decided PER SECTION: one section that never settles (a library emptied on Plex while its
+    mirrored items are kept) must not stop un-watches syncing in all the others."""
     item_id_by_rk = _rk_to_id(db)
     wanted = _enabled_section_keys(db)
+    # The sections we mirror something from: a total of 0 from one of those contradicts the mirror.
+    mirrored = set(
+        db.execute(
+            select(PlexLibrary.plex_key).join(PlexItem, PlexItem.library_id == PlexLibrary.id)
+        ).scalars()
+    )
     scanned = 0
+    settled: set[str] = set()
+    walked: set[str] = set()
     out: dict[int, tuple[str, int, datetime | None, datetime | None]] = {}
     for s in plex.sections():
         stype = s.get("type")
@@ -88,7 +102,10 @@ def _scan_plex_states(
         if wanted is not None and key not in wanted:
             continue
         item_type = _TYPE_MOVIE if stype == "movie" else _TYPE_EPISODE
-        for meta in _paginate(plex, key, item_type):
+        walked.add(key)
+        progress: dict = {}
+        before = scanned
+        for meta in _paginate(plex, key, item_type, progress):
             scanned += 1
             item_id = item_id_by_rk.get(str(meta.get("ratingKey") or ""))
             if item_id is None:
@@ -98,7 +115,15 @@ def _scan_plex_states(
             )
             if res is not None:
                 out[item_id] = res
-    return out, scanned
+        # A section the server SAYS is empty (total 0, nothing listed) settles too. Believed only for
+        # a section we mirror nothing from: one whose items we hold answering 0 is a Plex that lost
+        # the library for a moment, not one that un-watched all of it — so it is left
+        # unsettled, and its rows are not cleared.
+        listed = scanned > before
+        said_empty = progress.get("total") == 0 and not listed and key not in mirrored
+        if (progress.get("complete") and listed) or said_empty:
+            settled.add(key)
+    return out, scanned, settled, walked
 
 
 def import_owner_watch_state(db: Session, link: PlexLink) -> dict:
@@ -111,7 +136,8 @@ def import_owner_watch_state(db: Session, link: PlexLink) -> dict:
     stats = {"watched": 0, "in_progress": 0, "scanned": 0}
     try:
         with PlexClient(db) as plex:
-            scanned_states, stats["scanned"] = _scan_plex_states(db, plex)
+            # A partial scan is still safe here: the import only ever ADDS what Plex reports.
+            scanned_states, stats["scanned"], _settled, _walked = _scan_plex_states(db, plex)
             # One row per item (keyed by item_id → inherently de-duped) then bulk-UPSERT. An upsert is
             # idempotent AND concurrency-safe: a repeated/overlapping enable (the toggle firing twice
             # while the multi-second import runs) can't raise a duplicate-key error — it re-updates.
@@ -376,7 +402,7 @@ def _pull_apply(
         if st is not None and not st.synced_to_plex:
             st.synced_to_plex = True  # already matches Plex → just settle the flag, no churn
         return None
-    # Plex-authoritative (S2): on a genuine divergence for an item Plex knows about, Plex wins — no
+    # Plex-authoritative: on a genuine divergence for an item Plex knows about, Plex wins — no
     # last-write-wins tug-of-war. This is safe because prompt write-back (playback.py pushes the resume
     # position to Plex on the ~10s checkpoint, not just on the final flush) keeps Plex current, so a
     # local edit that's "newer" than Plex has almost always already been pushed and matched above by
@@ -590,6 +616,27 @@ def run_plex_watch_sync(db: Session) -> dict:
     return total
 
 
+def _clear_unwatched(db: Session, user_id: int, plex_states: dict, settled: set[str]) -> int:
+    """Clear the user's rows Plex no longer has a record for — the mirrored ones (Plex un-watched
+    them); hidden and dirty rows stay (a dirty one is re-pushed by `_repush_dirty`). Only rows of the
+    sections the scan SETTLED: a walk a Plex hiccup cut short leaves out rows it merely failed to
+    reach, and clearing those un-watched them here until the next complete run; an item of
+    a library the allowlist leaves out is absent from every scan, and was cleared on every run."""
+    n = 0
+    rows = db.execute(
+        select(PlexState)
+        .join(PlexItem, PlexItem.id == PlexState.item_id)
+        .join(PlexLibrary, PlexLibrary.id == PlexItem.library_id)
+        .where(PlexState.user_id == user_id, PlexLibrary.plex_key.in_(settled))
+    )
+    for st in rows.scalars().all():
+        if st.item_id in plex_states or st.status == "hidden" or not st.synced_to_plex:
+            continue
+        db.delete(st)
+        n += 1
+    return n
+
+
 def run_plex_watch_reconcile(db: Session) -> dict:
     """Scheduler job (default daily) — the authoritative full reconcile. Rescans every section and,
     per item, uses `synced_to_plex` to settle what the incremental feed can't:
@@ -608,7 +655,7 @@ def run_plex_watch_reconcile(db: Session) -> dict:
         try:
             with PlexClient(db) as plex:
                 _resolve_account_id(db, plex, link)
-                plex_states, scanned = _scan_plex_states(db, plex)
+                plex_states, scanned, settled, walked = _scan_plex_states(db, plex)
                 total["scanned"] += scanned
 
                 for item_id, (status, pos, watched_at, prog_at) in plex_states.items():
@@ -616,18 +663,19 @@ def run_plex_watch_reconcile(db: Session) -> dict:
                     if ch:
                         total["applied"] += 1
 
-                # Siftlode rows Plex no longer has a record for: clear the ones that were mirrored
-                # (Plex un-watched them); leave hidden + dirty (dirty gets re-pushed just below).
-                for st in (
-                    db.execute(select(PlexState).where(PlexState.user_id == link.user_id))
-                    .scalars()
-                    .all()
-                ):
-                    if st.item_id in plex_states or st.status == "hidden":
-                        continue
-                    if st.synced_to_plex:
-                        db.delete(st)
-                        total["cleared"] += 1
+                total["cleared"] += _clear_unwatched(db, link.user_id, plex_states, settled)
+                if unsettled := sorted(walked - settled):
+                    total["incomplete"] = total.get("incomplete", 0) + 1
+                    log.warning(
+                        "Plex sections %s were cut short or answered empty; their rows are not cleared"
+                        " this run",
+                        ", ".join(unsettled),
+                        extra=logs.event(
+                            "plex.watch.reconcile_incomplete",
+                            user_id=link.user_id,
+                            sections=unsettled,
+                        ),
+                    )
 
                 total["repushed"] += _repush_dirty(db, plex, link)
                 link.last_watch_sync_at = datetime.now(timezone.utc)

backend/app/progress.py

@@ -8,7 +8,7 @@ its own thread/context) never cross wires. When no sink is bound (e.g. a manual
 """
 
 import contextvars
-from typing import Callable, Optional
+from typing import Callable, Literal, Optional
 
 Sink = Callable[[int, Optional[int], Optional[str]], None]
 
@@ -24,6 +24,19 @@ def report(current: int, total: int | None = None, phase: str | None = None) ->
         sink(current, total, phase)
 
 
+class JobFailed(RuntimeError):
+    """A job's own failure, said so the reader's language can name it: the client reads
+    `scheduler.failure.<reason>` and interpolates `params` (keep them to values the reader may see —
+    numbers, ids, short codes). The message is the English line the log and `last_error` keep. An
+    upstream's own error (a timeout, a refused connection) is not one of these: it stays as raised,
+    being diagnostic."""
+
+    def __init__(self, reason: str, message: str, **params: object):
+        super().__init__(message)
+        self.reason = reason
+        self.params = params
+
+
 class bind:
     """Context manager that binds a progress sink for the duration of a job run."""
 
@@ -35,7 +48,9 @@ class bind:
         self._token = _sink.set(self._sink)
         return self
 
-    def __exit__(self, *exc) -> bool:
+    # Literal[False], not bool: this never swallows the job's exception, and saying so is what lets
+    # a type checker know a name assigned inside `with bind(...)` is bound after it.
+    def __exit__(self, *exc) -> Literal[False]:
         if self._token is not None:
             _sink.reset(self._token)
         return False

backend/app/quota.py

@@ -7,11 +7,13 @@ always be allowed; expensive backfill must yield to the remaining budget.
 
 import contextlib
 import contextvars
-from datetime import datetime, timezone
+from datetime import datetime, timedelta, timezone
+from typing import Any, cast
 from zoneinfo import ZoneInfo
 
-from sqlalchemy import func, select
+from sqlalchemy import delete, func, select
 from sqlalchemy.dialects.postgresql import insert as pg_insert
+from sqlalchemy.engine import CursorResult
 from sqlalchemy.orm import Session
 
 from app import sysconfig
@@ -96,6 +98,20 @@ def pacific_today():
     return datetime.now(_PACIFIC).date()
 
 
+def gc(db: Session, retention_days: int) -> int:
+    """Delete quota-history rows older than `retention_days` (<= 0 keeps them forever) and return how
+    many went. Flushes only; the cleanup job commits. Today's rows are never in reach: the per-user
+    search cap counts them, and the floor of two days outlasts any Pacific day (25 h at most)."""
+    if retention_days <= 0:
+        return 0
+    cutoff = datetime.now(timezone.utc) - timedelta(days=max(2, retention_days))
+    # A DELETE answers with a CursorResult; `execute` is typed as the general Result, without rowcount.
+    result = cast(
+        CursorResult[Any], db.execute(delete(QuotaEvent).where(QuotaEvent.created_at < cutoff))
+    )
+    return result.rowcount or 0
+
+
 def pacific_day_start_utc() -> datetime:
     """UTC instant of the current Pacific day's midnight — aligns 'today' with the budget reset."""
     start = datetime.now(_PACIFIC).replace(hour=0, minute=0, second=0, microsecond=0)

backend/app/realtime.py

@@ -1,68 +0,0 @@
-"""In-process WebSocket connection registry for live server→client push.
-
-Generic groundwork (notification module phase 2 is the first user): connections are kept in
-this process's memory, keyed by user_id, so a request handler can push an event to all of a
-user's open tabs/devices. Single-uvicorn-process assumption — if we ever run multiple workers,
-replace the in-memory map with a Redis pub/sub fan-out behind the same `push()` interface (same
-note as app.ratelimit).
-
-`push()` is intentionally SYNC-callable: ordinary (sync) route handlers run in a threadpool, so
-they schedule the actual send onto the WebSocket event loop via run_coroutine_threadsafe. The
-connect/disconnect/_push coroutines all run on that one loop, so the connection map is only ever
-mutated from a single thread.
-"""
-
-import asyncio
-from collections import defaultdict
-
-from starlette.websockets import WebSocket
-from app import logs
-
-log = logs.get_logger(__name__)
-
-
-class ConnectionManager:
-    def __init__(self) -> None:
-        self._conns: dict[int, set[WebSocket]] = defaultdict(set)
-        self._loop: asyncio.AbstractEventLoop | None = None
-
-    async def connect(self, user_id: int, ws: WebSocket) -> None:
-        self._loop = asyncio.get_running_loop()
-        await ws.accept()
-        self._conns[user_id].add(ws)
-
-    async def disconnect(self, user_id: int, ws: WebSocket) -> None:
-        conns = self._conns.get(user_id)
-        if conns:
-            conns.discard(ws)
-            if not conns:
-                self._conns.pop(user_id, None)
-
-    async def _push(self, user_id: int, payload: dict) -> None:
-        for ws in list(self._conns.get(user_id, ())):
-            try:
-                await ws.send_json(payload)
-            except Exception as exc:  # noqa: BLE001 — one dead socket must not stop the others
-                # Stale socket; it'll be cleaned up by its own disconnect handler. But this is
-                # also where a live update goes missing: the UI simply never changes, and there
-                # is no request behind it whose failure anyone could look up.
-                log.debug(
-                    "a live update could not be pushed to a socket",
-                    extra=logs.event(
-                        "realtime.push_failed", user_id=user_id, reason=type(exc).__name__
-                    ),
-                )
-
-    def push(self, user_id: int, payload: dict) -> None:
-        """Schedule a push onto the WebSocket event loop. Safe to call from a sync route
-        handler (threadpool). No-op if no WebSocket has ever connected (no loop captured)."""
-        loop = self._loop
-        if loop is None:
-            return
-        try:
-            asyncio.run_coroutine_threadsafe(self._push(user_id, payload), loop)
-        except Exception:  # noqa: BLE001 — a closed loop or socket must not fail the caller
-            log.debug("ws push failed", exc_info=True)
-
-
-manager = ConnectionManager()

backend/app/routes/_common.py

@@ -1,4 +1,4 @@
-"""Route-layer shared helpers for the API contract (R6 S1a).
+"""Route-layer shared helpers for the API contract.
 
 Small, dependency-light guards that several route modules had hand-rolled with subtle
 drift. Keeping them here (imported by the route modules) means the "fetch a row or 404"
@@ -60,14 +60,17 @@ def owned_or_404(
     return obj
 
 
-def youtube_refusal(exc: Exception) -> HTTPException:
+def youtube_refusal(exc: Exception, *, playlist: bool = False) -> HTTPException:
     """The answer to an action YouTube refused: a 422 whose reason the UI translates
     (`errors.youtube.<reason>`). One definition, because the routes had grown three shapes of it —
     a bare English string quoting the upstream message, a playlist-scoped reason, and nothing at
     all (a raw 500 from the manual syncs). The upstream failure is already logged where it
     happened (`youtube.api.error`). A failure that is not the Data API's (the search scraper's) is
-    YouTube being unavailable to us — the retryable answer."""
-    reason = refusal_reason(exc) if isinstance(exc, YouTubeError) else "unavailable"
+    YouTube being unavailable to us — the retryable answer. `playlist`: the route acts on one
+    playlist the user picked, so a missing playlist is THAT one, gone (`refusal_reason`)."""
+    reason = (
+        refusal_reason(exc, playlist=playlist) if isinstance(exc, YouTubeError) else "unavailable"
+    )
     return HTTPException(status_code=422, detail={"code": "youtube", "reason": reason})
 
 
@@ -125,7 +128,7 @@ async def capped_stream_to(
 
 
 def issue_refused(exc: gitea.GiteaError) -> HTTPException:
-    """The refusal an issue-tracker failure (E16) answers as — the send, the preview and the
+    """The refusal an issue-tracker failure answers as — the send, the preview and the
     admin's connection test alike. Literal reasons, one per case: the frontend's contract test reads
     them to require a sentence for each in both languages."""
     match exc.reason:

backend/app/routes/admin.py

@@ -1,5 +1,4 @@
-"""Admin-only onboarding: review access requests (Invites), approve/deny, manual add.
-Also the demo-account admin surface: the demo email whitelist + a manual state reset."""
+"""Admin-only onboarding: review access requests (Invites), approve/deny, manual add."""
 
 from collections.abc import Iterable
 from datetime import datetime, timezone
@@ -19,15 +18,12 @@ from app.auth import (
     count_admins,
     env_allowed_canonicals,
     find_user_by_email,
-    get_or_create_demo_user,
     is_allowed,
     operator_contact,
     purge_user,
 )
 from app.db import get_db
-from app.demo import reset_demo_state
 from app.models import (
-    DemoWhitelist,
     Invite,
     User,
 )
@@ -131,7 +127,7 @@ def _welcome_mail_parts(db: Session, email: str) -> tuple[str | None, bool, int]
     """
     hours = int(APPROVAL_VERIFY_TTL.total_seconds() // 3600)
     user = find_user_by_email(db, email)
-    if user is None or user.is_demo:
+    if user is None:
         return None, False, hours
     if user.email_verified or not email_mod.email_enabled(db):
         return None, True, hours
@@ -142,20 +138,21 @@ def _activate_user_by_email(db: Session, email: str) -> None:
     """Approving access activates a matching (e.g. pending password-registration) account so
     it can sign in. No-op if no such user exists yet (a Google user is created at first login).
 
-    E9.5 S1: this path deliberately does NOT revoke a preset password. Revoking here destroyed the
+    This path deliberately does NOT revoke a preset password. Revoking here destroyed the
     password a legitimate registrant had just chosen and made "your password was cleared for
     security" the DEFAULT experience, because approve-before-verify is the normal ordering.
 
-    S1 justified that by saying `password_login` refuses an unverified account BEFORE it looks at
-    `is_active`, so a pre-emptor could not sign in "approved or not". A review of the finished epic
-    showed that argument only holds while the address stays unverified — and S6's approval mail
-    hands the real owner a confirmation link precisely so they will end that state. Nothing replaced
-    the revoke, and that is deliberate: round 2 of the same review tried "the newest registration
-    owns an unverified account" and reverted it, because the rule is symmetric — an attacker can
-    always register last — and it handed anonymous callers a write onto a real account's credential.
-    The owner's recourse is password-reset, which proves the mailbox, rewrites the hash and bumps
-    the session epoch. The residual is logged as SEC-16. The revoke also stays on the Google-adoption path
-    (auth.revoke_preset_password_if_unverified), where ownership is actually attested."""
+    That was once justified by saying `password_login` refuses an unverified account BEFORE it
+    looks at `is_active`, so a pre-emptor could not sign in "approved or not". A review of the
+    finished epic showed that argument only holds while the address stays unverified — and the
+    approval mail hands the real owner a confirmation link precisely so they will end that state.
+    Nothing replaced the revoke, and that is deliberate: the same review later tried "the newest
+    registration owns an unverified account" and reverted it, because the rule is symmetric — an
+    attacker can always register last — and it handed anonymous callers a write onto a real
+    account's credential. The owner's recourse is password-reset, which proves the mailbox, rewrites
+    the hash and bumps the session epoch. The residual is logged as SEC-16. The revoke also stays on
+    the Google-adoption path (auth.revoke_untrusted_preset_password), where ownership is
+    actually attested."""
     # Canonical, with the literal spelling preferred when both exist: approving
     # `npeter83.temu@gmail.com` must activate the account that registered as `npeter83temu@gmail.com`,
     # since Gmail delivers both to one mailbox and the admin decided about the mailbox.
@@ -400,19 +397,18 @@ def _serialize_user(u: User) -> dict:
         "is_active": u.is_active,
         "is_suspended": u.is_suspended,
         "email_verified": u.email_verified,
-        "is_demo": u.is_demo,
         "has_password": u.password_hash is not None,
         # Distinguishes "this account never had a password" (Google-only) from "SEC-10 cleared it" —
         # the same distinction _send_reset_if_eligible needs, and the one an admin has to see before
         # deciding what kind of help someone needs.
         "password_revoked_at": iso(u.password_revoked_at),
         "has_google": u.google_sub is not None,
-        # Only when it DIFFERS from the account address (E9.5 S3): "connected" is not the same
+        # Only when it DIFFERS from the account address: "connected" is not the same
         # statement as "connected as someone else", and the second one is the one that explains a
         # confusing sign-in.
         "google_email": u.google_email if u.google_email != u.email else None,
         "created_at": iso(u.created_at),
-        # NULL = no sign-in since the column existed (0085), or the shared demo account.
+        # NULL = no sign-in since the column existed (0085).
         "last_login_at": iso(u.last_login_at),
     }
 
@@ -435,14 +431,12 @@ def set_user_role(
     admin: User = Depends(admin_user),
     db: Session = Depends(get_db),
 ) -> dict:
-    """Promote/demote a user. Guards: can't change your own role, can't touch the demo
-    account, and can't remove the last remaining admin. Emails the user when it actually changes."""
+    """Promote/demote a user. Guards: can't change your own role and can't remove the
+    last remaining admin. Emails the user when it actually changes."""
     role = body.role
     if role not in ("user", "admin"):
         raise refused(400, "invalid_role")
     target = get_or_404(db, User, user_id, what="user")
-    if target.is_demo:
-        raise refused(400, "demo_role")
     if target.id == admin.id:
         raise refused(400, "own_role")
     if (
@@ -490,14 +484,11 @@ def set_user_suspended(
     db: Session = Depends(get_db),
 ) -> dict:
     """Suspend or un-suspend an account. A suspended user can't sign in by any method and any
-    live session is dropped on its next request. Guards: can't suspend the demo account, can't
-    suspend yourself, and can't suspend the last active admin (that would lock admin out). On
+    live session is dropped on its next request. Guards: can't suspend yourself, and can't suspend the last active admin (that would lock admin out). On
     un-suspend the user is emailed they can return (suspend is announced reactively, on a blocked
     sign-in)."""
     suspended = body.suspended
     target = get_or_404(db, User, user_id, what="user")
-    if target.is_demo:
-        raise refused(400, "demo_suspend")
     if target.id == admin.id:
         raise refused(400, "own_suspend")
     if (
@@ -538,10 +529,8 @@ def admin_delete_user(
 ) -> dict:
     """Admin-triggered account deletion — the same full erasure a user can perform on themselves
     (cascade delete of all personal data + access-request cleanup + Google-grant revoke). Guards:
-    not the demo account, not yourself (use Settings → Account), and not the last admin."""
+    not yourself (use Settings → Account), and not the last admin."""
     target = get_or_404(db, User, user_id, what="user")
-    if target.is_demo:
-        raise refused(400, "demo_delete")
     if target.id == admin.id:
         raise refused(400, "own_delete")
     if (
@@ -563,10 +552,7 @@ def admin_delete_user(
     return {"deleted": user_id}
 
 
-# --- Demo account: email whitelist + state reset -------------------------------------
-
-
-# --- Stuck sign-ups: the admin lifeline (E9.5 S5, roadmap UX-25) ---------------------------------
+# --- Stuck sign-ups: the admin lifeline -----------------------------------------------------------
 
 
 def _pending_access_rows(db: Session) -> list[User]:
@@ -587,7 +573,6 @@ def _pending_access_rows(db: Session) -> list[User]:
         db.execute(
             select(User)
             .where(
-                User.is_demo.is_(False),
                 User.is_suspended.is_(False),
                 or_(User.is_active.is_(False), User.email_verified.is_(False)),
             )
@@ -625,7 +610,7 @@ def user_access(
     _: User = Depends(admin_user),
     db: Session = Depends(get_db),
 ) -> dict:
-    """Everything the account panel shows about why this person can or cannot get in (E12 S5).
+    """Everything the account panel shows about why this person can or cannot get in.
 
     Its own endpoint rather than extra columns on /users: `is_allowed` and `can_activate` each run a
     query, and the panel opens for ONE user at a time — folding them into the list would pay that cost
@@ -644,7 +629,7 @@ def create_set_password_link(
     admin: User = Depends(admin_user),
     db: Session = Depends(get_db),
 ) -> dict:
-    """Mint a one-time link that lets this account set a password, with no email involved (E12 S5).
+    """Mint a one-time link that lets this account set a password, with no email involved.
 
     The answer to SEC-15, SEC-16 and "no SMTP and someone forgot their password" at once. The admin
     hands the link over by whatever channel they trust; opening it sets a password and signs out
@@ -653,13 +638,8 @@ def create_set_password_link(
     Deliberately NOT gated on the account's state. It is useful for a pre-empted address, for a
     revoked password, for a Google-only account that wants one, and for a plain forgotten password on
     an instance with no mail — narrowing it to one of those would only make the admin's real case the
-    one it refuses. The demo account is the single exception: it is shared, so handing anyone a way to
-    change its password is handing them everyone else's session."""
+    one it refuses."""
     target = get_or_404(db, User, user_id, what="user")
-    if target.is_demo:
-        raise HTTPException(
-            status_code=400, detail={"code": "setpw", "reason": "demo", "email": target.email}
-        )
     url, expires_at = issue_set_password_link(db, target)
     # The URL itself is NOT recorded: the audit log is read by humans and kept long, and a live
     # credential-setting link is the last thing that belongs in it. That the link was issued, by whom,
@@ -692,8 +672,6 @@ def activate_user(
     already do exactly this by re-entering the address in the whitelist box (`add_invite` →
     `_activate_user_by_email`). What was missing was any way to know they had to."""
     target = get_or_404(db, User, user_id, what="user")
-    if target.is_demo:
-        raise refused(400, "demo_activate")
     if target.is_active:
         return _serialize_user(target)  # idempotent: a double click is not an error
     if not is_allowed(db, target.email):
@@ -731,7 +709,7 @@ def resend_user_verification(
     Not rate-limited: an admin is not a threat model to enumerate against, and the deliberate silence
     of the public endpoint is exactly what this exists to escape."""
     target = get_or_404(db, User, user_id, what="user")
-    if target.is_demo or target.email_verified:
+    if target.email_verified:
         raise HTTPException(
             status_code=409,
             detail={"code": "resend", "reason": "not_applicable", "email": target.email},
@@ -757,89 +735,6 @@ def resend_user_verification(
     return {"ok": True, "email": target.email}
 
 
-def _serialize_demo(row: DemoWhitelist) -> dict:
-    return {
-        "id": row.id,
-        "email": row.email,
-        "note": row.note,
-        "added_by": row.added_by,
-        "created_at": iso(row.created_at),
-    }
-
-
-@router.get("/demo/whitelist")
-def list_demo_whitelist(_: User = Depends(admin_user), db: Session = Depends(get_db)) -> list[dict]:
-    rows = (
-        db.execute(select(DemoWhitelist).order_by(DemoWhitelist.created_at.desc())).scalars().all()
-    )
-    return [_serialize_demo(r) for r in rows]
-
-
-class AddDemoWhitelistIn(BaseModel):
-    email: str = ""
-    note: str | None = None
-
-
-@router.post("/demo/whitelist")
-def add_demo_whitelist(
-    body: AddDemoWhitelistIn,
-    admin: User = Depends(admin_user),
-    db: Session = Depends(get_db),
-) -> dict:
-    """Add an email that may enter the shared demo account from the login page (no Google
-    sign-in). Idempotent — re-adding an existing email just returns it."""
-    email = body.email.strip().lower()
-    if not valid_email(email):
-        raise refused(400, "invalid_email")
-    row = db.execute(select(DemoWhitelist).where(DemoWhitelist.email == email)).scalar_one_or_none()
-    if row is None:
-        row = DemoWhitelist(
-            email=email,
-            note=(body.note or "").strip() or None,
-            added_by=admin.email,
-        )
-        db.add(row)
-        audit.record(
-            db, admin.id, AuditAction.DEMO_ADD, target_type="demo", target_id=email, summary=email
-        )
-        db.commit()
-    return _serialize_demo(row)
-
-
-@router.delete("/demo/whitelist/{entry_id}")
-def remove_demo_whitelist(
-    entry_id: int,
-    admin: User = Depends(admin_user),
-    db: Session = Depends(get_db),
-) -> dict:
-    row = db.get(DemoWhitelist, entry_id)
-    if row is not None:
-        audit.record(
-            db,
-            admin.id,
-            AuditAction.DEMO_REMOVE,
-            target_type="demo",
-            target_id=row.email,
-            summary=row.email,
-        )
-        db.delete(row)
-        db.commit()
-    return {"deleted": entry_id}
-
-
-@router.post("/demo/reset")
-def reset_demo(admin: User = Depends(admin_user), db: Session = Depends(get_db)) -> dict:
-    """Manually clean up the shared demo sandbox back to a baseline. A scheduled job does the
-    same automatically (admin-tunable interval); this is the admin's on-demand button."""
-    demo = get_or_create_demo_user(db)
-    seeded = reset_demo_state(db, demo)
-    audit.record(
-        db, admin.id, AuditAction.DEMO_RESET, summary=f"{seeded} sample playlists re-seeded"
-    )
-    db.commit()
-    return {"ok": True, "reset": True, "playlists_seeded": seeded}
-
-
 @router.post("/purge-discovery")
 def purge_discovery(admin: User = Depends(admin_user), db: Session = Depends(get_db)) -> dict:
     """Reclaim all un-kept discovery content NOW (ignoring the grace period): live-search result

backend/app/routes/admin_logs.py

@@ -1,4 +1,4 @@
-"""The admin diagnostics console's read side (E22.6): query `log_events`, the server's stored log.
+"""The admin diagnostics console's read side: query `log_events`, the server's stored log.
 
 Admin-only and read-only — rows leave the table through the retention job (`log_retention_days`),
 never through here. Reading is not audited: it changes nothing, and the console's live tail polls,

backend/app/routes/audit.py

@@ -66,6 +66,12 @@ def list_audit(
 def clear_audit(admin: User = Depends(admin_user), db: Session = Depends(get_db)) -> dict:
     """Clear the audit log. Leaves ONE row recording who cleared it + how many (tamper-evidence)."""
     n = db.execute(delete(AuditLog)).rowcount or 0
-    audit.record(db, admin.id, AuditAction.AUDIT_CLEAR, summary=f"Cleared {n} audit entries")
+    audit.record(
+        db,
+        admin.id,
+        AuditAction.AUDIT_CLEAR,
+        summary=f"Cleared {n} audit entries",
+        after={"cleared": n},  # the count as data: the log names it in the reader's language
+    )
     db.commit()
     return {"ok": True, "cleared": n}

backend/app/routes/auth.py

@@ -1,6 +1,6 @@
 """Auth HTTP endpoints (OAuth login/callback, register/verify, password, session mgmt).
 
-Split out of app/auth.py (R12 S2); the core auth surface (current_user, scopes, token
+Split out of app/auth.py; the core auth surface (current_user, scopes, token
 helpers) stays in app/auth.py, imported below."""
 
 import secrets
@@ -29,8 +29,8 @@ from app.auth import (
     WRITE_SCOPES,
     current_user,
     env_admin_canonicals,
+    env_allowed_canonicals,
     find_user_by_email,
-    get_or_create_demo_user,
     google_enabled,
     is_allowed,
     log,
@@ -39,7 +39,7 @@ from app.auth import (
 )
 from app.config import settings
 from app.db import SessionLocal, get_db
-from app.models import AuthToken, DemoWhitelist, Invite, OAuthToken, User, display_text
+from app.models import AuthToken, Invite, OAuthToken, User, display_text
 from app.ratelimit import RateLimiter, client_key
 from app.security import encrypt, hash_password, hash_token, verify_password
 from app.utils import canonical_email, valid_email
@@ -53,7 +53,7 @@ VERIFY_TTL = timedelta(hours=24)
 # in and sets no password.
 APPROVAL_VERIFY_TTL = timedelta(days=7)
 RESET_TTL = timedelta(hours=1)
-# An admin-issued set-password link (E12 S5) travels out of band — the admin has to actually reach the
+# An admin-issued set-password link travels out of band — the admin has to actually reach the
 # person and hand it over — so the one-hour reset window is far too short. A day matches VERIFY_TTL and
 # the affordance the admin was shown.
 ADMIN_SETPW_TTL = timedelta(hours=24)
@@ -72,7 +72,6 @@ _reset_limiter = RateLimiter(max_events=5, window_seconds=300)
 # Separate from _reset_limiter on purpose: opening a link must not eat the budget for REQUESTING one.
 _reset_check_limiter = RateLimiter(max_events=10, window_seconds=300)
 _verify_resend_limiter = RateLimiter(max_events=5, window_seconds=300)
-_request_access_limiter = RateLimiter(max_events=5, window_seconds=300)
 _suspend_email_limiter = RateLimiter(max_events=1, window_seconds=3600)
 
 
@@ -92,8 +91,6 @@ def _notify_suspended(background: BackgroundTasks, user: User) -> None:
         )
 
 
-_demo_limiter = RateLimiter(max_events=5, window_seconds=60)
-
 router = APIRouter(prefix="/auth", tags=["auth"])
 
 _oauth = OAuth()
@@ -161,24 +158,12 @@ def notify_admins_of_request(
             background.add_task(email_mod.send_admin_new_request, recipients, lang, requester)
 
 
-def is_demo_allowed(db: Session, email: str) -> bool:
-    """Whether this email may enter the shared demo account (no Google sign-in)."""
-    if not email:
-        return False
-    return (
-        db.execute(
-            select(DemoWhitelist).where(DemoWhitelist.email == email.lower())
-        ).scalar_one_or_none()
-        is not None
-    )
-
-
 class InviteRequest(NamedTuple):
     """What an access request settled on.
 
     `invite` is the row the request now lives under — which is NOT always the address that was just
     typed: one Gmail mailbox spelled two ways settles on whichever row already exists, and that row's
-    address is where mail will go. Callers that tell the user what happened need it (E9.5 S4).
+    address is where mail will go. Callers that tell the user what happened need it.
     `is_new` is the old boolean return in a name: true only when a row was created or a denied one
     reopened, i.e. exactly when admins should hear about it. A repeat request must not re-spam them.
     """
@@ -196,14 +181,14 @@ def upsert_pending_invite(db: Session, email: str, lang: str | None = None) -> I
     not reopen a decision that has already been made for that mailbox. The row is still STORED under
     the address as typed, because that is where mail has to go.
 
-    `lang` is the language of the surface the request came from — the sign-in page, or the locale
-    Google reported for a refused sign-in. The latest one wins: someone who asks again from a
-    Hungarian page is telling us something newer than last time. Anything unsupported is ignored
+    `lang` is the language the request came in — the locale Google reported for a refused sign-in.
+    The latest one wins: someone who asks again under another locale is telling us something newer
+    than last time. Anything unsupported is ignored
     rather than stored, so a forged body cannot put junk in the column.
 
     The commit contract is unchanged: this still commits ONLY when it writes something. Callers rely
     on that — `_register_account` persists its own flushed user precisely because a no-op here
-    commits nothing (E9.5 S6).
+    commits nothing.
     """
     email = (email or "").lower()
     if not valid_email(email):
@@ -215,6 +200,12 @@ def upsert_pending_invite(db: Session, email: str, lang: str | None = None) -> I
         .scalars()
         .all()
     )
+    # An address the env already lets in has nothing to ask for, and a pending row for it would be
+    # worse than nothing: `is_allowed` answers from the invite rows once any exist, so a stranger
+    # registering a sibling spelling (`b.ob@` beside an env-granted `bob@`) would lock the real
+    # owner out of a grant no admin could see.
+    if not rows and canon in env_allowed_canonicals():
+        return InviteRequest(None, False)
     # An approved row wins, so a denied duplicate can't mask an approval. Then a row that is not
     # DENIED, preferring the exact spelling — a denied sibling must never be picked while a pending
     # row exists, or the reopen branch below would fire and put one person in the admin's queue
@@ -254,7 +245,7 @@ def _access_notice_url(kind: str, email: str = "", inv: Invite | None = None) ->
 
     It rides the FRAGMENT, not the query string, for the same reason the verify/reset tokens do
     (SB3): a fragment never reaches the server, a proxy access log or a Referer header — and this
-    notice now carries the user's address. Naming it is the point of E9.5 S4: on 2026-09-03 a
+    notice now carries the user's address. Naming it is the point: on 2026-09-03 a
     registration as `npeter83.temu@` met an OAuth claim of `npeter83temu@`, and a screen that said
     "request recorded" without saying FOR WHAT hid the mismatch for a day.
 
@@ -306,9 +297,10 @@ class EmailPasswordIn(BaseModel):
     password: str | None = None
     # The language the sign-up page was in. Captured at registration because it is the ONLY moment
     # the choice exists anywhere the server can see: it lives in the browser until an account does,
-    # and a preference nobody recorded cannot be backfilled later (the lesson `google_email` taught
-    # in E9.5 S3 — those rows still read NULL until each user signs in again). Nothing reads it yet;
-    # the bilingual mail templates are E9.5 S7.
+    # and a preference nobody recorded cannot be backfilled later (as `google_email` showed: those
+    # rows still read NULL until each user signs in again). It is stored as the account's
+    # `preferences["language"]`, which the UI starts in and every mail to the account is written in
+    # (`mailtext.lang_for_user` / `lang_for_email`).
     lang: str | None = None
 
 
@@ -408,7 +400,7 @@ async def callback(
             return RedirectResponse(url=_access_notice_url("requested", email, req.invite))
         return RedirectResponse(url=_access_notice_url("denied"))
 
-    user = db.query(User).filter(User.google_sub == sub).one_or_none()
+    user = db.execute(select(User).where(User.google_sub == sub)).scalar_one_or_none()
     if user is None:
         # No account carries this Google identity yet. If one already exists for this MAILBOX —
         # e.g. an email+password registration — adopt the Google identity onto it instead of
@@ -421,7 +413,11 @@ async def callback(
         # (ids 13 and 14). Exact spelling still wins when both exist, so an account that already
         # matches literally is never passed over for a canonical sibling.
         canon = canonical_email(email)
-        candidates = db.query(User).filter(User.email_canonical == canon).order_by(User.id).all()
+        candidates = (
+            db.execute(select(User).where(User.email_canonical == canon).order_by(User.id))
+            .scalars()
+            .all()
+        )
         existing = next((u for u in candidates if u.email == email), None) or next(
             iter(candidates), None
         )
@@ -447,9 +443,10 @@ async def callback(
                 return RedirectResponse(url=_access_notice_url("denied"))
             user = existing
             user.google_sub = sub
-            # SEC-10: drop any password pre-set on this account while its email was still unverified,
-            # now that Google attests ownership (bumps the epoch to revoke any session it minted).
-            revoke_preset_password_if_unverified(user)
+            # SEC-10: drop any password nobody proved the mailbox for, now that Google attests
+            # ownership (bumps the epoch to revoke any session it minted). Runs before the
+            # activation below, which is what it reads.
+            revoke_untrusted_preset_password(user)
             # Google verified the email and is_allowed granted access, so finish activating a
             # previously pending password registration (otherwise current_user would bounce it).
             user.email_verified = True
@@ -480,7 +477,16 @@ async def callback(
     # default for a brand-new account (its role attribute is still unset before flush).
     # Canonical, like every other read of the env lists: a dotted or `+tag`ged ADMIN_EMAILS entry
     # would otherwise never bootstrap its admin, because Google answers with the plain primary.
+    #
+    # Becoming admin this way drops any password the account holds, on whichever sign-in it
+    # happens. Nothing here can tell who chose that password: without SMTP registration marks the
+    # address verified with no proof (and SMTP configured later never revisits it), and an admin's
+    # Activate vouches for access, not for the password. So a stranger who registered the address
+    # first, and was let in, would otherwise see their own live session become an admin one. The
+    # real owner, signed in through Google right now, sets a new password in Settings.
     if canonical_email(email) in env_admin_canonicals():
+        if user.role != "admin" and user.password_hash is not None:
+            _revoke_password(user)
         user.role = "admin"
     elif not user.role:
         user.role = "user"
@@ -515,7 +521,7 @@ def _sync_google_email(db: Session, user: User, email: str) -> None:
     of the spelling they chose, because sign-in matches the stored address first. Google returns the
     account's PRIMARY address, which need not be what they registered with — and can never contain a
     `+` at all — so without this guard every plus-tagged registration lost its tag on the first Google
-    sign-in. (E9.5 S3 made that reachable by matching adoption across spellings; the rewrite itself is
+    sign-in. (Matching adoption across spellings made that reachable; the rewrite itself is
     older.)
 
     Never overwrite with an address another account already owns either: that would hit the unique
@@ -525,7 +531,9 @@ def _sync_google_email(db: Session, user: User, email: str) -> None:
         return
     if canonical_email(user.email) == canonical_email(email):
         return  # same mailbox, their spelling — leave it alone
-    clash = db.query(User).filter(User.email == email, User.id != user.id).one_or_none()
+    clash = db.execute(
+        select(User).where(User.email == email, User.id != user.id)
+    ).scalar_one_or_none()
     if clash is None:
         user.email = email
 
@@ -582,9 +590,11 @@ def _complete_link(
     if link_uid != request.session.get("user_id") and link_uid not in accounts:
         return RedirectResponse(url="/?link=error")
     user = db.get(User, link_uid)
-    if user is None or user.is_demo:
+    if user is None:
         return RedirectResponse(url="/?link=error")
-    other = db.query(User).filter(User.google_sub == sub, User.id != user.id).one_or_none()
+    other = db.execute(
+        select(User).where(User.google_sub == sub, User.id != user.id)
+    ).scalar_one_or_none()
     if other is not None:
         return RedirectResponse(url="/?link=conflict")
     if user.google_sub is not None and user.google_sub != sub:
@@ -628,73 +638,12 @@ async def logout(request: Request):
     return JSONResponse({"ok": True, "switched": False})
 
 
-class AccessRequestIn(EmailIn):
-    # The language of the page this request was made from. Same reasoning as registration's `lang`
-    # (and the same shape): it is the ONLY moment this preference exists anywhere the server can
-    # see, since the requester has no account to carry it, and it decides what language the approval
-    # or denial mail is written in.
-    lang: str | None = None
-
-
-@router.post("/request-access")
-async def request_access(
-    body: AccessRequestIn,
-    request: Request,
-    background: BackgroundTasks,
-    db: Session = Depends(get_db),
-) -> dict:
-    """Public: ask for access. Idempotent — repeat requests for the same email don't
-    duplicate or re-spam admins (upsert answers `is_new=False` for an already-pending row).
-
-    Answers a uniform ``"pending"`` for EVERY outcome — already-allowed, newly-pending,
-    already-pending, and rate-limited alike. The old ``"approved"`` reply for an allow-listed
-    address made this a public enumeration oracle for the allow-list (which the rest of the module
-    is careful to avoid, cf. demo_login); an already-allowed user can still just sign in with Google.
-    Rate-limited per client IP like register/login/reset, and a block returns "pending" too so the
-    endpoint can't be probed for throttling either."""
-    email = (body.email or "").strip().lower()
-    if not valid_email(email):
-        raise refused(400, "invalid_email")
-    if not _request_access_limiter.allow(client_key(request)):
-        return {"ok": True, "status": "pending"}
-    if is_allowed(db, email):
-        return {"ok": True, "status": "pending"}  # already allowed — don't reveal it
-    req = upsert_pending_invite(db, email, body.lang)
-    if req.is_new:
-        notify_admins_of_request(db, background, email)
-    return {"ok": True, "status": "pending"}
-
-
-@router.post("/demo")
-def demo_login(body: EmailIn, request: Request, db: Session = Depends(get_db)) -> dict:
-    """Hidden demo entry: a whitelisted email logs straight into the shared demo account,
-    no Google OAuth. The login page fires this quietly (debounced) as the user types/pastes,
-    so there is no visible 'demo login' button. Rate-limited per IP; when blocked it answers
-    exactly like a non-match (authenticated=false) so the field can't be hammered as an
-    enumeration oracle. On a match it sets the session and the client reloads into the app."""
-    if not _demo_limiter.allow(client_key(request)):
-        return {"ok": True, "authenticated": False}
-    email = (body.email or "").strip().lower()
-    if valid_email(email) and is_demo_allowed(db, email):
-        demo = get_or_create_demo_user(db)
-        # Isolate the demo: drop any real-account wallet on this tab first, so the demo session
-        # can't switch to / act as a previously signed-in account via the multi-account header.
-        request.session.clear()
-        request.session["user_id"] = demo.id
-        _remember_epoch(request, demo)
-        # The demo key IS an address, so it is not in the line — see the denied branch above.
-        log.info("signed in", extra=logs.event("auth.login.ok", method="demo", uid=demo.id))
-        return {"ok": True, "authenticated": True}
-    return {"ok": True, "authenticated": False}
-
-
 def _establish_session(request: Request, user: User) -> None:
     """Sign `user` in for this browser session — the ONE sign-in path, shared by the Google callback
     and password login, so both join the same account switcher and both stamp `last_login_at` (the
     caller commits). Multi-session: every account that authenticated in this browser is remembered so
     the user can switch between them; only accounts that actually proved themselves HERE land in that
-    wallet (it is the proof they may be switched to). The demo entry deliberately does not come
-    through here: it clears the wallet instead, and its shared account has no meaningful last login."""
+    wallet (it is the proof they may be switched to)."""
     accounts = [a for a in (request.session.get("account_ids") or []) if a != user.id]
     accounts.append(user.id)
     request.session["account_ids"] = accounts[-MAX_SESSION_ACCOUNTS:]
@@ -784,7 +733,7 @@ def _consume_password_token(db: Session, raw: str | None) -> tuple[User, str] |
 
 
 def issue_set_password_link(db: Session, user: User) -> tuple[str, datetime]:
-    """An admin-issued, single-use link that lets `user` set a password without any email (E12 S5).
+    """An admin-issued, single-use link that lets `user` set a password without any email.
 
     One mechanism answers three things: SEC-15 (on a no-SMTP instance every registration is
     auto-verified, so a pre-empted address cannot be recovered by mail), SEC-16 (a pre-empted address
@@ -866,10 +815,10 @@ def _register_account(email: str, password: str, lang: str | None = None) -> Non
             # for the same input — so it is not a new capability, just the end of a dead end. An
             # account that is already confirmed still gets nothing: there is nothing to confirm.
             #
-            # And a re-send is ALL it does: the submitted password is discarded. E9.5's closing
+            # And a re-send is ALL it does: the submitted password is discarded. A closing
             # review tried "while the address is unverified, the newest registration owns it"
             # (replace the hash, kill the outstanding verify tokens) as a fix for address
-            # pre-emption, and round 2 showed it was worse than the problem — DO NOT REPEAT IT.
+            # pre-emption, and it proved worse than the problem — DO NOT REPEAT IT.
             # The rule is symmetric, and an attacker can always be last: it hands an unauthenticated
             # caller the power to overwrite a real registrant's chosen password and delete the
             # confirmation link already in their inbox, and on a SEC-10-stripped row (password_hash
@@ -878,7 +827,7 @@ def _register_account(email: str, password: str, lang: str | None = None) -> Non
             # primitive: password-reset accepts this account, proves the mailbox, rewrites the hash
             # and bumps the session epoch. The residual is logged as SEC-16.
             user = existing
-            if not user.email_verified and not user.is_demo and email_mod.email_enabled():
+            if not user.email_verified and email_mod.email_enabled():
                 send_verification_link(db, user)
             return
         # Without working SMTP we can't deliver a verification link, so email ownership can't gate
@@ -918,7 +867,7 @@ def _register_account(email: str, password: str, lang: str | None = None) -> Non
         db.commit()
         if email_ok:
             send_verification_link(db, user)
-        # Fired unconditionally, unlike the other two callers which gate on a fresh invite row: for
+        # Fired unconditionally, unlike the Google callback, which gates on a fresh invite row: for
         # an address the admin already approved, nothing lands in the Access-requests queue, so this
         # is the only place the account's existence is announced at all. Delivery is a separate
         # matter — with no working SMTP `_send` drops it, which is why the pre-approved registrant
@@ -928,7 +877,7 @@ def _register_account(email: str, password: str, lang: str | None = None) -> Non
 
 def issue_verification_link(db: Session, user: User, ttl: timedelta | None = None) -> str:
     """A fresh confirmation link for `user`, returned rather than sent — for a mail that carries the
-    link inside a larger message (the branching approval/invitation mails, E9.5 S6).
+    link inside a larger message (the branching approval/invitation mails).
 
     Same token machinery and the same fragment (SB3) as the standalone confirmation mail, so the two
     kinds of link cannot drift apart in shape or in how they are consumed."""
@@ -943,8 +892,8 @@ def send_verification_link(db: Session, user: User) -> bool:
     address is the one its owner chose to be reachable at, and a `+tag` there is usually deliberate.
     The token rides the URL FRAGMENT (#) so it can never reach a proxy/access log or a Referer (SB3).
 
-    One definition, three callers — registration, the self-service resend, and the admin-side resend
-    (E9.5 S5). The link's shape and TTL must not drift between them; a verification mail that differs
+    One definition, three callers — registration, the self-service resend, and the admin-side resend.
+    The link's shape and TTL must not drift between them; a verification mail that differs
     by who triggered it is a bug waiting to be discovered by exactly one user.
 
     Returns whether the mail actually went out, so a caller with a human waiting on the answer (the
@@ -967,8 +916,8 @@ def verify_email_confirm(body: TokenIn, db: Session = Depends(get_db)) -> dict:
     dropped the password this account registered with, so sign-in would answer with the deliberately
     uniform "Invalid email or password" — true, but unreadable to the person who just proved they own
     the mailbox. Saying so leaks nothing: holding a valid single-use verify token for this account
-    already proves that much. Since E9.5 S1 the common source of that state is gone (admin approval no
-    longer revokes), leaving accounts stripped before that change and the Google-adoption path, so this
+    already proves that much. The common source of that state is gone (admin approval no longer
+    revokes), leaving accounts stripped before that change and the Google-adoption path, so this
     flag is rare now — but it must stay, because those users have no other way back in."""
     user = _consume_token(db, body.token, "verify")
     if user is None:
@@ -1012,7 +961,7 @@ def _resend_verify_if_eligible(email: str) -> None:
         return
     with SessionLocal() as db:
         user = find_user_by_email(db, email)
-        if user is not None and not user.email_verified and not user.is_demo:
+        if user is not None and not user.email_verified:
             send_verification_link(db, user)
 
 
@@ -1028,7 +977,7 @@ def _login_refused(status: int, reason: str, user: User | None = None, **extra)
     that proof (`invalid`, `rate_limited`) carry nothing at all, so they stay byte-identical for a
     real account and an unknown one.
 
-    Since E9.5 S1 the two gates are order-independent, so `approved` really can be true while
+    The two gates are order-independent, so `approved` really can be true while
     `email_verified` is false, and vice versa — the SPA renders whichever is actually outstanding
     rather than assuming a sequence.
     """
@@ -1058,11 +1007,11 @@ def password_login(
     email = (body.email or "").strip().lower()
     password = body.password or ""
     user = find_user_by_email(db, email)
-    # Always run the KDF (against a decoy hash for an unknown / passwordless / demo account) so the
+    # Always run the KDF (against a decoy hash for an unknown / passwordless account) so the
     # response time is the same whether or not the email exists — no timing-based enumeration.
-    candidate_hash = user.password_hash if (user and not user.is_demo) else None
+    candidate_hash = user.password_hash if user else None
     password_ok = verify_password(password, candidate_hash or _DECOY_PASSWORD_HASH)
-    if user is None or user.is_demo or not password_ok:
+    if user is None or not password_ok:
         raise _login_refused(401, "invalid")
     if user.is_suspended:
         _notify_suspended(background, user)
@@ -1101,7 +1050,7 @@ def password_reset_request(
 
 def _send_reset_if_eligible(email: str) -> None:
     """Background worker for password_reset_request: issue a reset token + email the link, but ONLY
-    for a real (non-demo) password account. Runs after the response with its own DB session, so the
+    for a password account. Runs after the response with its own DB session, so the
     request's timing never reveals whether the account exists. Token rides the URL FRAGMENT (#) so it
     can't leak into proxy/access logs or a Referer (SB3).
 
@@ -1112,7 +1061,7 @@ def _send_reset_if_eligible(email: str) -> None:
     proves identity through the session rather than by mail."""
     with SessionLocal() as db:
         user = find_user_by_email(db, email)
-        if user is None or user.is_demo:
+        if user is None:
             return
         if user.password_hash or user.password_revoked_at:
             raw = _issue_token(db, user, "reset", RESET_TTL)
@@ -1200,37 +1149,49 @@ def bump_session_epoch(user: User) -> None:
     user.session_epoch = (user.session_epoch or 0) + 1
 
 
-def revoke_preset_password_if_unverified(user: User) -> None:
+def revoke_untrusted_preset_password(user: User) -> None:
     """SEC-10 (account pre-emption): a password set on an account whose email was NEVER verified is
     untrusted — whoever registered the address (to pre-empt the real owner) could have chosen it. When
     the real owner attests ownership through Google, drop that password and bump the epoch to revoke any
-    session it minted; the owner re-sets one via password-reset. No-op for an already-verified account,
-    so a legit no-SMTP registration is unaffected.
+    session it minted; the owner re-sets one via password-reset, or from Settings while the Google
+    session lasts.
 
-    The Google-adoption callback is the ONLY caller (E9.5 S1). Admin approval used to call this too, and
-    dropping it was right for the reason S1 gave — the revoke destroyed the password a legitimate
+    A NEVER-ACTIVATED account is untrusted too, verified or not: without SMTP, registration marks the
+    address verified with no proof at all. For an ADMIN_EMAILS / ALLOWED_EMAILS address no invite row
+    gates the registration, so a stranger who registered it first would otherwise be activated here
+    with their own password still working. (An account the callback makes ADMIN loses its password
+    whatever its state — see the ADMIN_EMAILS block there.)
+
+    The Google-adoption callback is the ONLY caller. Admin approval used to call this too, and
+    dropping it was right for the reason given — the revoke destroyed the password a legitimate
     registrant had just chosen, and since the admin normally decides before the user reads their mail,
     that made the "your password was cleared" detour the DEFAULT path rather than an edge case.
 
-    S1 also claimed the revoke was REDUNDANT there, because `password_login` refuses an unverified
+    The revoke was also called REDUNDANT there, because `password_login` refuses an unverified
     account before it reads `is_active`. That part was wrong, and the epic's own review caught it: the
-    gate holds only while the address is unverified, and S6's approval mail exists to get the real
+    gate holds only while the address is unverified, and the approval mail exists to get the real
     owner to end that state — at which point a pre-emptor's password works. Nothing replaced the
     revoke, and that is deliberate. The review tried "the newest registration owns an unverified
     account" and reverted it: the rule is symmetric, so an attacker can simply register last, and it
     would have handed anonymous callers a write onto a real account's credential. What the owner
     actually has is password-reset, which proves the mailbox, rewrites the hash and bumps the epoch.
     The residual is logged as SEC-16; re-adding a caller here would put "your password was cleared"
-    back in front of every legitimate registrant, which is the whole reason S1 removed it.
+    back in front of every legitimate registrant, which is the whole reason it was removed.
 
     `password_revoked_at` is what makes "the owner re-sets one via password-reset" actually true. The
     revoke leaves `password_hash IS NULL`, which is also what a Google-only account looks like — and
     _send_reset_if_eligible keyed off exactly that, so a revoked account silently got no reset link and
     was locked out for good (it could not sign in either). The stamp separates the two cases."""
-    if not user.email_verified:
-        user.password_hash = None
-        user.password_revoked_at = datetime.now(timezone.utc)
-        bump_session_epoch(user)
+    if not user.email_verified or (not user.is_active and user.password_hash is not None):
+        _revoke_password(user)
+
+
+def _revoke_password(user: User) -> None:
+    """Drop the account's password, stamp why (so password-reset still accepts it) and end every
+    session it signed in."""
+    user.password_hash = None
+    user.password_revoked_at = datetime.now(timezone.utc)
+    bump_session_epoch(user)
 
 
 @router.post("/logout-others")
@@ -1239,9 +1200,7 @@ def logout_others(
 ) -> dict:
     """SA4: sign every OTHER session for this account out — bump the epoch so all existing cookies
     are rejected, then re-stamp THIS cookie so the caller stays signed in. For "I left myself logged
-    in somewhere" / a suspected session compromise. The demo account (shared) is excluded."""
-    if user.is_demo:
-        raise HTTPException(status_code=403, detail={"code": "demo"})
+    in somewhere" / a suspected session compromise."""
     # Commit the bump BEFORE re-stamping this cookie, so a failed commit can't strand the current
     # session at a newer epoch than the DB (see set_password).
     bump_session_epoch(user)
@@ -1260,8 +1219,6 @@ async def link_google(
     """Start linking a Google account to the signed-in (e.g. password) account. Requests only
     the identity scopes — YouTube access is granted separately via /auth/upgrade. The callback
     sees `oauth_link_uid` and attaches the identity instead of creating a new account."""
-    if user.is_demo:
-        return RedirectResponse(url="/")
     client = google_oauth(db)
     if client is None:
         return RedirectResponse(url="/?link=error")
@@ -1287,8 +1244,6 @@ def set_password(
     """Set or change the signed-in account's password. Setting a first password (Google-only
     account) needs no current password — the session already proves identity. Changing an
     existing one requires the current password, so a hijacked session can't silently rotate it."""
-    if user.is_demo:
-        raise HTTPException(status_code=403, detail={"code": "demo"})
     new_password = body.password or ""
     if len(new_password) < PASSWORD_MIN_LEN:
         raise refused(400, "weak_password", min=PASSWORD_MIN_LEN)
@@ -1319,10 +1274,6 @@ async def upgrade(
     (unsubscribe). The shared callback stores whatever Google returns, so `can_read` /
     `can_write` flip on once granted. Anything other than an explicit `write` defaults to
     the least-privileged read scope."""
-    # This is the browser-facing redirect entry point, so the shared demo account is bounced
-    # straight home (no YouTube link is ever attached to it) rather than shown a raw 403.
-    if user.is_demo:
-        return RedirectResponse(url="/")
     client = google_oauth(db)
     if client is None:
         return RedirectResponse(url="/")

backend/app/routes/channels.py

@@ -6,11 +6,13 @@ from datetime import datetime, timezone
 
 from fastapi import APIRouter, BackgroundTasks, Depends
 from pydantic import BaseModel
-from sqlalchemy import func, select, update
+from sqlalchemy import Select, func, select, update
 from sqlalchemy.orm import Session
 
+from app.collation import collated, sorted_words
+from app.mailtext import lang_for_user
 from app import logs, quota
-from app.auth import admin_user, current_user, require_human, require_write_scope
+from app.auth import admin_user, current_user, require_write_scope
 from app.db import get_db
 from app.models import (
     LIVE_OR_UPCOMING,
@@ -38,33 +40,40 @@ router = APIRouter(prefix="/api/channels", tags=["channels"])
 log = logs.get_logger(__name__)
 
 
+def channel_aggregates(channel_ids: list[str]) -> Select:
+    """Per-channel stored count, last upload and total duration, in one grouped pass. Every column it
+    reads is in `ix_videos_channel_pub_cover` (migration 0095), so Postgres answers it from the index
+    alone; a test holds that (test_channel_list_index_only)."""
+    return (
+        select(
+            Video.channel_id,
+            func.count(Video.id),
+            func.max(Video.published_at),
+            func.coalesce(func.sum(Video.duration_seconds), 0),
+        )
+        .where(Video.channel_id.in_(channel_ids))
+        .group_by(Video.channel_id)
+    )
+
+
 @router.get("")
 def list_channels(user: User = Depends(current_user), db: Session = Depends(get_db)) -> list[dict]:
     rows = db.execute(
         select(Subscription, Channel)
         .join(Channel, Channel.id == Subscription.channel_id)
         .where(Subscription.user_id == user.id)
-        .order_by(Subscription.priority.desc(), func.lower(Channel.title))
+        .order_by(Subscription.priority.desc(), collated(Channel.title, lang_for_user(user)))
     ).all()
 
-    # Per-channel aggregates over the shared catalog in ONE grouped pass: stored count, last
-    # upload, total duration. Cheap — a single grouped scan, so computed on read rather than
-    # denormalised (measure-first: reads aren't frequent/hot enough to warrant cached columns).
-    # (The Normal/Shorts/Live breakdown moved to the channel-detail page in R14, so it is no longer
+    # Per-channel aggregates over the shared catalog (channel_aggregates), computed on read rather than
+    # denormalised: an Index Only Scan over a ~20 MB index, where it used to read every video's heap page
+    # (681 MB on prod, a 2–6 s first open once something else had pushed it out of memory).
+    # (The Normal/Shorts/Live breakdown moved to the channel-detail page, so it is no longer
     # computed here — the manager table dropped its N/S/L column.)
     channel_ids = [c.id for _, c in rows]
     agg: dict[str, dict] = {}
     if channel_ids:
-        for cid, n, last_up, total_dur in db.execute(
-            select(
-                Video.channel_id,
-                func.count(Video.id),
-                func.max(Video.published_at),
-                func.coalesce(func.sum(Video.duration_seconds), 0),
-            )
-            .where(Video.channel_id.in_(channel_ids))
-            .group_by(Video.channel_id)
-        ).all():
+        for cid, n, last_up, total_dur in db.execute(channel_aggregates(channel_ids)).all():
             agg[cid] = {
                 "stored": n,
                 "last_video_at": iso(last_up),
@@ -139,7 +148,7 @@ def _discovery_rows(db: Session, user: User) -> list:
     return db.execute(
         q.group_by(Channel.id).order_by(
             func.count(func.distinct(PlaylistItem.video_id)).desc(),
-            func.lower(Channel.title),
+            collated(Channel.title, lang_for_user(user)),
         )
     ).all()
 
@@ -154,8 +163,8 @@ def discover_channels(
     can judge a channel BEFORE subscribing — but we do NOT pull its videos. Most-present
     channels come first; subscribing is a separate action."""
     # Resolve & cache the user's own channel id so it can be excluded — you don't subscribe
-    # to yourself. Lazy + cached; needs OAuth, so demo / token-less users just skip it.
-    if user.yt_channel_id is None and user.token is not None and not user.is_demo:
+    # to yourself. Lazy + cached; needs OAuth, so token-less users just skip it.
+    if user.yt_channel_id is None and user.token is not None:
         try:
             with (
                 quota.attribute(user.id, quota.QuotaAction.CHANNELS_DISCOVER),
@@ -183,8 +192,7 @@ def discover_channels(
     # (fresh subscriber counts/thumbnails carry through without a re-query). Only the title can
     # change, which affects the tie-break order → re-sort in Python instead of re-hitting the DB.
     need = [ch.id for ch, _, _ in rows if ch.details_synced_at is None][:DISCOVERY_ENRICH_CAP]
-    # Skip the quota-spending enrich for the demo account (it just sees the stub metadata).
-    if need and not user.is_demo:
+    if need:
         try:
             with (
                 quota.attribute(user.id, quota.QuotaAction.CHANNELS_DISCOVER),
@@ -192,10 +200,14 @@ def discover_channels(
             ):
                 apply_channel_details(db, yt.get_channels(need), need)
             db.commit()
-            # Match _discovery_rows' ORDER BY exactly: video-count DESC, then title (NULLs LAST,
-            # as Postgres orders them — coercing NULL→"" would wrongly float untitled channels up).
+            # Match _discovery_rows' ORDER BY exactly: video-count DESC, then title in the reader's
+            # alphabet (NULLs LAST, as Postgres orders them — coercing NULL→"" would wrongly float
+            # untitled channels up).
+            titles = [r[0].title for r in rows if r[0].title is not None]
+            alpha = {w: i for i, w in enumerate(sorted_words(db, titles, lang_for_user(user)))}
             rows = sorted(
-                rows, key=lambda r: (-int(r[1]), r[0].title is None, (r[0].title or "").lower())
+                rows,
+                key=lambda r: (-int(r[1]), r[0].title is None, alpha.get(r[0].title or "", 0)),
             )
         except YouTubeError as exc:
             log.warning(
@@ -264,7 +276,7 @@ def _channel_detail_dict(
         "explored": explored,
         "blocked": blocked,
         "stored_videos": stored,
-        # Per-type breakdown (Normal / Shorts / Live) — R14 moved it off the manager table onto the
+        # Per-type breakdown (Normal / Shorts / Live) — moved off the manager table onto the
         # channel page. Same aggregation the manager list uses (count_normal = stored − shorts − live).
         "count_normal": count_normal,
         "count_short": count_short,
@@ -298,12 +310,12 @@ def channel_detail(
     Lazily fills the About metadata (view count, join date, banner, links) the first time it's
     needed (a stub) and refreshes it once it is stale (`DETAILS_MAX_AGE`), which is how a channel nobody subscribes to (an explored one, outside the
     subscription import) gets a banner that still resolves. The enrich uses the API key (1 unit,
-    cached after), so it's skipped for the quota-less demo account."""
+    cached after)."""
     channel = get_or_404(db, Channel, channel_id, what="channel")
     # No `published_at is None` sentinel: a terminated channel never gets one, so it re-fetched on
     # every view; rows enriched before that field existed are long past DETAILS_MAX_AGE anyway.
     stale = details_stale(channel, datetime.now(timezone.utc))
-    if stale and not user.is_demo:
+    if stale:
         try:
             with (
                 quota.attribute(user.id, quota.QuotaAction.CHANNELS_DISCOVER),
@@ -337,7 +349,7 @@ def channel_detail(
     )
     blocked = _is_blocked(db, user, channel_id)
     # Stored total + the Normal/Shorts/Live split in one grouped pass (same filters as the manager
-    # list), so the channel page can show the per-type breakdown R14 moved off the table.
+    # list), so the channel page can show the per-type breakdown that moved off the table.
     n_all, n_short, n_live = db.execute(
         select(
             func.count(Video.id),
@@ -376,7 +388,7 @@ def channel_detail(
 def explore_channel(
     channel_id: str,
     page_token: str | None = None,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     """Open an un-subscribed channel for browsing. Records the exploration (a per-user,
@@ -384,7 +396,7 @@ def explore_channel(
     drives the later cleanup), flags the channel ephemeral unless someone organically follows
     it, then ingests one page of its uploads (newest-first) marked via_explore + enriched.
     Returns the next page token so the page can "load more". Guarded against the shared
-    backfill quota reserve; blocked for the demo account (it spends shared quota)."""
+    backfill quota reserve."""
     channel = get_or_404(db, Channel, channel_id, what="channel")
     if _is_blocked(db, user, channel_id):
         raise refused(403, "channel_blocked")
@@ -423,7 +435,7 @@ def explore_channel(
                 channel = db.get(Channel, channel_id)
             result = explore_ingest_page(db, yt, channel, page_token)
     except YouTubeError as exc:
-        raise youtube_refusal(exc)
+        raise youtube_refusal(exc) from exc
     return {
         "ok": True,
         "channel_id": channel_id,
@@ -480,10 +492,7 @@ def update_channel(
         sub.priority = body.priority
     if body.hidden is not None:
         sub.hidden = body.hidden
-    # The demo account is SHARED, so it must not persist deep_requested (a quota-relevant flag) onto
-    # the shared subscription — skip the write entirely for demo, which also keeps deep_turned_on
-    # False so the immediate backfill below can never fire for it.
-    if body.deep_requested is not None and not user.is_demo:
+    if body.deep_requested is not None:
         # Opt this channel into (or out of) full-history backfill. The deep scheduler
         # picks the flag up on its next run; turning it off won't undo already-fetched
         # videos, it just stops further deep paging if nobody else still wants it.
@@ -494,7 +503,7 @@ def update_channel(
 
     # Opting a channel into full history should give an immediate feed: if we haven't even
     # fetched its recent uploads yet, do that one channel now (cheap) instead of waiting for
-    # the scheduler. (deep_turned_on is only ever True for a non-demo user — see the guard above.)
+    # the scheduler.
     if deep_turned_on and channel is not None and channel.recent_synced_at is None:
         with quota.attribute(user.id, quota.QuotaAction.VIDEOS_BACKFILL_RECENT):
             run_recent_backfill(db, [channel], max_channels=1)
@@ -584,7 +593,7 @@ def subscribe(
             if "subscriptionDuplicate" in exc.reason.split(","):
                 yt_sub_id = ""
             else:
-                raise youtube_refusal(exc)
+                raise youtube_refusal(exc) from exc
         # Enrich a stub channel (title/thumbnail/subscriber count) so it shows up properly right
         # away, on BOTH the normal and the desync path — a resurfaced stub still needs metadata to
         # render. Keep the client open (we're still inside the `with`); a metadata failure here
@@ -645,7 +654,7 @@ def unsubscribe(
     except YouTubeError as exc:
         # 422 (a real, explainable error → speaking modal), not 502 which the client treats as
         # a transient "connection lost" gateway blip.
-        raise youtube_refusal(exc)
+        raise youtube_refusal(exc) from exc
     db.delete(sub)
     db.commit()
     return {"ok": True, "unsubscribed": channel_id}
@@ -659,7 +668,7 @@ def list_blocked(user: User = Depends(current_user), db: Session = Depends(get_d
             select(Channel)
             .join(BlockedChannel, BlockedChannel.channel_id == Channel.id)
             .where(BlockedChannel.user_id == user.id)
-            .order_by(func.lower(Channel.title))
+            .order_by(collated(Channel.title, lang_for_user(user)))
         )
         .scalars()
         .all()

backend/app/routes/config.py

@@ -139,7 +139,7 @@ def test_issue_tracker(
     _: User = Depends(admin_user),
     db: Session = Depends(get_db),
 ) -> dict:
-    """Check the configured issue tracker (E16): the repository answers with this token, takes
+    """Check the configured issue tracker: the repository answers with this token, takes
     issues, and is PRIVATE — the one it must be, since an issue carries its reporter's log."""
     try:
         return {"ok": True, **gitea.check(db)}

backend/app/routes/debug_sessions.py

@@ -1,9 +1,8 @@
-"""Debug sessions (E16): Start → reproduce → Stop, the screenshots, and the evidence under review.
+"""Debug sessions: Start → reproduce → Stop, the screenshots, and the evidence under review.
 
 The session itself — what tags a request, its cap, its level — is `app/debugsession.py`. These routes
 are the user's handle on their OWN sessions only: another user's id 404s (`owned_or_404`), including
-for an admin, who reads everyone's lines in the diagnostics console instead. The shared demo account
-cannot record: its log would interleave every visitor under one account.
+for an admin, who reads everyone's lines in the diagnostics console instead.
 
 Screenshots arrive as a raw image body (`image/png`, `image/jpeg`, `image/webp`), not a multipart
 form — one image per request, capped while it arrives, and kept only if its bytes really are one of
@@ -21,7 +20,7 @@ from sqlalchemy.orm import Session
 
 from app import audit, debugsession, logs, logsink
 from app.audit import AuditAction
-from app.auth import require_human
+from app.auth import current_user
 from app.db import get_db, release_read_connection
 from app.issues import bundle, gitea
 from app.ratelimit import RateLimiter
@@ -51,7 +50,7 @@ _new_sessions = RateLimiter(max_events=20, window_seconds=3600)
 @router.post("")
 def start_session(
     body: StartIn,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     """Start recording, or join the recording already running (another tab started it)."""
@@ -79,7 +78,7 @@ def start_session(
 
 @router.post("/{sid}/stop")
 def stop_session(
-    sid: str, user: User = Depends(require_human), db: Session = Depends(get_db)
+    sid: str, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     session = owned_or_404(db, DebugSession, sid, user, what="debug_session")
     if session.ended_at is None:
@@ -93,7 +92,7 @@ def stop_session(
 
 
 @router.get("")
-def list_sessions(user: User = Depends(require_human), db: Session = Depends(get_db)) -> dict:
+def list_sessions(user: User = Depends(current_user), db: Session = Depends(get_db)) -> dict:
     sessions = db.scalars(
         select(DebugSession)
         .where(DebugSession.user_id == user.id)
@@ -105,14 +104,14 @@ def list_sessions(user: User = Depends(require_human), db: Session = Depends(get
 
 @router.get("/{sid}")
 def get_session(
-    sid: str, user: User = Depends(require_human), db: Session = Depends(get_db)
+    sid: str, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     return _view(db, owned_or_404(db, DebugSession, sid, user, what="debug_session"))
 
 
 @router.get("/{sid}/events")
 def session_events(
-    sid: str, user: User = Depends(require_human), db: Session = Depends(get_db)
+    sid: str, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     """The rows an issue from this session would carry — browser and server, in time order."""
     session = owned_or_404(db, DebugSession, sid, user, what="debug_session")
@@ -141,7 +140,7 @@ class DraftIn(BaseModel):
 def save_draft(
     sid: str,
     body: DraftIn,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     session = owned_or_404(db, DebugSession, sid, user, what="debug_session")
@@ -162,7 +161,7 @@ def save_draft(
 
 @router.get("/{sid}/export")
 def export_session(
-    sid: str, user: User = Depends(require_human), db: Session = Depends(get_db)
+    sid: str, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> Response:
     """The issue as a zip — the text, the two logs and the screenshots — for an instance without an
     issue tracker, or for sending it by hand. Built from the SAVED draft: what the composer shows is
@@ -182,7 +181,7 @@ def export_session(
     )
 
 
-# --- the issue tracker (S4) -----------------------------------------------------------------
+# --- the issue tracker ----------------------------------------------------------------------
 # Issues a user may open per hour — counted once the tracker is ASKED to open one (an answer that
 # goes missing may still have created it, so an attempt cannot be given back); finishing a half-sent
 # one, or an attempt refused before the tracker was asked, is not counted. The composer sends one at
@@ -208,9 +207,7 @@ def _issues_rate_limited() -> HTTPException:
 
 
 @router.post("/{sid}/issue")
-def send_issue(
-    sid: str, user: User = Depends(require_human), db: Session = Depends(get_db)
-) -> dict:
+def send_issue(sid: str, user: User = Depends(current_user), db: Session = Depends(get_db)) -> dict:
     """Send the session's issue to the tracker — from the SAVED draft, as the export is built — or
     finish a send that stopped half-way (the issue exists; only what is missing is sent)."""
     session = owned_or_404(db, DebugSession, sid, user, what="debug_session")
@@ -302,7 +299,7 @@ class PreviewIn(BaseModel):
 def preview_issue(
     sid: str,
     body: PreviewIn,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     """The body as the tracker will render it. Gitea sanitises its own output; the composer still
@@ -323,14 +320,14 @@ def _shot_view(shot: DebugScreenshot) -> dict:
 
 @router.get("/{sid}/screenshots")
 def list_screenshots(
-    sid: str, user: User = Depends(require_human), db: Session = Depends(get_db)
+    sid: str, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     session = owned_or_404(db, DebugSession, sid, user, what="debug_session")
     return {"items": [_shot_view(s) for s in debugsession.screenshots(db, session)]}
 
 
 def _upload_target(
-    sid: str, user: User = Depends(require_human), db: Session = Depends(get_db)
+    sid: str, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> DebugSession:
     return owned_or_404(db, DebugSession, sid, user, what="debug_session")
 
@@ -393,7 +390,7 @@ def _screenshot_of(db: Session, session: DebugSession, shot_id: int) -> DebugScr
 
 @router.get("/{sid}/screenshots/{shot_id}")
 def get_screenshot(
-    sid: str, shot_id: int, user: User = Depends(require_human), db: Session = Depends(get_db)
+    sid: str, shot_id: int, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> Response:
     shot = _session_screenshot(db, sid, shot_id, user)
     # `nosniff`: the type was checked against the bytes at upload, and the browser must not guess.
@@ -406,7 +403,7 @@ def get_screenshot(
 
 @router.delete("/{sid}/screenshots/{shot_id}")
 def delete_screenshot(
-    sid: str, shot_id: int, user: User = Depends(require_human), db: Session = Depends(get_db)
+    sid: str, shot_id: int, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     session = owned_or_404(db, DebugSession, sid, user, what="debug_session")
     shot = _screenshot_of(db, session, shot_id)

backend/app/routes/diagnostics.py

@@ -1,4 +1,4 @@
-"""The browser's log, sent while its user has a raised diagnostics level (E22 S5).
+"""The browser's log, sent while its user has a raised diagnostics level.
 
 The client keeps its log in memory (frontend `lib/log.ts`). Only after the user consented to Részletes
 or Teljes in Settings → Diagnosztika does it stream that log here, in batches, and each batch lands in
@@ -14,7 +14,7 @@ Nothing from the body is trusted as-is:
     modified client would not;
   * an event below the consented level is dropped (a Részletes consent does not store what only
     Teljes may carry), and so is the whole batch once the level has lapsed — where an open debug
-    session the request names counts as consent to ITS level (E16), and only its own events keep
+    session the request names counts as consent to ITS level, and only its own events keep
     its tag;
   * the timestamp is clamped to the consent window, so a row cannot dodge the retention job by
     claiming to come from the future.
@@ -63,7 +63,7 @@ class ClientEvent(BaseModel):
     attrs: dict[str, Any] | None = None
     request_id: str | None = Field(default=None, pattern=r"^[A-Za-z0-9._-]{8,64}$")
     tab_id: str | None = Field(default=None, pattern=r"^[A-Za-z0-9_-]{1,32}$")
-    # The debug session the event was captured in (E16); kept only when it is the one this request
+    # The debug session the event was captured in; kept only when it is the one this request
     # names in `X-Debug-Session` and the auth dependency found it the caller's own and open.
     debug_session_id: str | None = Field(default=None, pattern=r"^[0-9a-f]{16}$")
     version: str | None = Field(default=None, max_length=64)
@@ -138,7 +138,7 @@ def ingest_events(
     body: ClientEventsIn = Depends(_parsed_events),
     db: Session = Depends(get_db),
 ) -> dict:
-    # A debug session the request names raises the level for its own events (E16): starting one is
+    # A debug session the request names raises the level for its own events: starting one is
     # the consent to that level for its duration, whatever the user's own level is.
     session = debugsession.from_header(db, user, request.headers.get(debugsession.HEADER))
     level, _ = debugsession.level_for(user, session)

backend/app/routes/discover.py

@@ -1,10 +1,10 @@
-"""E24 · the Home surface's HTTP shell.
+"""The Home surface's HTTP shell.
 
 Thin by construction: every decision — what counts as in progress, what fits a time budget, which
 tier the hero comes from — lives in `app/discover/`, so the scheduler, a future CLI or a test can ask
 the same questions without a request. This file validates, gates and serializes.
 
-The consolidated `GET /api/discover` mirrors the Plex Discover route's one-payload shape (E5 S4): a
+The consolidated `GET /api/discover` mirrors the Plex Discover route's one-payload shape: a
 landing page that fires four requests to answer one question would spend its whole budget on the
 thing it exists to remove.
 """
@@ -17,12 +17,13 @@ from sqlalchemy import select
 from sqlalchemy.exc import IntegrityError
 from sqlalchemy.orm import Session
 
-from app.auth import current_user, require_human
+from app.auth import current_user
 from app.db import get_db
 from app.discover import hero as hero_mod
 from app.discover import pins, rails
 from app.discover.budget import Budget, DEFAULT_BUDGET
 from app.models import ComfortPin, User
+from app.plex.cards import norm_lang
 from app.routes._common import owned_or_404, not_found
 
 router = APIRouter(prefix="/api/discover", tags=["discover"])
@@ -45,6 +46,7 @@ class ReorderIn(BaseModel):
 def get_discover(
     budget: Budget = DEFAULT_BUDGET,
     seed: int = 0,
+    lang: str = "en",
     user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
@@ -52,13 +54,14 @@ def get_discover(
 
     `resume` and `shelf` are built once and handed to the hero, so the two rails the hero draws from
     are not queried twice for a single page load."""
-    resume = rails.continue_watching(db, user)
+    lang = norm_lang(lang)
+    resume = rails.continue_watching(db, user, lang=lang)
     # Plex off → the shelf rail (and so the hero's comfort tier) goes YouTube-only like the rails
     # beside it; the pins are kept and `GET /pins` still lists them (`pins.shelf`).
-    shelf = pins.shelf(db, user, include_plex=rails.plex_available(db))
+    shelf = pins.shelf(db, user, include_plex=rails.plex_available(db), lang=lang)
     return {
         "budget": budget,
-        "hero": hero_mod.pick(db, user, budget, seed, resume=resume, shelf=shelf),
+        "hero": hero_mod.pick(db, user, budget, seed, resume=resume, shelf=shelf, lang=lang),
         "continue_watching": resume,
         "comfort": shelf,
     }
@@ -69,7 +72,7 @@ def get_discover(
 # never reaches SQL — `hero.pick` uses it only as set membership over an already-bounded candidate
 # pool (the resume rail's `RAIL_LIMIT` rows, the shelf, `SURPRISE_DRAW` surprise draws) — so there is
 # no scan for a crafted URL to unbound, and the set is built in full before this slice anyway. The
-# comment that claimed otherwise was corrected in round 2; the request line's own length is what
+# comment that claimed otherwise was later corrected; the request line's own length is what
 # actually bounds a hostile `exclude`.
 MAX_EXCLUDE = 32
 
@@ -80,6 +83,7 @@ def get_pick(
     seed: int = 0,
     exclude: str = "",
     start: int = 0,
+    lang: str = "en",
     user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
@@ -99,32 +103,38 @@ def get_pick(
     return {
         "budget": budget,
         "hero": hero_mod.pick(
-            db, user, budget, seed, exclude=set(list(skip)[:MAX_EXCLUDE]), start=start
+            db,
+            user,
+            budget,
+            seed,
+            exclude=set(list(skip)[:MAX_EXCLUDE]),
+            start=start,
+            lang=norm_lang(lang),
         ),
     }
 
 
 @router.get("/pins")
 def list_pins(
+    lang: str = "en",
     user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     """The shelf on its own — for the shelf editor, which needs it without the rest of the page."""
-    return {"items": pins.shelf(db, user)}
+    return {"items": pins.shelf(db, user, lang=norm_lang(lang))}
 
 
 @router.post("/pins")
 def add_pin(
     body: PinIn,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     """Pin something to the comfort shelf.
 
-    `require_human`, like every other personal-state write: the shared demo account must not be able
-    to curate a communal shelf. Re-pinning something already on the shelf is a no-op that returns the
-    existing pin rather than a 409 — the gesture is "make sure this is on my shelf", and the unique
-    constraint is the mechanism, not the message."""
+    Re-pinning something already on the shelf is a no-op that returns the existing pin rather than
+    a 409 — the gesture is "make sure this is on my shelf", and the unique constraint is the
+    mechanism, not the message."""
     resolved = pins.resolve_ref(db, user, body.kind, body.ref)
     if resolved is None:
         raise not_found("item")
@@ -159,7 +169,7 @@ def add_pin(
             # between `resolve_ref` confirming it and this INSERT (a Plex sync dropping a removed
             # item), which trips the FK. Semantically that is the same "no such thing" the pre-check
             # returns above — re-raising handed the user a 500 they could do nothing with.
-            raise not_found("item")
+            raise not_found("item") from None
         return {"ok": True, "id": winner.id, "created": False}
     return {"ok": True, "id": pin.id, "created": True}
 
@@ -167,7 +177,7 @@ def add_pin(
 @router.delete("/pins/{pin_id}")
 def remove_pin(
     pin_id: int,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     pin = owned_or_404(db, ComfortPin, pin_id, user, what="pin")
@@ -179,7 +189,7 @@ def remove_pin(
 @router.post("/pins/reorder")
 def reorder_pins(
     body: ReorderIn,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     """Rewrite the shelf order from the list of pin ids the client shows.

backend/app/routes/downloads.py

@@ -1,9 +1,8 @@
 """Download center HTTP API — the per-user surface over the worker/cache/quota layers.
 
-All routes are behind `require_human` (the shared demo account can't spend server disk / needs a
-real identity). Admin routes add `admin_user`. Live job state is polled by the frontend (the
-worker is a separate process); this module never downloads — it enqueues, manages job state,
-serves finished files (range-aware, with the user's custom display name), and shares them.
+All routes are behind `current_user`; admin routes add `admin_user`. Live job state is polled by the
+frontend (the worker is a separate process); this module never downloads — it enqueues, manages job
+state, serves finished files (range-aware, with the user's custom display name), and shares them.
 """
 
 import os
@@ -21,8 +20,10 @@ from pydantic import BaseModel
 from sqlalchemy import func, select
 from sqlalchemy.orm import Session
 
+from app.mailtext import lang_for_user
+from app.collation import collated
 from app import audit, sysconfig
-from app.auth import admin_user, require_human
+from app.auth import admin_user, current_user
 from app.config import settings
 from app.db import get_db, release_read_connection
 from app.downloads import credentials as dlcreds
@@ -31,6 +32,7 @@ from app.downloads import links as linksmod
 from app.downloads import quota, service, storage
 from app.downloads import upload as uploadmod
 from app.downloads import urlguard
+from app.downloads.status import JobStatus
 from app.models import (
     DownloadJob,
     DownloadLink,
@@ -130,8 +132,8 @@ def _serialize(job: DownloadJob, asset: MediaAsset | None) -> dict:
         "job_kind": job.job_kind,
         "source_job_id": job.source_job_id,
         "edit_spec": job.edit_spec,
-        "can_download": ready and job.status == "done",
-        "expired": job.asset_id is None and job.status == "done",
+        "can_download": ready and job.status == JobStatus.DONE,
+        "expired": job.asset_id is None and job.status == JobStatus.DONE,
         "asset": None
         if asset is None
         else {
@@ -210,7 +212,7 @@ def _serialize_profile(p: DownloadProfile) -> dict:
 
 
 @router.get("/profiles")
-def list_profiles(user: User = Depends(require_human), db: Session = Depends(get_db)) -> list[dict]:
+def list_profiles(user: User = Depends(current_user), db: Session = Depends(get_db)) -> list[dict]:
     rows = (
         db.execute(
             select(DownloadProfile)
@@ -232,7 +234,7 @@ class CreateProfileIn(BaseModel):
 
 @router.post("/profiles")
 def create_profile(
-    body: CreateProfileIn, user: User = Depends(require_human), db: Session = Depends(get_db)
+    body: CreateProfileIn, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     name = body.name.strip()
     spec = body.spec
@@ -260,7 +262,7 @@ class UpdateProfileIn(BaseModel):
 def update_profile(
     profile_id: int,
     body: UpdateProfileIn,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     p = db.get(DownloadProfile, profile_id)
@@ -282,7 +284,7 @@ def update_profile(
 
 @router.delete("/profiles/{profile_id}")
 def delete_profile(
-    profile_id: int, user: User = Depends(require_human), db: Session = Depends(get_db)
+    profile_id: int, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     p = db.get(DownloadProfile, profile_id)
     if p is None or p.is_builtin or p.user_id != user.id:
@@ -298,7 +300,7 @@ def delete_profile(
 
 
 @router.get("/credentials")
-def list_credentials(user: User = Depends(require_human), db: Session = Depends(get_db)) -> dict:
+def list_credentials(user: User = Depends(current_user), db: Session = Depends(get_db)) -> dict:
     return {"hosts": dlcreds.status_list(db, user.id)}
 
 
@@ -310,7 +312,7 @@ class SetCredentialIn(BaseModel):
 def set_credential(
     host_key: str,
     body: SetCredentialIn,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     """Store (encrypted) this user's cookies.txt for one supported host. Write-only: the value is
@@ -320,7 +322,7 @@ def set_credential(
     try:
         cleaned = dlcreds.validate_cookies(body.cookies)
     except dlcreds.InvalidCookies as e:
-        raise HTTPException(status_code=400, detail={"code": "cookies", "reason": str(e)})
+        raise HTTPException(status_code=400, detail={"code": "cookies", "reason": str(e)}) from e
     dlcreds.upsert_credential(db, user.id, host_key, cleaned)
     audit.record(
         db,
@@ -335,7 +337,7 @@ def set_credential(
 
 @router.delete("/credentials/{host_key}")
 def delete_credential(
-    host_key: str, user: User = Depends(require_human), db: Session = Depends(get_db)
+    host_key: str, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     if not dlcreds.is_supported_host(host_key):
         raise not_found("host")
@@ -386,7 +388,7 @@ class EnqueueDownloadIn(BaseModel):
 
 @router.post("")
 def enqueue_download(
-    body: EnqueueDownloadIn, user: User = Depends(require_human), db: Session = Depends(get_db)
+    body: EnqueueDownloadIn, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     source_kind, source_ref = resolve_source(body.source)
     # SEC-12: a non-YouTube source is a raw URL the worker will fetch server-side, so refuse the
@@ -405,14 +407,14 @@ def enqueue_download(
     try:
         job = service.enqueue(db, user.id, source_kind, source_ref, spec, profile_id, display_name)
     except quota.QuotaExceeded as e:
-        raise HTTPException(status_code=422, detail=_quota_detail(e))
+        raise HTTPException(status_code=422, detail=_quota_detail(e)) from e
     return _serialize_job(db, job)
 
 
 # The upload's filename and display name travel as HEADERS, not as multipart parts. That is not a
 # style choice: a DECLARED body parameter (`file: UploadFile = File(...)`) is read and spooled by
-# FastAPI *before the first dependency runs*, so `require_human` could not gate it and an anonymous
-# caller could make the process write the whole per-file allowance to temp, per request (P1; SEC-20
+# FastAPI *before the first dependency runs*, so the auth dependency could not gate it and an anonymous
+# caller could make the process write the whole per-file allowance to temp, per request (SEC-20
 # bounded the number but could not stop the writing). Reading the body ourselves, from a dependency
 # declared AFTER the auth one, is the only shape that closes it — the same one
 # `POST /api/debug-sessions/{sid}/screenshots` uses, and `test_body_limit.py` now enforces it for
@@ -452,13 +454,13 @@ def _header_text(request: Request, name: str) -> str:
 
 async def spool_upload(
     request: Request,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> AsyncIterator[SpooledUpload]:
     """Stream the request body to a staging temp file, enforcing the per-file cap as it arrives.
 
     `user` is taken but unused: it is the whole point. Dependencies resolve in declaration order, so
-    naming `require_human` here makes the auth check a precondition of reading the body at all — a
+    naming `current_user` here makes the auth check a precondition of reading the body at all — a
     stranger is refused before a byte moves. The `yield` shape owns the temp file's life: ingest
     MOVES it into place on success (the unlink is then a no-op), and on any other exit — cap hit,
     empty, quota, probe or move error, a client that vanished mid-upload — it is removed."""
@@ -497,11 +499,12 @@ def upload_download(
     upload: SpooledUpload = Depends(spool_upload),
     db: Session = Depends(get_db),
 ) -> dict:
-    """Ingest a local file into the caller's library (E7 S5). Any file type is accepted; media
+    """Ingest a local file into the caller's library. Any file type is accepted; media
     (video/audio) gets the full function set, a generic file is download-only. The body IS the file
     (see the header note above); `spool_upload` has already authenticated the caller and streamed it
     to a temp path under the download volume, so what is left here is the non-yt-dlp ingest which
-    probes + places it and returns an already-`done` job.
+    probes + places it and returns its job — `done`, or `queued` for the worker's phone-safe
+    conversion (`downloads.upload`).
 
     A sync (threadpool) handler on purpose: the ingest does blocking work — an ffprobe of up to 30s
     plus sync DB writes — so it must NOT run on the event loop. The streaming half is async because
@@ -518,7 +521,7 @@ def upload_download(
             display_name=upload.display_name,
         )
     except quota.QuotaExceeded as e:
-        raise HTTPException(status_code=422, detail=_quota_detail(e))
+        raise HTTPException(status_code=422, detail=_quota_detail(e)) from e
     return _serialize_job(db, job)
 
 
@@ -530,7 +533,7 @@ class EnqueueEditIn(BaseModel):
 
 @router.post("/edit")
 def enqueue_edit(
-    body: EnqueueEditIn, user: User = Depends(require_human), db: Session = Depends(get_db)
+    body: EnqueueEditIn, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     """Queue a phase-2 editor derivative (trim/crop) of a download this user can access.
 
@@ -538,16 +541,16 @@ def enqueue_edit(
     clip is a NEW per-user derivative in the editor's own library (counts against their quota); the
     source file is only read, never modified, so editing a shared video is safe."""
     source_job = _accessible_job(db, user, body.source_job_id)
-    if source_job.status != "done" or source_job.asset_id is None:
+    if source_job.status != JobStatus.DONE or source_job.asset_id is None:
         raise refused(409, "edit_unfinished")
     spec = body.edit_spec
     display_name = display_text(body.display_name)
     try:
         job = service.enqueue_edit(db, user.id, source_job, spec, display_name)
     except quota.QuotaExceeded as e:
-        raise HTTPException(status_code=422, detail=_quota_detail(e))
+        raise HTTPException(status_code=422, detail=_quota_detail(e)) from e
     except service.EditError as e:
-        raise HTTPException(status_code=400, detail={"code": "edit", "reason": e.reason})
+        raise HTTPException(status_code=400, detail={"code": "edit", "reason": e.reason}) from e
     return _serialize_job(db, job)
 
 
@@ -558,9 +561,7 @@ def _quota_detail(e: quota.QuotaExceeded) -> dict:
 
 
 @router.get("")
-def list_downloads(
-    user: User = Depends(require_human), db: Session = Depends(get_db)
-) -> list[dict]:
+def list_downloads(user: User = Depends(current_user), db: Session = Depends(get_db)) -> list[dict]:
     jobs = (
         db.execute(
             select(DownloadJob)
@@ -576,7 +577,7 @@ def list_downloads(
 
 
 @router.get("/usage")
-def my_usage(user: User = Depends(require_human), db: Session = Depends(get_db)) -> dict:
+def my_usage(user: User = Depends(current_user), db: Session = Depends(get_db)) -> dict:
     """The caller's download quota, plus the per-file upload cap.
 
     The cap rides here rather than in `quota.usage` because it is not a quota value — it is a
@@ -594,17 +595,22 @@ def my_usage(user: User = Depends(require_human), db: Session = Depends(get_db))
 
 
 # Priority when a source has several jobs (e.g. a done one plus a fresh queued one).
-_INDEX_RANK = {"done": 4, "running": 3, "paused": 2, "queued": 1}
+_INDEX_RANK: dict[str, int] = {
+    JobStatus.DONE: 4,
+    JobStatus.RUNNING: 3,
+    JobStatus.PAUSED: 2,
+    JobStatus.QUEUED: 1,
+}
 
 
 @router.get("/index")
-def my_download_index(user: User = Depends(require_human), db: Session = Depends(get_db)) -> dict:
+def my_download_index(user: User = Depends(current_user), db: Session = Depends(get_db)) -> dict:
     """source_ref -> best active status, for the feed's per-card "downloaded / queued" badge.
     Small + cheap so the feed can poll it without loading the full job list."""
     rows = db.execute(
         select(DownloadJob.source_ref, DownloadJob.status).where(
             DownloadJob.user_id == user.id,
-            DownloadJob.status.in_(("queued", "running", "paused", "done")),
+            DownloadJob.status.in_(tuple(_INDEX_RANK)),
         )
     ).all()
     out: dict[str, str] = {}
@@ -615,9 +621,7 @@ def my_download_index(user: User = Depends(require_human), db: Session = Depends
 
 
 @router.get("/shared")
-def shared_with_me(
-    user: User = Depends(require_human), db: Session = Depends(get_db)
-) -> list[dict]:
+def shared_with_me(user: User = Depends(current_user), db: Session = Depends(get_db)) -> list[dict]:
     jobs = (
         db.execute(
             select(DownloadJob)
@@ -677,7 +681,7 @@ class UpdateDownloadMetaIn(BaseModel):
 def update_download_meta(
     job_id: int,
     body: UpdateDownloadMetaIn,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     """Update a download's display metadata: title, channel name + link, and extra reference URLs.
@@ -700,7 +704,7 @@ def update_download_meta(
 
 @router.post("/{job_id}/pause")
 def pause_download(
-    job_id: int, user: User = Depends(require_human), db: Session = Depends(get_db)
+    job_id: int, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     job = _own_job_for_update(db, user, job_id)
     service.pause(db, job)
@@ -709,7 +713,7 @@ def pause_download(
 
 @router.post("/{job_id}/resume")
 def resume_download(
-    job_id: int, user: User = Depends(require_human), db: Session = Depends(get_db)
+    job_id: int, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     job = _own_job_for_update(db, user, job_id)
     service.resume(db, job)
@@ -718,7 +722,7 @@ def resume_download(
 
 @router.post("/{job_id}/redownload")
 def redownload_download(
-    job_id: int, user: User = Depends(require_human), db: Session = Depends(get_db)
+    job_id: int, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     """Force a fresh re-download of a finished (or failed) job, rebuilding the file + Plex tree.
 
@@ -737,7 +741,7 @@ def redownload_download(
         # Neither has a re-fetchable source: an editor clip is derived locally, an upload IS the
         # source. Re-running yt-dlp against source_kind='upload'/'edit' would just fail.
         raise refused(409, "not_redownloadable")
-    if job.status in ("queued", "running", "paused"):
+    if job.status in (JobStatus.QUEUED, JobStatus.RUNNING, JobStatus.PAUSED):
         raise refused(409, "still_downloading")
     service.redownload(db, job)
     return _serialize_job(db, job)
@@ -745,7 +749,7 @@ def redownload_download(
 
 @router.post("/{job_id}/cancel")
 def cancel_download(
-    job_id: int, user: User = Depends(require_human), db: Session = Depends(get_db)
+    job_id: int, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     job = _own_job_for_update(db, user, job_id)
     service.cancel(db, job)
@@ -754,7 +758,7 @@ def cancel_download(
 
 @router.delete("/{job_id}")
 def delete_download(
-    job_id: int, user: User = Depends(require_human), db: Session = Depends(get_db)
+    job_id: int, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     job = _own_job_for_update(db, user, job_id)
     service.delete(db, job)
@@ -784,7 +788,7 @@ def _accessible_job(db: Session, user: User, job_id: int) -> DownloadJob:
 def download_file(
     job_id: int,
     request: Request,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ):
     job = _accessible_job(db, user, job_id)
@@ -823,7 +827,7 @@ def download_file(
 
 @router.get("/{job_id}/storyboard")
 def storyboard_meta(
-    job_id: int, user: User = Depends(require_human), db: Session = Depends(get_db)
+    job_id: int, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     """Filmstrip geometry for the editor's scrub track, generating the sprite on first use.
 
@@ -848,7 +852,7 @@ def storyboard_meta(
 
 
 @router.get("/{job_id}/poster.jpg")
-def poster_image(job_id: int, user: User = Depends(require_human), db: Session = Depends(get_db)):
+def poster_image(job_id: int, user: User = Depends(current_user), db: Session = Depends(get_db)):
     """The locally-generated poster frame for a download whose source had no thumbnail."""
     job = _accessible_job(db, user, job_id)
     asset = db.get(MediaAsset, job.asset_id) if job.asset_id else None
@@ -862,7 +866,7 @@ def poster_image(job_id: int, user: User = Depends(require_human), db: Session =
 
 @router.get("/{job_id}/storyboard.jpg")
 def storyboard_image(
-    job_id: int, user: User = Depends(require_human), db: Session = Depends(get_db)
+    job_id: int, user: User = Depends(current_user), db: Session = Depends(get_db)
 ):
     job = _accessible_job(db, user, job_id)
     asset = db.get(MediaAsset, job.asset_id) if job.asset_id else None
@@ -885,11 +889,11 @@ class ShareDownloadIn(BaseModel):
 def share_download(
     job_id: int,
     body: ShareDownloadIn,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     job = owned_or_404(db, DownloadJob, job_id, user, what="download")
-    if job.status != "done" or job.asset_id is None:
+    if job.status != JobStatus.DONE or job.asset_id is None:
         raise refused(409, "share_unfinished")
     target = body.email.strip().lower()
     if not target:
@@ -897,7 +901,7 @@ def share_download(
     recipient = db.execute(
         select(User).where(func.lower(User.email) == target)
     ).scalar_one_or_none()
-    # Uniform "no user" for a missing OR non-messageable (demo/suspended/deactivated) target, so
+    # Uniform "no user" for a missing OR non-messageable (suspended/deactivated) target, so
     # sharing can't be used to probe which addresses exist or to reach a suspended account.
     if not is_messageable_user(recipient):
         raise not_found("recipient")
@@ -917,7 +921,14 @@ def share_download(
             user_id=recipient.id,
             type="download_shared",
             title=(job.display_name or "A download") + " was shared with you",
-            data={"kind": "download_shared", "from": user.email, "job_id": job_id},
+            # The client names the notice in the reader's language from `data`; the title is the
+            # English fallback only.
+            data={
+                "kind": "download_shared",
+                "from": user.email,
+                "job_id": job_id,
+                "title": job.display_name,
+            },
             commit=False,
         )
         db.commit()
@@ -928,7 +939,7 @@ def share_download(
 def unshare_download(
     job_id: int,
     email: str,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     owned_or_404(
@@ -952,7 +963,7 @@ def unshare_download(
 
 @router.delete("/shared/{job_id}")
 def remove_shared_with_me(
-    job_id: int, user: User = Depends(require_human), db: Session = Depends(get_db)
+    job_id: int, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     """Remove a download that was shared WITH this user from their 'Shared with me' list. Deletes
     only the recipient's share grant — the owner's job and the physical file are untouched (this is
@@ -973,15 +984,15 @@ def remove_shared_with_me(
 
 @router.get("/recipients")
 def share_recipients(
-    user: User = Depends(require_human), db: Session = Depends(get_db)
+    user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> list[dict]:
     """Registered users this user can share a download with (excludes self + any non-messageable
-    account — demo, suspended, or deactivated — so the picker never leaks those addresses)."""
+    account — suspended or deactivated — so the picker never leaks those addresses)."""
     rows = (
         db.execute(
             select(User)
             .where(User.id != user.id, *messageable_clauses())
-            .order_by(func.lower(User.email))
+            .order_by(collated(User.email, lang_for_user(user)))
         )
         .scalars()
         .all()
@@ -1012,7 +1023,7 @@ def _link_expiry(days: int | None) -> datetime | None:
 
 @router.get("/{job_id}/links")
 def list_links(
-    job_id: int, user: User = Depends(require_human), db: Session = Depends(get_db)
+    job_id: int, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> list[dict]:
     owned_or_404(db, DownloadJob, job_id, user, what="download")
     rows = (
@@ -1029,7 +1040,7 @@ def list_links(
 
 class CreateLinkIn(BaseModel):
     password: str = ""
-    # D1 (E7 S3): a newly shared download is downloadable by default — the common intent is "let the
+    # A newly shared download is downloadable by default — the common intent is "let the
     # recipient have the file", not stream-only. Forward-only: this only sets the default for links
     # minted from now on; existing links keep whatever `allow_download` they were created with (the
     # column default stays False, so no back-catalogue link silently flips). The owner can still
@@ -1042,11 +1053,11 @@ class CreateLinkIn(BaseModel):
 def create_link(
     job_id: int,
     body: CreateLinkIn,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     job = owned_or_404(db, DownloadJob, job_id, user, what="download")
-    if job.status != "done" or job.asset_id is None:
+    if job.status != JobStatus.DONE or job.asset_id is None:
         raise refused(409, "share_unfinished")
     pw = body.password.strip()
     link = DownloadLink(
@@ -1073,7 +1084,7 @@ class UpdateLinkIn(BaseModel):
 def update_link(
     link_id: int,
     body: UpdateLinkIn,
-    user: User = Depends(require_human),
+    user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     link = _own_link(db, user, link_id)
@@ -1094,7 +1105,7 @@ def update_link(
 
 @router.delete("/links/{link_id}")
 def revoke_link(
-    link_id: int, user: User = Depends(require_human), db: Session = Depends(get_db)
+    link_id: int, user: User = Depends(current_user), db: Session = Depends(get_db)
 ) -> dict:
     link = _own_link(db, user, link_id)
     db.delete(link)
@@ -1133,10 +1144,10 @@ def admin_list_downloads(
 
 @admin_router.get("/storage")
 def admin_storage(admin: User = Depends(admin_user), db: Session = Depends(get_db)) -> dict:
-    ready = db.execute(
-        select(func.count(), func.coalesce(func.sum(MediaAsset.size_bytes), 0)).where(
-            MediaAsset.status == "ready"
-        )
+    # Every file on disk (`quota.ON_DISK`): ready ones, and uploads waiting for or failed in their
+    # conversion — the same disk the footprint and the GC cap count.
+    on_disk = db.execute(
+        select(func.count(), func.coalesce(func.sum(MediaAsset.size_bytes), 0)).where(quota.ON_DISK)
     ).one()
     total_assets = db.execute(select(func.count()).select_from(MediaAsset)).scalar()
     per_user = db.execute(
@@ -1149,8 +1160,8 @@ def admin_storage(admin: User = Depends(admin_user), db: Session = Depends(get_d
         for u in db.execute(select(User).where(User.id.in_([uid for uid, _ in per_user]))).scalars()
     }
     return {
-        "ready_files": ready[0],
-        "total_bytes": int(ready[1]),
+        "files_on_disk": on_disk[0],
+        "total_bytes": int(on_disk[1]),
         "total_assets": total_assets,
         "total_cap_bytes": sysconfig.get_int(db, "download_total_max_bytes"),
         "per_user": [

backend/app/routes/feed.py

@@ -10,6 +10,7 @@ from sqlalchemy.orm.exc import StaleDataError
 from app import quota
 from app.auth import current_user
 from app.db import get_db
+from app.mailtext import lang_for_user
 from app.feedquery import (
     decode_cursor,
     encode_cursor,
@@ -18,8 +19,11 @@ from app.feedquery import (
     filtered_query,
     keyset_predicate,
     order_by,
+    rank_window_start,
     serialize,
     sort_keys,
+    windowed_rank,
+    windowed_relevance_rows,
 )
 from app.models import (
     POSITION_MAX_S,
@@ -108,30 +112,45 @@ def get_feed(
     db: Session = Depends(get_db),
 ) -> dict:
     query, rank_expr = filtered_query(db, user, **params)
-    keys = sort_keys(sort, seed, rank_expr)
-
-    # Expose each key column on the row so we can build the next cursor from the last item.
-    query = query.add_columns(*[k["expr"].label(f"_k{i}") for i, k in enumerate(keys)])
-    query = query.order_by(*order_by(keys))
+    lang = lang_for_user(user)
+    keys = sort_keys(sort, seed, rank_expr, lang)
 
+    vals, cursor_id, window = None, None, None
     if cursor:
         # The read-model raises a plain ValueError on a malformed cursor; translate it to the
         # HTTP 400 here so feedquery stays web-framework-free.
         try:
-            vals, cursor_id = decode_cursor(keys, cursor)
+            vals, cursor_id, window = decode_cursor(keys, cursor)
         except ValueError:
             # `from None`: the ValueError is an expected bad-input signal, not an error to chain.
             raise refused(400, "invalid_cursor") from None
-        for k, v in zip(keys, vals):
-            k["value"] = v
-        query = query.where(keyset_predicate(keys, cursor_id))
+    ranked = sort == "relevance" and rank_expr is not None
+    if ranked and not cursor:
+        window = rank_window_start(db, query)
+    if ranked and window is not None:
+        keys = sort_keys(sort, seed, windowed_rank(rank_expr, window), lang)
 
-    rows = db.execute(query.limit(limit + 1)).all()
+    # Expose each key column on the row so we can build the next cursor from the last item.
+    query = query.add_columns(*[k["expr"].label(f"_k{i}") for i, k in enumerate(keys)])
+    if vals is not None:
+        for k, v in zip(keys, vals, strict=True):
+            k["value"] = v
+    if ranked and window is not None:
+        rows = windowed_relevance_rows(db, query, keys, window, cursor_id, limit + 1)
+    else:
+        query = query.order_by(*order_by(keys))
+        if cursor_id is not None:
+            query = query.where(keyset_predicate(keys, cursor_id))
+        rows = db.execute(query.limit(limit + 1)).all()
     has_more = len(rows) > limit
     page = rows[:limit]
     return {
         "items": [serialize(r) for r in page],
-        "next_cursor": encode_cursor(keys, page[-1]) if (has_more and page) else None,
+        "next_cursor": (
+            encode_cursor(keys, page[-1], window if ranked else None)
+            if (has_more and page)
+            else None
+        ),
         "has_more": has_more,
         "limit": limit,
     }
@@ -216,7 +235,7 @@ def get_feed_years(
     user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
-    """Per-year video counts for the year control (item 8 / E8·S2): the quick recent-year chips and the
+    """Per-year video counts for the year control: the quick recent-year chips and the
     "more…" per-year-counts dialog. Every filter EXCEPT the publish-date narrowing applies, so picking a
     year re-queries the feed but not this list. Response: {"years": [{"year": <int>, "count": <int>}]},
     newest year first."""
@@ -400,11 +419,7 @@ def get_video_detail(
             "duration_seconds": v.duration_seconds,
         }
 
-    # The DB path above is free; the off-catalog YouTube lookup below spends a quota unit, so the
-    # demo account can't reach it — it just sees "unknown" for a video we don't already store.
-    if user.is_demo:
-        raise not_found("video")
-
+    # The DB path above is free; the off-catalog YouTube lookup below spends a quota unit.
     try:
         with (
             quota.attribute(user.id, quota.QuotaAction.VIDEOS_LOOKUP),
@@ -412,7 +427,7 @@ def get_video_detail(
         ):
             items = yt.get_videos([video_id])
     except YouTubeError as exc:
-        raise youtube_refusal(exc)
+        raise youtube_refusal(exc) from exc
     if not items:
         raise not_found("video")
 

backend/app/routes/me.py

@@ -1,6 +1,6 @@
 from typing import Any
 
-from fastapi import APIRouter, BackgroundTasks, Depends, HTTPException, Request
+from fastapi import APIRouter, BackgroundTasks, Depends, Request
 from pydantic import BaseModel, RootModel
 from sqlalchemy import func, select
 from sqlalchemy.orm import Session
@@ -40,7 +40,7 @@ def my_accounts(
     """The accounts that have authenticated in this browser session (switchable without a
     new Google sign-in). The active one is flagged."""
     ids = request.session.get("account_ids") or [user.id]
-    rows = {u.id: u for u in db.query(User).filter(User.id.in_(ids)).all()}
+    rows = {u.id: u for u in db.execute(select(User).where(User.id.in_(ids))).scalars()}
     out = []
     for uid in ids:  # preserve recency order from the session
         u = rows.get(uid)
@@ -131,7 +131,6 @@ def get_me(
         "display_name": user.display_name,
         "avatar_url": user.avatar_url,
         "role": user.role,
-        "is_demo": user.is_demo,
         "has_google": user.google_sub is not None,
         # The address Google answers with, when it DIFFERS from the account's own. Same string for
         # most people, and then this is None so no surface has to decide whether to show it — the
@@ -141,13 +140,16 @@ def get_me(
         ),
         "has_password": user.password_hash is not None,
         # The account's latest sign-in on any device — shown in Settings → Account, so a sign-in the
-        # user does not recognise is something they can see (null: none since 0.88, or the demo).
+        # user does not recognise is something they can see (null: none since 0.88).
         "last_login_at": iso(user.last_login_at),
         # Instance-wide: whether Google sign-in / YouTube connect is available at all (the UI hides
         # YouTube-access affordances and the onboarding nudge when it isn't configured).
         "google_enabled": google_enabled(db),
         # Whether the optional Plex module is enabled (UI shows the Plex feed source when so).
         "plex_enabled": sysconfig.get_bool(db, "plex_enabled"),
+        # Which live-search source runs, so the search button can say what a search costs: "scrape"
+        # spends no API quota, "api" spends 100 units a page.
+        "search_source": sysconfig.get_str(db, "search_source"),
         "can_read": has_read_scope(user),
         "can_write": has_write_scope(user),
         "pending_invites": pending_invites,
@@ -156,12 +158,12 @@ def get_me(
         "pending_activations": pending_activations,
         "email_configured": email_configured,
         "preferences": user.preferences or {},
-        # The consented diagnostics level in force and when it lapses (E22 S5). The client applies
+        # The consented diagnostics level in force and when it lapses. The client applies
         # it to its own log and streams that log while it is raised.
         "diagnostics": diagnostics.state(db, user),
-        # The live debug session, if one is recording (E16) — a reload resumes it.
+        # The live debug session, if one is recording — a reload resumes it.
         "debug_session": _live_session(db, user),
-        # Where a bug report goes (E16 S4): the tracker's repository, or None — download only.
+        # Where a bug report goes: the tracker's repository, or None — download only.
         "issues": gitea.target(db),
     }
 
@@ -181,15 +183,8 @@ def delete_my_account(
     """GDPR self-service erasure: permanently delete the signed-in account and ALL its personal
     data — OAuth tokens, subscriptions, tags, watch/save/hide states, playlists, notifications —
     which all cascade on the users row. Quota-audit events are kept but anonymised (FK SET NULL).
-    The shared demo account can't be deleted, and the last remaining admin can't delete itself
-    (that would lock everyone out of the admin surfaces)."""
-    # Structured, like the other three demo refusals: `req` leaves 403 to the caller and
-    # `notifyActionError` discards a plain-string detail in favour of the caller's translated
-    # fallback, so a hand-written English sentence here reaches a Hungarian UI as a generic
-    # "couldn't do that" — or, on this affordance, as nothing at all. `reason` distinguishes it
-    # from the other sites' bare `{code: "demo"}` so the UI can say what specifically is refused.
-    if user.is_demo:
-        raise HTTPException(status_code=403, detail={"code": "demo", "reason": "delete_account"})
+    The last remaining admin can't delete itself (that would lock everyone out of the admin
+    surfaces)."""
     if user.role == "admin" and count_admins(db) <= 1:
         raise refused(400, "only_admin_delete")
     user_id = user.id
@@ -239,10 +234,7 @@ def update_diagnostics(
 ) -> dict:
     """Raise the diagnostics level (the user consented in Settings → Diagnosztika) or return to the
     base. The expiry is the server's: a raised level always lapses after
-    `diagnostics.ELEVATION_HOURS`, whatever the client asks. The shared demo account cannot raise
-    it — its log would interleave every visitor's session under one account."""
-    if user.is_demo and body.level != diagnostics.BASE:
-        raise HTTPException(status_code=403, detail={"code": "demo", "reason": "diagnostics"})
+    `diagnostics.ELEVATION_HOURS`, whatever the client asks."""
     diagnostics.set_level(user, body.level)
     db.commit()
     # The rest of THIS request, and the process's elevation registry, follow the new level now.

backend/app/routes/messages.py

@@ -1,459 +0,0 @@
-"""Direct messaging (notification module, phase 2): end-to-end encrypted user-to-user chat
-plus server-authored system messages, with live WebSocket delivery.
-
-E2EE model: every private message is encrypted IN THE BROWSER under a key both parties derive
-from ECDH(their private key, the other's public key). The server only ever stores ciphertext +
-iv and acts as a key directory (public keys) and an opaque blob store (each user's private key,
-wrapped client-side with a passphrase the server never sees). So not even an admin can read a
-user-to-user conversation. `kind="system"` messages (the welcome; future announcements) are
-plain server boilerplate — non-private, readable without any key setup, shown as a "Siftlode"
-conversation.
-
-Every endpoint is gated by `require_human` (the shared demo account can't message). Metadata
-(who messaged whom, when, read state) is NOT encrypted — standard for E2EE and needed for
-routing and unread counts.
-"""
-
-from datetime import datetime, timezone
-
-from fastapi import APIRouter, Depends, WebSocket, WebSocketDisconnect
-from pydantic import BaseModel
-from sqlalchemy import and_, case, delete, func, or_, select
-from sqlalchemy.orm import Session
-
-from app.auth import require_human, resolved_user_id
-from app.db import SessionLocal, get_db
-from app.models import Message, MessageKey, User
-from app.ratelimit import RateLimiter
-from app.realtime import manager
-from app.security import verify_password
-from app.userscope import is_messageable_user, messageable_clauses
-from app.utils import iso
-from app.refusal import refused
-from app.routes._common import not_found
-
-router = APIRouter(prefix="/api/messages", tags=["messages"])
-
-MAX_CIPHERTEXT = 8000  # base64 of a generous plaintext + AES-GCM overhead
-SYSTEM_PARTNER_ID = 0  # synthetic conversation partner for system messages
-SYSTEM_NAME = "Siftlode"
-_send_limiter = RateLimiter(max_events=30, window_seconds=60)
-_reset_key_limiter = RateLimiter(max_events=5, window_seconds=300)
-
-# The welcome is rendered server-side at creation time, so it carries its own translations
-# (one per supported UI language). Kept short and non-private (every new user gets the same text).
-WELCOME = {
-    "en": (
-        "Welcome to Siftlode! 👋\n\n"
-        "This is your private message inbox. Conversations with other members are end-to-end "
-        "encrypted — only you and the person you're chatting with can read them, not even an "
-        "admin. To start messaging someone, set up secure messaging with a passphrase first.\n\n"
-        "Enjoy your Feed!"
-    ),
-    "hu": (
-        "Üdv a Siftlode-on! 👋\n\n"
-        "Ez a privát üzenet-postafiókod. A többi taggal folytatott beszélgetések végpontig "
-        "titkosítottak — csak te és a beszélgetőpartnered olvashatja őket, még az admin sem. "
-        "Ha üzenni szeretnél valakinek, előbb állítsd be a biztonságos üzenetküldést egy "
-        "jelszóval.\n\n"
-        "Jó böngészést!"
-    ),
-}
-
-
-class KeySetupIn(BaseModel):
-    public_key: str
-    wrapped_private_key: str
-    salt: str
-    wrap_iv: str
-    key_check: str
-
-
-class KeyResetIn(BaseModel):
-    # Present only for password accounts; a Google-only account has no password to re-verify.
-    current_password: str | None = None
-
-
-class SendIn(BaseModel):
-    recipient_id: int
-    ciphertext: str
-    iv: str
-
-
-def _user_label(u: User) -> str:
-    return u.display_name or u.email.split("@")[0]
-
-
-def _serialize_user(u: User) -> dict:
-    return {"id": u.id, "name": _user_label(u), "avatar_url": u.avatar_url}
-
-
-def _system_partner() -> dict:
-    return {"id": SYSTEM_PARTNER_ID, "name": SYSTEM_NAME, "avatar_url": None}
-
-
-def _serialize_msg(m: Message) -> dict:
-    return {
-        "id": m.id,
-        "kind": m.kind,
-        "sender_id": m.sender_id,
-        "recipient_id": m.recipient_id,
-        "body": m.body,  # only set for system messages
-        "ciphertext": m.ciphertext,
-        "iv": m.iv,
-        "read_at": iso(m.read_at),
-        "created_at": iso(m.created_at),
-    }
-
-
-def ensure_welcome(db: Session, user: User) -> None:
-    """Create the one-time system welcome message for a (human) user the first time it's needed.
-    Idempotent: guarded on whether any system message already exists for them."""
-    if user.is_demo:
-        return
-    exists = db.scalar(
-        select(Message.id).where(Message.recipient_id == user.id, Message.kind == "system").limit(1)
-    )
-    if exists:
-        return
-    lang = (user.preferences or {}).get("language")
-    body = WELCOME.get(lang, WELCOME["en"])
-    m = Message(kind="system", sender_id=None, recipient_id=user.id, body=body)
-    db.add(m)
-    db.commit()
-    db.refresh(m)
-    manager.push(user.id, {"type": "message", "message": _serialize_msg(m)})
-
-
-# --- E2EE key directory / blob store -----------------------------------------
-
-
-@router.get("/keys/me")
-def my_key(user: User = Depends(require_human), db: Session = Depends(get_db)) -> dict:
-    """My own key bundle. `configured` says whether secure messaging is set up; if so the
-    wrapped blob + params are returned so this device can unlock with the passphrase."""
-    k = db.get(MessageKey, user.id)
-    if k is None:
-        return {"configured": False}
-    return {
-        "configured": True,
-        "public_key": k.public_key,
-        "wrapped_private_key": k.wrapped_private_key,
-        "salt": k.salt,
-        "wrap_iv": k.wrap_iv,
-        "key_check": k.key_check,
-    }
-
-
-@router.post("/keys")
-def setup_keys(
-    payload: KeySetupIn,
-    user: User = Depends(require_human),
-    db: Session = Depends(get_db),
-) -> dict:
-    """Store this user's public key + passphrase-wrapped private key. Set-once: changing the
-    key would orphan every existing conversation key, so re-setup is rejected here."""
-    if db.get(MessageKey, user.id) is not None:
-        raise refused(409, "messaging_set_up")
-    db.add(
-        MessageKey(
-            user_id=user.id,
-            public_key=payload.public_key,
-            wrapped_private_key=payload.wrapped_private_key,
-            salt=payload.salt,
-            wrap_iv=payload.wrap_iv,
-            key_check=payload.key_check,
-        )
-    )
-    db.commit()
-    return {"ok": True, "configured": True}
-
-
-@router.post("/keys/reset")
-def reset_keys(
-    payload: KeyResetIn,
-    user: User = Depends(require_human),
-    db: Session = Depends(get_db),
-) -> dict:
-    """Forgotten-passphrase escape hatch: drop this user's secure-messaging key so they can set a
-    NEW passphrase (the set-once guard in setup_keys otherwise makes a lost passphrase a permanent
-    wall). Destroys the user's entire E2EE history BY DESIGN — every conversation key is derived
-    from this keypair, so replacing it orphans every stored ciphertext in both directions. Those
-    now-undecryptable `user` messages are deleted here (a clean slate for both parties) rather than
-    left as permanent "can't decrypt" clutter; `system` messages (plaintext) are untouched.
-
-    An account WITH a password must re-authenticate: a hijacked session must not be able to silently
-    reset messaging and then read the victim's future messages under a new passphrase. A password-
-    less (Google-only) account has nothing to re-verify, so it relies on its live session plus the
-    client-side danger confirmation."""
-    if not _reset_key_limiter.allow(str(user.id)):
-        raise refused(429, "too_many_attempts")
-    if user.password_hash and not verify_password(
-        payload.current_password or "", user.password_hash
-    ):
-        raise refused(403, "wrong_password")
-    # Delete the now-dead ciphertext first (both directions), then the wrapped keypair itself so a
-    # fresh setup is allowed. Order doesn't matter (no FK between them), but do it in one commit.
-    db.execute(
-        delete(Message).where(
-            Message.kind == "user",
-            or_(Message.sender_id == user.id, Message.recipient_id == user.id),
-        )
-    )
-    key = db.get(MessageKey, user.id)
-    if key is not None:
-        db.delete(key)
-    db.commit()
-    return {"ok": True, "configured": False}
-
-
-@router.get("/keys/{user_id}")
-def public_key(
-    user_id: int, user: User = Depends(require_human), db: Session = Depends(get_db)
-) -> dict:
-    """A member's public key, for deriving the pairwise conversation key. MB2 (like get_thread):
-    resolve the target only if they're messageable OR we already share a thread, else the same
-    "User not found" as a non-existent id — so this can't be probed to enumerate user_ids or to
-    fetch keys for suspended/deactivated/demo accounts."""
-    cond = and_(
-        Message.kind == "user",
-        or_(
-            and_(Message.sender_id == user.id, Message.recipient_id == user_id),
-            and_(Message.sender_id == user_id, Message.recipient_id == user.id),
-        ),
-    )
-    p = db.get(User, user_id)
-    has_history = db.scalar(select(Message.id).where(cond).limit(1))
-    if p is None or (not is_messageable_user(p) and not has_history):
-        raise not_found("user")
-    k = db.get(MessageKey, user_id)
-    if k is None:
-        raise refused(404, "no_messaging")
-    return {"user_id": user_id, "public_key": k.public_key}
-
-
-# --- directory + conversations -----------------------------------------------
-
-
-@router.get("/users")
-def list_recipients(
-    user: User = Depends(require_human), db: Session = Depends(get_db)
-) -> list[dict]:
-    """Member directory for the recipient picker: messageable members (except me) who have set
-    up secure messaging (so a conversation key can be derived)."""
-    rows = (
-        db.execute(
-            select(User)
-            .join(MessageKey, MessageKey.user_id == User.id)
-            .where(User.id != user.id, *messageable_clauses())
-            .order_by(func.lower(func.coalesce(User.display_name, User.email)))
-        )
-        .scalars()
-        .all()
-    )
-    return [_serialize_user(u) for u in rows]
-
-
-@router.get("/conversations")
-def list_conversations(user: User = Depends(require_human), db: Session = Depends(get_db)) -> dict:
-    """One entry per conversation: each user partner (E2EE) plus the system "Siftlode" thread,
-    with the latest message + my unread count, newest first. Lazily creates the welcome on the
-    first open. User-message previews are ciphertext — the client decrypts them."""
-    ensure_welcome(db, user)
-    # The conversation partner for a message, from this user's perspective (system → 0).
-    partner_expr = case(
-        (Message.kind == "system", SYSTEM_PARTNER_ID),
-        (Message.sender_id == user.id, Message.recipient_id),
-        else_=Message.sender_id,
-    )
-    mine = or_(Message.sender_id == user.id, Message.recipient_id == user.id)
-    # Latest message per conversation in ONE query (DISTINCT ON, Postgres) instead of loading the
-    # user's whole message history into Python — bounded by the number of conversations.
-    latest = (
-        db.execute(
-            select(Message)
-            .where(mine)
-            .distinct(partner_expr)
-            .order_by(partner_expr, Message.created_at.desc())
-        )
-        .scalars()
-        .all()
-    )
-    # Unread incoming, grouped per conversation.
-    unread = dict(
-        db.execute(
-            select(partner_expr, func.count())
-            .where(mine, Message.recipient_id == user.id, Message.read_at.is_(None))
-            .group_by(partner_expr)
-        ).all()
-    )
-
-    def _pid(m: Message) -> int:
-        if m.kind == "system":
-            return SYSTEM_PARTNER_ID
-        return m.recipient_id if m.sender_id == user.id else m.sender_id
-
-    user_ids = [_pid(m) for m in latest if m.kind != "system"]
-    partners = {}
-    if user_ids:
-        partners = {
-            u.id: u for u in db.execute(select(User).where(User.id.in_(user_ids))).scalars().all()
-        }
-    items = []
-    for m in latest:
-        pid = _pid(m)
-        if pid == SYSTEM_PARTNER_ID:
-            partner = _system_partner()
-        elif pid in partners:
-            partner = _serialize_user(partners[pid])
-        else:
-            continue
-        items.append(
-            {
-                "partner": partner,
-                "last_message": _serialize_msg(m),
-                "unread": int(unread.get(pid, 0)),
-            }
-        )
-    items.sort(key=lambda x: x["last_message"]["created_at"] or "", reverse=True)
-    return {"items": items}
-
-
-@router.get("/conversations/{partner_id}")
-def get_thread(
-    partner_id: int,
-    mark_read: bool = True,
-    user: User = Depends(require_human),
-    db: Session = Depends(get_db),
-) -> dict:
-    """Full thread with a partner (or the system thread when partner_id == 0), oldest first.
-    Stamps unread incoming messages read unless mark_read=false."""
-    if partner_id == SYSTEM_PARTNER_ID:
-        ensure_welcome(db, user)
-        partner = _system_partner()
-        cond = and_(Message.recipient_id == user.id, Message.kind == "system")
-    else:
-        p = db.get(User, partner_id)
-        cond = and_(
-            Message.kind == "user",
-            or_(
-                and_(Message.sender_id == user.id, Message.recipient_id == partner_id),
-                and_(Message.sender_id == partner_id, Message.recipient_id == user.id),
-            ),
-        )
-        # MB2: don't leak arbitrary user identities. Resolve a partner only if they're messageable
-        # OR we already share a thread (so a conversation with a since-suspended/deactivated user
-        # still opens). Otherwise the same 404 as a non-existent id → no enumeration oracle.
-        has_history = db.scalar(select(Message.id).where(cond).limit(1))
-        if p is None or (not is_messageable_user(p) and not has_history):
-            raise not_found("user")
-        partner = _serialize_user(p)
-    msgs = (
-        db.execute(select(Message).where(cond).order_by(Message.created_at.asc())).scalars().all()
-    )
-    if mark_read:
-        now = datetime.now(timezone.utc)
-        changed = False
-        for m in msgs:
-            if m.recipient_id == user.id and m.read_at is None:
-                m.read_at = now
-                changed = True
-        if changed:
-            db.commit()
-            # MB3: ping this user's OTHER tabs/devices that their unread dropped so the badge
-            # refreshes live instead of waiting for the next 20s poll. The client re-fetches the
-            # count on this signal, so no number is carried here.
-            manager.push(user.id, {"type": "unread"})
-    return {"partner": partner, "items": [_serialize_msg(m) for m in msgs]}
-
-
-@router.post("")
-def send_message(
-    payload: SendIn,
-    user: User = Depends(require_human),
-    db: Session = Depends(get_db),
-) -> dict:
-    """Send an end-to-end-encrypted message. The server stores only the ciphertext the client
-    produced, then pushes it live to the recipient (and the sender's other devices)."""
-    if not _send_limiter.allow(f"msg:{user.id}"):
-        raise refused(429, "message_rate")
-    ct = (payload.ciphertext or "").strip()
-    iv = (payload.iv or "").strip()
-    if not ct or not iv:
-        raise refused(400, "message_empty")
-    if len(ct) > MAX_CIPHERTEXT:
-        raise refused(400, "message_too_long")
-    if payload.recipient_id == user.id:
-        raise refused(400, "message_self")
-    recipient = db.get(User, payload.recipient_id)
-    if not is_messageable_user(recipient):
-        raise not_found("recipient")
-    if db.get(MessageKey, recipient.id) is None:
-        raise refused(404, "no_messaging")
-    m = Message(kind="user", sender_id=user.id, recipient_id=recipient.id, ciphertext=ct, iv=iv)
-    db.add(m)
-    db.commit()
-    db.refresh(m)
-    # Carry both parties so each recipient can open/flash the right dock window (the partner is
-    # whichever of from/to isn't them).
-    event = {
-        "type": "message",
-        "message": _serialize_msg(m),
-        "from": _serialize_user(user),
-        "to": _serialize_user(recipient),
-    }
-    manager.push(recipient.id, event)
-    manager.push(user.id, event)  # the sender's other open devices
-    return _serialize_msg(m)
-
-
-@router.get("/unread_count")
-def unread_count(user: User = Depends(require_human), db: Session = Depends(get_db)) -> dict:
-    n = db.scalar(
-        select(func.count())
-        .select_from(Message)
-        .where(Message.recipient_id == user.id, Message.read_at.is_(None))
-    )
-    return {"count": n or 0}
-
-
-# --- live delivery -----------------------------------------------------------
-
-
-@router.websocket("/ws")
-async def messages_ws(ws: WebSocket) -> None:
-    """Live push channel: authenticated by the session cookie, registers the connection so
-    sent messages reach this user's open tabs instantly. We don't consume client→server frames
-    (sending goes through POST /api/messages); the receive loop only detects disconnect."""
-    # Which signed-in account this socket acts as — the SHARED per-tab resolution (a wallet-gated
-    # ?account=, the session default when none is named, nobody for an account no longer signed in
-    # here). A browser WebSocket can't send the X-Siftlode-Account header, so resolved_user_id falls
-    # through to the ?account= query param.
-    uid, _ = resolved_user_id(ws)
-    if not uid:
-        await ws.close(code=1008)
-        return
-    db = SessionLocal()
-    try:
-        user = db.get(User, uid)
-        # SA4: honour server-side session revocation on the live channel too — reject a cookie whose
-        # recorded epoch is behind the account's current one (mirrors current_user). Without this a
-        # copied cookie could keep receiving pushes after a password reset / "log out everywhere".
-        epochs = ws.session.get("epochs") or {}
-        epoch_ok = user is not None and int(epochs.get(str(uid), 0)) == (user.session_epoch or 0)
-        ok = epoch_ok and is_messageable_user(user)
-    finally:
-        db.close()
-    if not ok:
-        await ws.close(code=1008)
-        return
-    await manager.connect(uid, ws)
-    try:
-        while True:
-            await ws.receive_text()
-    # silent-ok: the ordinary way a chat socket ends — a closed tab. The `finally` below is the
-    # whole handling, and nothing failed.
-    except WebSocketDisconnect:
-        pass
-    finally:
-        await manager.disconnect(uid, ws)

backend/app/routes/notifications.py

@@ -1,9 +1,5 @@
 """Per-user notification inbox (phase 1): the durable, server-backed center that coexists
-with the client-side transient bell. Read-only listing plus read/dismiss/clear mutations.
-
-All endpoints use `current_user` (not `require_human`): notifications and their read/dismiss
-state are personal UI state with no quota or YouTube involvement, so the shared demo account
-manages its own inbox normally."""
+with the client-side transient bell. Read-only listing plus read/dismiss/clear mutations."""
 
 from fastapi import APIRouter, Depends, Query
 from sqlalchemy import func, select, update

backend/app/routes/playlists.py

@@ -342,7 +342,7 @@ def sync_youtube(
                 return {"ok": True, **pull_user_playlists(db, user)}
         except YouTubeError as exc:
             db.rollback()
-            raise youtube_refusal(exc)
+            raise youtube_refusal(exc) from exc
 
 
 @router.post("/{playlist_id}/revert-youtube")
@@ -369,10 +369,10 @@ def revert_youtube(
             with quota.attribute(user.id, quota.QuotaAction.PLAYLISTS_PULL):
                 return {"ok": True, **repull_playlist(db, user, pl)}
         except PlaylistGone:
-            raise not_found("playlist")
+            raise not_found("playlist") from None
         except YouTubeError as exc:
             db.rollback()
-            raise youtube_refusal(exc)
+            raise youtube_refusal(exc, playlist=True) from exc
 
 
 def _pushable(db: Session, user: User, playlist_id: int) -> Playlist:
@@ -398,7 +398,7 @@ def push_plan(
         with quota.attribute(user.id, quota.QuotaAction.PLAYLISTS_PUSH):
             plan = plan_merge(db, user, pl)
     except YouTubeError as exc:
-        raise youtube_refusal(exc)
+        raise youtube_refusal(exc, playlist=True) from exc
     plan["remaining_today"] = quota.remaining_today(db)
     plan["affordable"] = plan["units_estimate"] <= plan["remaining_today"]
     return plan
@@ -436,10 +436,10 @@ def push(
                     raise _refused(429, "quota_exceeded")
                 return {"ok": True, **merge_playlist(db, user, pl, live=live)}
         except PlaylistGone:
-            raise not_found("playlist")
+            raise not_found("playlist") from None
         except YouTubeError as exc:
             db.rollback()
-            raise youtube_refusal(exc)
+            raise youtube_refusal(exc, playlist=True) from exc
 
 
 @router.get("/{playlist_id}")
@@ -539,7 +539,7 @@ def delete_playlist(
                     with YouTubeClient(db, user) as yt:
                         yt.delete_playlist(pl.yt_playlist_id)
             except YouTubeError as exc:
-                raise youtube_refusal(exc)
+                raise youtube_refusal(exc, playlist=True) from exc
         # The playlist first, as every item write: the cascade deletes the items before the list row.
         lock_playlists(db, pl.id)
         db.delete(pl)  # items cascade

backend/app/routes/plex/__init__.py

@@ -1,10 +1,10 @@
 """Plex route package — assembles the admin/browse/playback sub-routers.
 
-Read endpoints (libraries/browse/show/image) are available to ANY authenticated user (demo +
-pure-Siftlode included) — this is an access layer to the Plex library WITHOUT requiring a plex.tv
+Read endpoints (libraries/browse/show/image) are available to ANY authenticated user
+(pure-Siftlode included) — this is an access layer to the Plex library WITHOUT requiring a plex.tv
 account, and playback is a local file. Admin endpoints test the connection and run the catalog sync.
 
-Split out of the former single routes/plex.py (R12 S1). `router` keeps the same shape the app
+Split out of the former single routes/plex.py. `router` keeps the same shape the app
 wiring expects (`plex_routes.router`), so main.py is unchanged.
 """
 

backend/app/routes/plex/_shared.py

@@ -4,6 +4,7 @@ from sqlalchemy.orm import Session
 
 from app import logs
 from app.models import PlexItem
+from app.plex.cards import norm_lang as _norm_lang  # noqa: F401 — the package's historic name
 from app.plex.lookups import by_rating_key
 from app.routes._common import not_found
 
@@ -15,8 +16,3 @@ def item_or_404(db: Session, rating_key: str) -> PlexItem:
     if it is None:
         raise not_found("plex_item")
     return it
-
-
-def _norm_lang(lang: str | None) -> str:
-    """The overlay has an EN base + an optional HU twin; anything other than `hu` serves EN."""
-    return "hu" if lang == "hu" else "en"

backend/app/routes/plex/browse.py

@@ -10,10 +10,13 @@ from sqlalchemy import (
     func,
     literal,
     null,
+    or_,
     select,
 )
 from sqlalchemy.orm import Session, aliased
 
+from app.mailtext import lang_for_user
+from app.collation import collated, sorted_words
 from app import sysconfig
 from app.auth import current_user
 from app.db import get_db
@@ -25,7 +28,7 @@ from app.models import (
     User,
 )
 from app.plex import rollup
-from app.plex.cards import episode_card
+from app.plex.cards import display_title, episode_card
 from app.plex.query import (
     FIELD_COLUMNS,
     PLAYBACK_CLASSES,
@@ -54,14 +57,15 @@ from app.utils import to_tsquery_str
 router = APIRouter(prefix="/api/plex", tags=["plex"])
 
 
-def sort_column(u, sort: str):
+def sort_column(u, sort: str, lang: str = "en"):
     """The browse grid's sort key -> orderable column on the union subquery. Module-level (not inline
     in the handler) so a test can pin its vocabulary to `PLEX_SORT_KEYS` — a key the Sort control
     offers but this cannot order would quietly fall back to `added_at` and look like a broken control.
-    An unknown key (a stale bookmark, a hand-edited URL) deliberately keeps that fallback."""
+    An unknown key (a stale bookmark, a hand-edited URL) deliberately keeps that fallback. The title sorts
+    in the reader's alphabet (`collation`)."""
     return {
         "added": u.c.added_at,
-        "title": func.lower(u.c.title),
+        "title": collated(u.c.title, lang),
         "year": u.c.year,
         "rating": u.c.rating,
         "duration": u.c.duration_s,
@@ -78,7 +82,7 @@ def list_libraries(user: User = Depends(current_user), db: Session = Depends(get
         db.execute(
             select(PlexLibrary)
             .where(PlexLibrary.overlay_of.is_(None))
-            .order_by(PlexLibrary.kind.desc(), PlexLibrary.title)
+            .order_by(PlexLibrary.kind.desc(), collated(PlexLibrary.title, lang_for_user(user)))
         )
         .scalars()
         .all()
@@ -237,7 +241,7 @@ def unified_library(
         if tsq is not None:
             order = [u.c.rank.desc()]
         else:
-            col = sort_column(u, sort)
+            col = sort_column(u, sort, lang)
             order = [(col.asc() if sort_dir == "asc" else col.desc()).nullslast()]
         order.append(u.c.rk.desc())
         rows = db.execute(select(u).order_by(*order).offset(offset).limit(limit)).all()
@@ -281,26 +285,32 @@ def unified_library(
     episodes = []
     if tsq is not None and want_shows and field in ("all", "episode"):
         ep_st = aliased(PlexState)
-        ep_vec = (
-            func.to_tsvector(TS_CONFIG, func.coalesce(PlexItem.title, ""))
-            if field == "episode"
-            else PlexItem.search_vector
-        )
+        # In HU an episode is found by its Hungarian title too (the overlay fills it) — and
+        # still by its English one, like `vector_match` does for the grid.
+        if field == "episode":
+            ep_vec = func.to_tsvector(TS_CONFIG, func.coalesce(PlexItem.title, ""))
+            ep_vec_hu = func.to_tsvector(TS_CONFIG, func.coalesce(PlexItem.title_hu, ""))
+        else:
+            ep_vec, ep_vec_hu = PlexItem.search_vector, PlexItem.search_vector_hu
+        hit = ep_vec.op("@@")(tsq)
+        if lang == "hu":
+            hit = or_(ep_vec_hu.op("@@")(tsq), hit)
+            ep_vec = ep_vec_hu
         eps = db.execute(
-            select(PlexItem, ep_st, PlexShow.title.label("show_title"))
+            select(PlexItem, ep_st, PlexShow)
             .outerjoin(ep_st, and_(ep_st.item_id == PlexItem.id, ep_st.user_id == user.id))
             .outerjoin(PlexShow, PlexShow.id == PlexItem.show_id)
             .where(
                 PlexItem.kind == "episode",
                 PlexItem.library_id.in_(show_lib_ids),
-                ep_vec.op("@@")(tsq),
+                hit,
             )
             .order_by(func.ts_rank(ep_vec, tsq).desc())
             .limit(24)
         ).all()
-        for e, s, show_title in eps:
-            card = episode_card(e, s)
-            card["show_title"] = show_title
+        for e, s, sh in eps:
+            card = episode_card(e, s, lang=lang)
+            card["show_title"] = display_title(sh, lang) if sh is not None else None
             episodes.append(card)
 
     return {
@@ -348,6 +358,24 @@ def facets(
 
     uid = user.id
 
+    # The shows that pass the watch-state scope, asked ONCE. The per-user episode rollup behind it is a
+    # full pass over the user's episodes (~10 ms on a prod-sized library), and each of the show groups
+    # (genres, ratings, studios, years, rating max) asked it again: half of a filtered facets call.
+    # The same filter on the same rows, so the counts do not change
+    # (test_plex_facets_grid_parity).
+    shows_in_scope: list[int] | None = None
+    if show_ids and show in ("watched", "in_progress", "unwatched"):
+        agg = rollup.agg_subquery(db, uid).subquery()
+        in_scope = select(PlexShow.id).outerjoin(agg, agg.c.show_id == PlexShow.id)
+        in_scope = apply_show_state_filter(
+            in_scope,
+            func.coalesce(agg.c.total, 0),
+            func.coalesce(agg.c.watched, 0),
+            func.coalesce(agg.c.inprog, 0),
+            show,
+        )
+        shows_in_scope = list(db.scalars(in_scope.where(PlexShow.library_id.in_(show_ids))))
+
     def wsq(query, model, ids, pp, kind_movie, with_dur):
         """Apply the (self-excluded) metadata filters + library scope + the per-user watch-state to a
         facet aggregation query, so the facets reflect exactly the visible result set."""
@@ -366,14 +394,8 @@ def facets(
             st = aliased(PlexState)
             query = query.outerjoin(st, and_(st.item_id == model.id, st.user_id == uid))
             query = apply_movie_state_filter(query, st, show)
-        else:
-            if show in ("watched", "in_progress", "unwatched"):
-                agg = rollup.agg_subquery(db, uid).subquery()
-                query = query.outerjoin(agg, agg.c.show_id == model.id)
-                tot = func.coalesce(agg.c.total, 0)
-                wat = func.coalesce(agg.c.watched, 0)
-                inp = func.coalesce(agg.c.inprog, 0)
-                query = apply_show_state_filter(query, tot, wat, inp, show)
+        elif shows_in_scope is not None:
+            query = query.where(model.id.in_(shows_in_scope))
         return query
 
     genre_counts: dict[str, int] = {}
@@ -523,16 +545,19 @@ def facets(
     for c in split_csv(f.playable):
         if c in PLAYBACK_CLASSES:
             playable_counts.setdefault(c, 0)
+    # Names in the reader's alphabet: the genres are Hungarian words for a Hungarian reader.
+    alpha = {w: i for i, w in enumerate(sorted_words(db, [*genre_counts, *studio_counts], lang))}
     return {
         "genres": [
-            {"value": g, "count": c} for g, c in sorted(genre_counts.items(), key=lambda kv: kv[0])
+            {"value": g, "count": c}
+            for g, c in sorted(genre_counts.items(), key=lambda kv: alpha[kv[0]])
         ],
         "content_ratings": [
             {"value": cr, "count": c} for cr, c in sorted(cr_counts.items(), key=lambda kv: -kv[1])
         ],
         "studios": [
             {"value": s, "count": c}
-            for s, c in sorted(studio_counts.items(), key=lambda kv: (-kv[1], kv[0]))
+            for s, c in sorted(studio_counts.items(), key=lambda kv: (-kv[1], alpha[kv[0]]))
         ],
         # Fixed good→heavy order (not by count), so the class list reads as a severity scale.
         "playable": [

backend/app/routes/plex/collections.py

@@ -9,6 +9,8 @@ from pydantic import BaseModel
 from sqlalchemy import func, select
 from sqlalchemy.orm import Session
 
+from app.mailtext import lang_for_user
+from app.collation import collated, sorted_words
 from app import logs
 from app.auth import admin_user, current_user
 from app.db import get_db
@@ -22,10 +24,10 @@ from app.models import (
 )
 from app.plex import rollup
 from app.plex import sync as plex_sync
-from app.plex.cards import movie_card, show_card
+from app.plex.cards import display_title, movie_card, norm_lang, show_card
 from app.plex.client import PlexClient, PlexError, PlexNotConfigured
 from app.plex.lookups import by_rating_key
-from app.plex.query import content_libs, runtime_by_key, top_art_keys
+from app.plex.query import content_libs, runtime_by_key, title_col, top_art_keys
 from app.utils import like_escape
 from app.refusal import refused
 from app.routes._common import not_found
@@ -142,7 +144,7 @@ def collections(
         db.execute(
             query.order_by(
                 PlexCollection.child_count.desc().nullslast(),
-                func.lower(PlexCollection.title),
+                collated(PlexCollection.title, lang_for_user(user)),
             )
         )
         .scalars()
@@ -208,16 +210,24 @@ MEMBER_SORT_KEYS: dict[str, Callable[[Member], Any]] = {
 DEFAULT_MEMBER_SORT = "year"
 
 
-def sort_members(members: list[Member], sort: str, sort_dir: str) -> list[Member]:
+def sort_members(
+    members: list[Member], sort: str, sort_dir: str, alpha: dict[str, int] | None = None
+) -> list[Member]:
     """Order a collection's members by one of `PLEX_SORT_KEYS`.
 
     Two properties worth stating because they are easy to get wrong with a plain `reverse=True`:
     members with no value for the key sink to the END in BOTH directions (matching the browse grid's
     `nullslast()` — a movie with no rating is not "the worst rated"), and ties always resolve
     title-ascending, so flipping the direction reverses the ranking rather than the alphabet inside
-    it. An unknown key falls back to the page's default instead of returning an arbitrary order."""
+    it. An unknown key falls back to the page's default instead of returning an arbitrary order.
+
+    `alpha` ranks each title in the reader's alphabet (`collation.sorted_words`); without it, titles
+    compare as Python strings, which puts every accented initial after "z"."""
+    by_title = (lambda m: alpha[m.title]) if alpha is not None else (lambda m: m.title)
     key = MEMBER_SORT_KEYS.get(sort, MEMBER_SORT_KEYS[DEFAULT_MEMBER_SORT])
-    ordered = sorted(members, key=lambda m: m.title)  # stable secondary, applied first
+    if key is MEMBER_SORT_KEYS["title"]:
+        key = by_title
+    ordered = sorted(members, key=by_title)  # stable secondary, applied first
     have = [m for m in ordered if key(m) is not None]
     missing = [m for m in ordered if key(m) is None]
     have.sort(key=key, reverse=sort_dir != "asc")
@@ -229,6 +239,7 @@ def collection_detail(
     rating_key: str,
     sort: str = DEFAULT_MEMBER_SORT,
     sort_dir: str = "asc",
+    lang: str = "en",
     user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
@@ -237,6 +248,7 @@ def collection_detail(
     control means the same thing on both surfaces; the default is release year ascending, which is
     what this page did before it had a control. Read-only here; edits go through the admin mutation
     endpoints below."""
+    lang = norm_lang(lang)
     col = by_rating_key(db, PlexCollection, rating_key)
     if col is None:
         raise not_found("collection")
@@ -244,7 +256,7 @@ def collection_detail(
         db.execute(
             select(PlexItem)
             .where(PlexItem.kind == "movie", PlexItem.collection_keys.contains([rating_key]))
-            .order_by(PlexItem.year.asc().nullslast(), func.lower(PlexItem.title))
+            .order_by(PlexItem.year.asc().nullslast(), collated(title_col(PlexItem, lang), lang))
         )
         .scalars()
         .all()
@@ -253,7 +265,7 @@ def collection_detail(
         db.execute(
             select(PlexShow)
             .where(PlexShow.collection_keys.contains([rating_key]))
-            .order_by(PlexShow.year.asc().nullslast(), func.lower(PlexShow.title))
+            .order_by(PlexShow.year.asc().nullslast(), collated(title_col(PlexShow, lang), lang))
         )
         .scalars()
         .all()
@@ -285,7 +297,7 @@ def collection_detail(
         Member(
             kind="movie",
             obj=m,
-            title=(m.title or "").lower(),
+            title=display_title(m, lang).lower(),
             year=m.year,
             rating=m.rating,
             added=m.added_at,
@@ -298,7 +310,7 @@ def collection_detail(
         Member(
             kind="show",
             obj=sh,
-            title=(sh.title or "").lower(),
+            title=display_title(sh, lang).lower(),
             year=sh.year,
             rating=sh.rating,
             added=sh.added_at,
@@ -308,11 +320,12 @@ def collection_detail(
         )
         for sh in shows
     ]
-    members = sort_members(members, sort, sort_dir)
+    alpha = {w: i for i, w in enumerate(sorted_words(db, [m.title for m in members], lang))}
+    members = sort_members(members, sort, sort_dir, alpha)
     items = [
-        movie_card(m.obj, mstate.get(m.obj.id))
+        movie_card(m.obj, mstate.get(m.obj.id), lang=lang)
         if m.kind == "movie"
-        else show_card(m.obj, sstate.get(m.obj.id, "new"))
+        else show_card(m.obj, sstate.get(m.obj.id, "new"), lang=lang)
         for m in members
     ]
     years = [m.year for m in members if m.year is not None]
@@ -415,7 +428,7 @@ def create_collection(
         db.commit()
     except (PlexError, PlexNotConfigured) as e:
         db.rollback()
-        raise _write_or_pending_error(wrote, e)
+        raise _write_or_pending_error(wrote, e) from e
     return _collection_card(col)
 
 
@@ -437,7 +450,7 @@ def collection_add_item(
         db.commit()
     except (PlexError, PlexNotConfigured) as e:
         db.rollback()
-        raise _write_or_pending_error(wrote, e)
+        raise _write_or_pending_error(wrote, e) from e
     return _collection_card(col)
 
 
@@ -459,7 +472,7 @@ def collection_remove_item(
         db.commit()
     except (PlexError, PlexNotConfigured) as e:
         db.rollback()
-        raise _write_or_pending_error(wrote, e)
+        raise _write_or_pending_error(wrote, e) from e
     return _collection_card(col)
 
 
@@ -489,7 +502,7 @@ def collection_rename(
         db.commit()
     except (PlexError, PlexNotConfigured) as e:
         db.rollback()
-        raise _write_or_pending_error(wrote, e)
+        raise _write_or_pending_error(wrote, e) from e
     return _collection_card(col)
 
 

backend/app/routes/plex/detail.py

@@ -1,14 +1,16 @@
 """Plex show/item detail: the show drill-down + the per-item player payload (markers, cast, nav)."""
 
+import re
 import threading
 import time
 from concurrent.futures import ThreadPoolExecutor
 from datetime import datetime, timezone
 
 from fastapi import APIRouter, Depends
-from sqlalchemy import func, select
+from sqlalchemy import select
 from sqlalchemy.orm import Session
 
+from app.collation import collated, sorted_words
 from app import logs
 from app.auth import current_user
 from app.db import get_db
@@ -24,6 +26,7 @@ from app.models import (
 from app.plex import detail as plex_detail
 from app.plex import people as plex_people
 from app.plex import rollup, subtitles
+from app.plex.query import title_col
 from app.plex import tmdb as plex_tmdb
 from app.plex.cards import episode_card, movie_card, show_card
 from app.plex.client import PlexClient, PlexError, PlexNotConfigured
@@ -107,7 +110,8 @@ def show_detail(
     db: Session = Depends(get_db),
 ) -> dict:
     """A show's seasons + episodes (the drill-down view) with per-user watch state."""
-    hu = _norm_lang(lang) == "hu"
+    lang = _norm_lang(lang)
+    hu = lang == "hu"
     sh = by_rating_key(db, PlexShow, rating_key)
     if sh is None:
         raise not_found("show")
@@ -144,19 +148,19 @@ def show_detail(
     by_season: dict[int | None, list] = {}
     all_cards: list[dict] = []  # every episode, in season/episode order → show-level rollup
     for e in eps:
-        card = episode_card(e, states.get(e.id))
+        card = episode_card(e, states.get(e.id), lang=lang)
         by_season.setdefault(e.season_id, []).append(card)
         all_cards.append(card)
     show_roll = rollup.rollup(all_cards)
 
-    # Cast + IMDb + tagline are served from the DB row (persisted by the cast backfill, E5 S5b). The one
+    # Cast + IMDb + tagline are served from the DB row (persisted by the cast backfill). The one
     # remaining live call is `related` — a Plex recommendation hub we don't mirror — which also carries
     # the metadata used to lazily fill a not-yet-enriched show. Any live failure just drops `related` and
     # the lazy fill; the row-served detail below still renders (outage-resilient).
     related: list[dict] = []
     try:
         meta, related_raw = _show_enrichment(db, sh.rating_key)
-        related = _relatedshow_cards(db, user.id, related_raw, exclude=sh.id)
+        related = _relatedshow_cards(db, user.id, related_raw, exclude=sh.id, lang=lang)
         if sh.people_enriched_at is None and meta:
             plex_people.enrich_row_people(db, sh, plex_people.cast_from_meta(meta))
             plex_detail.apply_rich_detail(sh, meta)  # imdb/tagline/full_cast for future views
@@ -209,8 +213,8 @@ def show_detail(
             "first": show_roll["first"],
             "episode_count": show_roll["episode_count"],
             "collection_keys": sh.collection_keys or [],
-            # TMDB enrichment (E5 S2), served from the row.
-            "tmdb": plex_tmdb.serve_block(db, sh),
+            # TMDB enrichment, served from the row.
+            "tmdb": plex_tmdb.serve_block(db, sh, lang=lang),
         },
         "seasons": [_season_block(se, by_season.get(se.id, [])) for se in seasons],
         "related": related,
@@ -234,7 +238,9 @@ def _season_block(se: PlexSeason, cards: list[dict]) -> dict:
     }
 
 
-def _relatedshow_cards(db: Session, user_id: int, related: list[dict], exclude: int) -> list[dict]:
+def _relatedshow_cards(
+    db: Session, user_id: int, related: list[dict], exclude: int, lang: str
+) -> list[dict]:
     """Map Plex 'related' metadata to show cards for the shows we actually mirror (so they're openable),
     excluding the current show. Order preserved; capped."""
     rks = [str(m.get("ratingKey")) for m in related if m.get("ratingKey")]
@@ -254,13 +260,13 @@ def _relatedshow_cards(db: Session, user_id: int, related: list[dict], exclude:
         if sh is None or rk in seen:
             continue
         seen.add(rk)
-        out.append(show_card(sh, statuses.get(sh.id, "new")))
+        out.append(show_card(sh, statuses.get(sh.id, "new"), lang=lang))
         if len(out) >= 20:
             break
     return out
 
 
-def _collection_strips(db: Session, it: PlexItem, user_id: int) -> list[dict]:
+def _collection_strips(db: Session, it: PlexItem, user_id: int, lang: str) -> list[dict]:
     """The collections this movie belongs to, each with its member titles (playable cards) — the
     'Avatar Collection' strip on the info page. Pure local lookup (no Plex call)."""
     keys = it.collection_keys or []
@@ -273,7 +279,8 @@ def _collection_strips(db: Session, it: PlexItem, user_id: int) -> list[dict]:
         ).scalars()
     }
     # Smallest (most specific — e.g. a franchise) first; the big auto-lists (IMDb/TMDb Popular) last.
-    ordered = sorted((c for c in cols.values()), key=lambda c: (c.child_count or 0, c.title))
+    alpha = {w: i for i, w in enumerate(sorted_words(db, [c.title for c in cols.values()], lang))}
+    ordered = sorted(cols.values(), key=lambda c: (c.child_count or 0, alpha[c.title]))
     out = []
     for col in ordered:
         rk = col.rating_key
@@ -285,7 +292,9 @@ def _collection_strips(db: Session, it: PlexItem, user_id: int) -> list[dict]:
                     PlexItem.kind == "movie",
                     PlexItem.collection_keys.contains([rk]),
                 )
-                .order_by(PlexItem.year.asc().nullslast(), func.lower(PlexItem.title))
+                .order_by(
+                    PlexItem.year.asc().nullslast(), collated(title_col(PlexItem, lang), lang)
+                )
                 .limit(60)
             )
             .scalars()
@@ -305,7 +314,7 @@ def _collection_strips(db: Session, it: PlexItem, user_id: int) -> list[dict]:
                 "id": rk,
                 "title": col.title,
                 "source": _collection_source(col),
-                "items": [movie_card(m, states.get(m.id)) for m in members],
+                "items": [movie_card(m, states.get(m.id), lang=lang) for m in members],
             }
         )
     return out
@@ -339,6 +348,26 @@ def _episode_nav(db: Session, it: PlexItem) -> tuple[str | None, str | None]:
     return (keys[i - 1] if i > 0 else None), (keys[i + 1] if i < len(keys) - 1 else None)
 
 
+# The technical tail of Plex's track title — "English (AC3 5.1)" → "AC3 5.1".
+_TRACK_TAIL = re.compile(r"\(([^()]+)\)\s*$")
+
+
+def _track_naming(s: dict) -> dict:
+    """What the player names a track from. Plex's `displayTitle` is in the SERVER's language
+    ("English (AC3 5.1)", "magyar Forced", "中文 (EAC3 Stereo)"), so it is sent only as the fallback
+    `label`; the client builds the name in the reader's language from the language tag, the
+    forced / SDH flags and the codec tail. No title and no language → no label (the client numbers it)."""
+    title = s.get("displayTitle") or s.get("language") or None
+    tail = _TRACK_TAIL.search(title) if title else None
+    return {
+        "label": title,
+        "lang_tag": s.get("languageTag") or None,
+        "forced": bool(s.get("forced")),
+        "hearing_impaired": bool(s.get("hearingImpaired")),
+        "detail": tail.group(1).strip() if tail else None,
+    }
+
+
 @router.get("/item/{rating_key}")
 def item_detail(
     rating_key: str,
@@ -384,9 +413,7 @@ def item_detail(
                 audio_streams.append(
                     {
                         "ord": len(audio_streams),
-                        "label": s.get("displayTitle")
-                        or s.get("language")
-                        or f"Audio {len(audio_streams) + 1}",
+                        **_track_naming(s),
                         "language": s.get("languageTag") or s.get("language"),
                         "default": bool(s.get("selected") or s.get("default")),
                     }
@@ -396,9 +423,7 @@ def item_detail(
                 subtitle_streams.append(
                     {
                         "ord": len(subtitle_streams),
-                        "label": s.get("displayTitle")
-                        or s.get("language")
-                        or f"Subtitle {len(subtitle_streams) + 1}",
+                        **_track_naming(s),
                         "language": s.get("languageTag") or s.get("language"),
                         "codec": codec,
                         # Text subs (srt/ass/…) can be served as WebVTT tracks; image subs (PGS/VobSub)
@@ -468,7 +493,7 @@ def item_detail(
             "genres": (it.genres_hu or it.genres if hu else it.genres) or [],
             "directors": it.directors or [],
             "studio": it.studio,
-            "tagline": it.tagline,
+            "tagline": (it.tagline_hu or it.tagline) if hu else it.tagline,
         }
     else:
         rich = plex_detail.rich_detail(meta)
@@ -512,7 +537,7 @@ def item_detail(
         "episode_number": it.episode_number,
         "prev_id": prev_id,
         "next_id": next_id,
-        "collections": _collection_strips(db, it, user.id),
-        # TMDB enrichment (E5 S2), served from the row — movies only (episodes aren't enriched).
-        "tmdb": plex_tmdb.serve_block(db, it) if it.kind == "movie" else None,
+        "collections": _collection_strips(db, it, user.id, lang),
+        # TMDB enrichment, served from the row — movies only (episodes aren't enriched).
+        "tmdb": plex_tmdb.serve_block(db, it, lang=lang) if it.kind == "movie" else None,
     }

backend/app/routes/plex/discover.py

@@ -1,11 +1,11 @@
-"""Plex Discover: the rail-surface read-model (E5 S4).
+"""Plex Discover: the rail-surface read-model.
 
 A consolidated payload for the dedicated Discover landing — a stack of horizontal rails:
 Continue watching · Recently added · Browse-by-genre tiles · Because you watched X — plus a
 separate, reshuffling Surprise-me sample. Serve-from-DB only: the recommendation rail reads the
-recs the S2 TMDB engine already persisted (`PlexItem.recommendations`), never a live TMDB call, and
+recs the TMDB engine already persisted (`PlexItem.recommendations`), never a live TMDB call, and
 every card comes from the same `movie_card`/`show_card` serializers the grid uses (so rails inherit
-S1's `art` + S3's rating/summary for free). The rails carry EN metadata (like the other non-grid
+their `art` + rating/summary for free). The rails carry EN metadata (like the other non-grid
 card paths); the HU overlay coalesces only in the main grid.
 
 The sidebar metadata filters (genre / rating / year / collection / …) narrow EVERY rail too, so the
@@ -18,11 +18,12 @@ from fastapi import APIRouter, Depends, Query
 from sqlalchemy import and_, func, or_, select
 from sqlalchemy.orm import Session
 
+from app.collation import sorted_words
 from app.auth import current_user
 from app.db import get_db
 from app.models import PlexItem, PlexShow, PlexState, User
 from app.plex import rollup
-from app.plex.cards import leaf_card, movie_card, show_card
+from app.plex.cards import display_title, leaf_card, movie_card, show_card
 from app.plex.query import (
     LibraryFilters,
     apply_meta_filters,
@@ -61,14 +62,14 @@ def _movie_meta(q, p: dict, lang: str, f: LibraryFilters):
     return q
 
 
-def _movie_cards(db: Session, uid: int, items: list[PlexItem]) -> list[dict]:
+def _movie_cards(db: Session, uid: int, items: list[PlexItem], lang: str) -> list[dict]:
     sts = rollup.item_state_map(db, uid, [it.id for it in items])
-    return [movie_card(it, sts.get(it.id)) for it in items]
+    return [movie_card(it, sts.get(it.id), lang=lang) for it in items]
 
 
-def _show_cards(db: Session, uid: int, shows: list[PlexShow]) -> list[dict]:
+def _show_cards(db: Session, uid: int, shows: list[PlexShow], lang: str) -> list[dict]:
     statuses = rollup.state_map(db, uid, [s.id for s in shows])
-    return [show_card(s, statuses.get(s.id, "new")) for s in shows]
+    return [show_card(s, statuses.get(s.id, "new"), lang=lang) for s in shows]
 
 
 def _recently_added(
@@ -113,8 +114,8 @@ def _recently_added(
         else []
     )
     # Merge the two type-homogeneous lists by added_at, then serialize each side in one batch so the
-    # per-card state lookups stay batched (no N+1). Same `or NO_TS` key as the two rails below — the
-    # E24 review found this one still on the older `(is not None, value)` form, i.e. one rule in two
+    # per-card state lookups stay batched (no N+1). Same `or NO_TS` key as the two rails below — a
+    # review found this one still on the older `(is not None, value)` form, i.e. one rule in two
     # spellings, with the two comments explaining it differently.
     tagged = [(m.added_at, "movie", m) for m in movies] + [(s.added_at, "show", s) for s in shows]
     tagged.sort(key=lambda t: t[0] or rollup.NO_TS, reverse=True)
@@ -123,8 +124,14 @@ def _recently_added(
     # share one), so the merge doesn't depend on object identity staying alive.
     top_movies = [o for _, kind, o in top if kind == "movie"]
     top_shows = [o for _, kind, o in top if kind == "show"]
-    mcards = {m.rating_key: c for m, c in zip(top_movies, _movie_cards(db, uid, top_movies))}
-    scards = {s.rating_key: c for s, c in zip(top_shows, _show_cards(db, uid, top_shows))}
+    mcards = {
+        m.rating_key: c
+        for m, c in zip(top_movies, _movie_cards(db, uid, top_movies, lang), strict=True)
+    }
+    scards = {
+        s.rating_key: c
+        for s, c in zip(top_shows, _show_cards(db, uid, top_shows, lang), strict=True)
+    }
     return [mcards[o.rating_key] if kind == "movie" else scards[o.rating_key] for _, kind, o in top]
 
 
@@ -156,10 +163,10 @@ def _recently_watched(
             .order_by(PlexState.watched_at.desc().nullslast())
             .limit(RAIL_LIMIT)
         ).all():
-            rows.append((st.watched_at, leaf_card(it, st)))
+            rows.append((st.watched_at, leaf_card(it, st, lang=lang)))
     if show_ids:
         eq = (
-            select(PlexItem, PlexState, PlexShow.title, PlexShow.rating_key)
+            select(PlexItem, PlexState, PlexShow)
             .join(PlexState, and_(PlexState.item_id == PlexItem.id, PlexState.user_id == uid))
             .join(PlexShow, PlexShow.id == PlexItem.show_id)
             .where(
@@ -170,8 +177,11 @@ def _recently_watched(
             .order_by(PlexState.watched_at.desc().nullslast())
             .limit(RAIL_LIMIT)
         )
-        for ep, st, sh_title, sh_rk in db.execute(eq).all():
-            rows.append((st.watched_at, leaf_card(ep, st, show_title=sh_title, show_rk=sh_rk)))
+        for ep, st, sh in db.execute(eq).all():
+            card = leaf_card(
+                ep, st, show_title=display_title(sh, lang), show_rk=sh.rating_key, lang=lang
+            )
+            rows.append((st.watched_at, card))
     rows.sort(key=lambda r: r[0] or rollup.NO_TS, reverse=True)
     return [card for _, card in rows[:RAIL_LIMIT]]
 
@@ -214,7 +224,7 @@ def _continue_watching(
             .limit(RAIL_LIMIT)
         ).all()
         for it, st in mrows:
-            rows.append((st.progress_updated_at, movie_card(it, st)))
+            rows.append((st.progress_updated_at, movie_card(it, st, lang=lang)))
 
     if show_ids:
         # Reuse the shared per-show episode rollup (total/watched/inprog) rather than re-inlining it, plus
@@ -234,7 +244,7 @@ def _continue_watching(
             .limit(RAIL_LIMIT)
         ).all()
         for sh, resumed_at in srows:
-            rows.append((resumed_at, show_card(sh, "in_progress")))
+            rows.append((resumed_at, show_card(sh, "in_progress", lang=lang)))
 
     rows.sort(key=lambda r: r[0] or rollup.NO_TS, reverse=True)
     return [card for _, card in rows[:RAIL_LIMIT]]
@@ -261,7 +271,8 @@ def _genre_tiles(
         expanded = base.subquery()
         for g, c in db.execute(select(expanded.c.g, func.count()).group_by(expanded.c.g)).all():
             counts[g] = counts.get(g, 0) + c
-    top = sorted(counts.items(), key=lambda kv: (-kv[1], kv[0]))[:GENRE_TILES]
+    alpha = {w: i for i, w in enumerate(sorted_words(db, list(counts), lang))}
+    top = sorted(counts.items(), key=lambda kv: (-kv[1], alpha[kv[0]]))[:GENRE_TILES]
     # A 2×2 poster collage per tile: the top movies in that genre (color hash is the client fallback).
     art = top_art_keys(
         db,
@@ -310,7 +321,7 @@ def _because_you_watched(
     )
     if anchor is None:
         return None
-    resolved = resolve_recs(db, anchor.recommendations)
+    resolved = resolve_recs(db, anchor.recommendations, lang=lang)
     # Preserve TMDB's recommendation order, keep only in-library hits, and drop the anchor itself.
     rks = [
         r["rating_key"]
@@ -346,9 +357,13 @@ def _because_you_watched(
     ordered_shows = [shows[rk] for rk in rks if rk in shows]
     # rating_key is globally unique in Plex, so it keys the serialized cards back safely (no id()).
     mcards = {
-        m.rating_key: c for m, c in zip(ordered_movies, _movie_cards(db, uid, ordered_movies))
+        m.rating_key: c
+        for m, c in zip(ordered_movies, _movie_cards(db, uid, ordered_movies, lang), strict=True)
+    }
+    scards = {
+        s.rating_key: c
+        for s, c in zip(ordered_shows, _show_cards(db, uid, ordered_shows, lang), strict=True)
     }
-    scards = {s.rating_key: c for s, c in zip(ordered_shows, _show_cards(db, uid, ordered_shows))}
     items: list[dict] = []
     for rk in rks:
         if rk in mcards:
@@ -357,7 +372,10 @@ def _because_you_watched(
             items.append(scards[rk])
     if not items:
         return None
-    return {"anchor": {"id": anchor.rating_key, "title": anchor.title}, "items": items}
+    return {
+        "anchor": {"id": anchor.rating_key, "title": display_title(anchor, lang)},
+        "items": items,
+    }
 
 
 @router.get("/discover")
@@ -456,8 +474,8 @@ def surprise(
             .scalars()
             .all()
         )
-    mcards = _movie_cards(db, user.id, movies)
-    scards = _show_cards(db, user.id, shows)
+    mcards = _movie_cards(db, user.id, movies, lang)
+    scards = _show_cards(db, user.id, shows, lang)
     # Interleave movie/show cards, then trim to `limit`.
     items: list[dict] = []
     for i in range(max(len(mcards), len(scards))):

backend/app/routes/plex/people.py

@@ -7,7 +7,8 @@ from fastapi import APIRouter, BackgroundTasks, Depends
 from sqlalchemy import func, or_, select
 from sqlalchemy.orm import Session
 
-from app import logs
+from app.collation import collated
+from app import logs, sysconfig
 from app.auth import current_user
 from app.db import SessionLocal, get_db
 from app.models import (
@@ -18,22 +19,28 @@ from app.models import (
 )
 from app.plex import rollup
 from app.plex import tmdb as plex_tmdb
-from app.plex.cards import movie_card, show_card
-from app.plex.query import content_libs, split_csv
+from app.plex.cards import movie_card, norm_lang, show_card
+from app.plex.query import content_libs, split_csv, title_col
 
 router = APIRouter(prefix="/api/plex", tags=["plex"])
 log = logs.get_logger(__name__)
 
 
+# How many "Known for" titles the person page shows (the stored pool is deeper; see plex.tmdb).
+_KNOWN_FOR_SHOWN = 15
+
+
 def _resolve_known_for(
     db: Session,
     known_for: list | None,
     uid: int,
     movie_lib_ids: list[int],
     show_lib_ids: list[int],
+    lang: str,
 ) -> list[dict]:
-    """The person's TMDB "Known for" credits (top by popularity) that are IN our library, as cards, in
-    popularity order. Matched by the canonical guid `tmdb:<id>`; out-of-library credits are dropped."""
+    """The person's TMDB "Known for" credits (most-voted first) that are IN our library, as cards, in
+    that order, at most _KNOWN_FOR_SHOWN. Matched by the canonical guid `tmdb:<id>`; out-of-library
+    credits are dropped."""
     if not known_for:
         return []
     # Match each credit to a library row by guid, bucketed by media_type (movie ids and tv ids collide
@@ -50,10 +57,10 @@ def _resolve_known_for(
         # A credit resolves against its OWN media_type's bucket only.
         if k.get("media_type") == "tv":
             if g in shows:
-                out.append(show_card(shows[g], show_statuses.get(shows[g].id, "new")))
+                out.append(show_card(shows[g], show_statuses.get(shows[g].id, "new"), lang=lang))
         elif g in movies:
-            out.append(movie_card(movies[g], m_states.get(movies[g].id)))
-    return out
+            out.append(movie_card(movies[g], m_states.get(movies[g].id), lang=lang))
+    return out[:_KNOWN_FOR_SHOWN]
 
 
 @router.get("/people/cards")
@@ -159,6 +166,7 @@ def _enrich_person_bg(person_id: int) -> None:
 def person_detail(
     name: str,
     background_tasks: BackgroundTasks,
+    lang: str = "en",
     user: User = Depends(current_user),
     db: Session = Depends(get_db),
 ) -> dict:
@@ -168,21 +176,27 @@ def person_detail(
     people who share a name share a page. Each section is capped at _PERSON_SECTION_MAX. An unknown name
     simply returns empty sections.
 
-    E5 S6: the first time a page is opened, the person's TMDB bio + "Known for" credits are lazily
-    fetched (from the S2-linked `tmdb_person_id`) and persisted, then served from the row on every open.
+    The first time a page is opened, the person's TMDB bio + "Known for" credits are lazily
+    fetched (from the linked `tmdb_person_id`) and persisted, then served from the row on every open.
     The fetch runs OFF the request path (a background task); the first open serves whatever's in the row
-    (bio possibly still null) and the bio lands from the row by the next open.
+    with `enriching` set, and the page asks again until the bio lands. The people_enrich job
+    fetches everyone ahead of time, so this lazy path is only for a person the job hasn't reached yet.
     """
+    lang = norm_lang(lang)
     person = db.execute(select(PlexPerson).where(PlexPerson.name == name)).scalars().first()
     # Lazy one-time bio enrich, scheduled AFTER the response so the TMDB call is never on the serve path.
     # Gated by a per-person cooldown so concurrent/rapid opens (or repeated opens while TMDB is degraded)
     # can't pile background tasks onto the shared threadpool.
-    if (
+    # `enriching` tells the page the bio is on its way: it asks again shortly instead of showing
+    # an empty page until the next open. True while a fetch is pending, including one an earlier open
+    # already reserved.
+    enriching = (
         person is not None
-        and person.tmdb_person_id
+        and bool(person.tmdb_person_id)
         and person.tmdb_enriched_at is None
-        and _should_schedule_enrich(person.id)
-    ):
+        and bool(sysconfig.get_str(db, "tmdb_api_key"))
+    )
+    if enriching and person is not None and _should_schedule_enrich(person.id):
         background_tasks.add_task(_enrich_person_bg, person.id)
     movie_lib_ids, show_lib_ids = content_libs(db)
 
@@ -199,14 +213,16 @@ def person_detail(
                         PlexItem.directors.contains([name]),
                     ),
                 )
-                .order_by(PlexItem.year.desc().nullslast(), func.lower(PlexItem.title))
+                .order_by(
+                    PlexItem.year.desc().nullslast(), collated(title_col(PlexItem, lang), lang)
+                )
                 .limit(_PERSON_SECTION_MAX)
             )
             .scalars()
             .all()
         )
         states = rollup.item_state_map(db, user.id, [it.id for it in items])
-        movies = [movie_card(it, states.get(it.id)) for it in items]
+        movies = [movie_card(it, states.get(it.id), lang=lang) for it in items]
 
     shows: list[dict] = []
     if show_lib_ids:
@@ -220,28 +236,34 @@ def person_detail(
                         PlexShow.directors.contains([name]),
                     ),
                 )
-                .order_by(PlexShow.year.desc().nullslast(), func.lower(PlexShow.title))
+                .order_by(
+                    PlexShow.year.desc().nullslast(), collated(title_col(PlexShow, lang), lang)
+                )
                 .limit(_PERSON_SECTION_MAX)
             )
             .scalars()
             .all()
         )
         statuses = rollup.state_map(db, user.id, [s.id for s in show_rows])
-        shows = [show_card(sh, statuses.get(sh.id, "new")) for sh in show_rows]
+        shows = [show_card(sh, statuses.get(sh.id, "new"), lang=lang) for sh in show_rows]
 
     known_for = _resolve_known_for(
-        db, person.known_for if person else None, user.id, movie_lib_ids, show_lib_ids
+        db, person.known_for if person else None, user.id, movie_lib_ids, show_lib_ids, lang
     )
     return {
         "name": name,
         "photo": person.photo_thumb if person else None,
-        "biography": person.biography if person else None,
+        # The Hungarian biography for a Hungarian reader, the instance-language one when TMDB has none.
+        "biography": ((person.biography_hu if lang == "hu" else None) or person.biography)
+        if person
+        else None,
         "birthday": person.birthday if person else None,
         "deathday": person.deathday if person else None,
         "place_of_birth": person.place_of_birth if person else None,
         "known_for_department": person.known_for_department if person else None,
         "also_known_as": (person.also_known_as if person else None) or [],
         "known_for": known_for,
+        "enriching": enriching,
         "movies": movies,
         "shows": shows,
     }

backend/app/routes/plex/playback.py

@@ -1,6 +1,5 @@
 """Plex playback: image/subtitle proxies, HLS streaming, and watch-state writes."""
 
-import hashlib
 import threading
 from datetime import datetime, timedelta, timezone
 
@@ -66,7 +65,7 @@ def image(rating_key: str, request: Request, variant: str = "thumb") -> Response
     brief session only to resolve the image key + read the Plex config, then RELEASES it before the
     fetch. Auth is the signed-session check (no DB user-load)."""
     _require_session(request)
-    v = "art" if variant == "art" else "thumb"
+    v = variant if variant in imgcache.SIZES else "thumb"
     w, h = imgcache.SIZES[v]
     # Size is part of the cache key: Plex resizes server-side (see PlexClient.image_bytes), so we cache
     # the small variant. The width tag also means changing a target size just re-caches, never serves a
@@ -78,7 +77,7 @@ def image(rating_key: str, request: Request, variant: str = "thumb") -> Response
     # Cold miss: short-lived session for the key + Plex client config, released before the slow fetch.
     try:
         with SessionLocal() as db:
-            key = imgcache.key_for(db, str(rating_key), v)
+            key = imgcache.key_for(db, str(rating_key), imgcache.picture(v))
             if not key:
                 raise not_found("image")
             plex = PlexClient(db)  # reads base URL + token now; the fetch below needs no DB
@@ -170,7 +169,7 @@ _PROGRESS_MIN_S = 5
 
 _FINISH_MARGIN_S = 30
 
-# S2 write-back throttle: the resume position is now pushed to Plex on the periodic (non-final)
+# Write-back throttle: the resume position is now pushed to Plex on the periodic (non-final)
 # checkpoint too — not only on the final pause/unmount flush — so Plex (the source of truth) stays
 # current in near-real-time. To keep a single watch from spraying `/:/timeline` calls, a push is
 # rate-limited to at most one per item per user every _PUSH_THROTTLE_S seconds; a `final` checkpoint
@@ -214,24 +213,31 @@ def _resume_push_superseded(user_id: int, item_id: int, intent: datetime) -> boo
 
 @router.get("/person-image")
 def person_image(u: str, request: Request) -> Response:
-    """Proxy a cast/crew photo from Plex's public metadata CDN (host-whitelisted; no token needed).
-    Signed-session auth (no DB) so it, like /image, never holds a pool connection during the fetch."""
+    """Proxy a cast/crew photo from Plex's public metadata CDN (only a headshot-shaped URL on its host;
+    no token needed), resized through Plex and cached (plex.people.cache_portrait; the people_enrich job
+    fills the same cache ahead of time). Signed-session auth; like /image, the DB session is released
+    before the fetch."""
     _require_session(request)
-    if not u.startswith(plex_people.PERSON_IMG_HOST):
+    if not plex_people.is_portrait_url(u):
         raise refused(400, "image_host")
-    cache_key = "person_" + hashlib.sha1(u.encode()).hexdigest()
-    hit = imgcache.read(cache_key)
+    hit = plex_people.portrait_cached(u)
     if hit:
         return imgcache.response(*hit)
     try:
-        with httpx.Client(timeout=10.0, follow_redirects=True) as c:
-            r = c.get(u)
-            r.raise_for_status()
-    except httpx.HTTPError:
-        raise refused(502, "image_fetch_failed")
-    ct = r.headers.get("content-type", "image/jpeg")
-    imgcache.write(cache_key, r.content, ct)
-    return imgcache.response(r.content, ct)
+        with SessionLocal() as db:
+            plex: PlexClient | None = PlexClient(db)
+    # silent-ok: Plex off is a setting; without its resize the original comes straight from the CDN.
+    except PlexNotConfigured:
+        plex = None
+    try:
+        if plex is None:
+            data, ct = plex_people.cache_portrait(None, u)
+        else:
+            with plex:
+                data, ct = plex_people.cache_portrait(plex, u)
+    except (PlexError, httpx.HTTPError):
+        raise refused(502, "image_fetch_failed") from None
+    return imgcache.response(data, ct)
 
 
 def _push_resume_guarded(
@@ -354,7 +360,7 @@ def item_progress(
 ) -> dict:
     """Checkpoint the resume position while playing (near the end → mark watched + clear it).
 
-    S2 write-back: the resume position is mirrored to Plex (the source of truth) on the periodic
+    Write-back: the resume position is mirrored to Plex (the source of truth) on the periodic
     checkpoint too, not only on the final flush — throttled to ≥_PUSH_THROTTLE_S per item (see
     `_should_push_resume`). `final` (sent on pause / seek / pagehide / unmount) pushes immediately and
     resets the throttle window, so Plex sees a settled position the instant the user pauses or seeks."""
@@ -690,7 +696,7 @@ def stream_hls_file(
     (viewer, rating_key) and a seek-restart replaces the viewer's registered one, while the DIRECTORY
     is per (key, viewer, offset, tag) — so resolving by the key alone handed the old player the new
     session's `seg_12.ts`, produced from a different offset under `-copyts`: wrong
-    bytes with a 200, which the player has no way to notice (deep-review round 5). A request for a
+    bytes with a 200, which the player has no way to notice. A request for a
     session that is no longer the registered one is a 404, and the player reclaims at the playhead.
 
     That reclaim is also why a request with no live session is no longer auto-started here: the
@@ -722,7 +728,7 @@ def stream_hls_file(
     # `proc` so a session whose ffmpeg has already exited answers at once instead of polling out the
     # whole timeout: this is a SYNC route on the ~40-slot threadpool, holding a pooled DB connection
     # with it, so every such wait is a worker (and a connection) parked for 20 s — an ordinary seek
-    # past the produced region was enough to do it (round 5).
+    # past the produced region was enough to do it.
     if not plex_stream.wait_for(path, timeout, proc=s.proc):
         raise refused(404, "not_ready")
     try: