- Every state and Watch-later write goes through features/video/videoWrites,
which marks the never-refetched lists (related lists, live searches) once
the server has it; a contract test holds it. The Notification Center's
undo, the feed's reset and the Watch-later toggles skipped it before.
- Confirmed marks are also kept apart (qk.videoMarks) and laid over the
related rows: a Show more page landing later rebuilt them from the pages
it read when it started, and dropped a mark made meanwhile.
- Restart on a related video starts at 0 (the session's stored position
won before), and on the one already playing it seeks.
- The watched toggle remembers each video's mark: Back showed the opened
video's launch-time state after a related one was marked.
- The help sheet leaves out the player's keys while a layer above it has
them waiting.
- After a refusal a retry cannot answer, neither Show more (related) nor
Load more (live search) is offered; a failed live-search page keeps the
results already read and says why under them.
- The related cap message counts requests, as the cap does.
Checked first and logged after search.list, requests sent together at the
last free slot all passed, and each spent 100 units of the shared budget.
Every live search now claims its slot (quota.claim_action) whatever the
source, and gives it back when it showed nothing. The cap counts the
zero-unit claims only, so search.list's own spend event is accounting,
not a second slot. log_action, the unlocked writer, goes.
Tests: the API source reads one page per request, is refused before the
call when the budget cannot pay, counts one slot per search and refunds a
failed one; the lock test now proves the count happens after the lock.
A related video played in place is the player's item now: its title and its own watched /
playlist / download controls, its progress checkpointed and resumed, marked watched at its end —
and its end does not step the queue. Back returns to the opened video, where it was left.
The related list:
- its marks announce through the shared notice (the Notification Center finds them), join the
host's player session (the feed refetches on close), and stay per write, so an earlier answer no
longer undoes a later mark still on its way;
- every host's state write patches the cached related lists, which are never refetched;
- kept for the session (gcTime), a retry asks for the page that failed, a list read so far with
nothing unseen says so, a refusal a retry cannot answer offers none, and the followed mark names
itself on hover.
Turned off on the instance (a limit of 0), the tab, the drawer button and the live-search buttons
are left out.
Keys: the player's keys and its focus reclaim defer to a layer above it (the drawer); going wide
moves the drawer's list into the panel's tab; `j` goes on from the slot of a row removed inside the
top layer; the help sheet lists the list keys over a layer that holds a list.
The watch frame alone read as "nothing related" — so a related column YouTube moved, or a list
drawn by renderers this does not read, answered an empty list instead of an error. Now only the
"isn't available" promo YouTube draws for a video it no longer has is an empty page; anything else
it cannot read is a ScrapeError.
The flattening of a video renderer and the read of the continuation token are one each, in
innertube, for the search and the related scrape.
A related list's cursor cost nothing, and a cursor is a free string: a replayed one read and
enriched without end. Every request now takes a slot of the user's daily cap — taken under a lock
before the work (`quota.claim_action`), so requests sent together cannot all pass the last free
one, and given back when the request fails outright (`quota.refund_action`). The live search's
scrape source takes its slots the same way.
A limit of 0 turns the feature off: a distinct 409 `related_disabled` / `search_disabled`, and
`/api/me` says `related_enabled` / `search_enabled` so the UI can leave it out.
A videos.list that fails no longer reads as an empty page: `ingest_candidates` raises, and the
search and the related list answer the refusal (the shared budget spent says "quota") instead of
"no results". The two routes' page loops are one (`_gather`).
`clear` deletes only what the caller found: a related list's rows carry no find, and any user
could name them.
The list keys are the page's and an open layer silences them, so the
player's related videos could only be reached by Tab. A layer now names
its root element; with the focus on a row inside the topmost one, j/k,
the arrows and Enter walk that list, the walk stays inside the layer,
and the player's seek and volume arrows leave a focused row alone.
The player shows what YouTube lists beside the video, without what the
user has watched or hidden. Wide, a Related tab beside the description;
stacked, a button beside the info opens them in a drawer. A row plays in
the same player (Back returns) and the list stays the opened video's.
A switch picks on-topic or Hungarian-trending lists, kept on the account.
Each related row says whether the user follows its channel; the row is
the video card's narrow form with its whole menu. VideoMenu moves to
features/video, where the player can reach it.
GET /api/search/related/{video_id} reads the list YouTube shows beside a
video from the website's InnerTube "next" endpoint (no quota), and brings
it into the catalog the way live-search results come in. It leaves out
what the user watched or hid, here or in their Takeout history (unless
taken back to "new" since), blocked channels and the video itself, and
pages on until 20 are gathered (at most 4 pages). region=topical|hu picks
the locale the list is filled in. Its own per-user daily cap
(related_daily_limit_per_user, one event per opened list); a failed
scrape answers 422 youtube/unavailable.
The InnerTube request (live key, consent cookie, errors) moves into
youtube/innertube.py, shared by the search scrape and the related one.
A subscription pulled in from YouTube left an explored channel flagged as an exploration's, so its
later videos came in hidden from the Library with their watched marks waiting for good; a subscribe
landing while an exploration's videos came in left those flagged too. One step, settle_explored,
now holds the rule by who follows the channel now: the subscribe here, the subscription import, the
watched-channel videos and the explore page all end with it. Migration 0116 clears what the gap
left behind.
The list's "may still grow" count no longer counts a channel it already shows from a stub row, and
the comments on the cleanup's keep rule and on the in-memory daily shares say what they do.
- A channel someone follows: the videos its exploration brings in (the list's
dialog, the channel page's explore) come in as the catalog's own, not
flagged via_explore, which subscribing would never clear and which kept
their watched marks waiting for good.
- The explore cleanup keeps the row of a channel a Takeout history shows
watched (it still reclaims the videos), so a looked-at channel no longer
drops off the list.
- The Shorts job takes up only users whose list still has lengths to ask or
channels to name, by each user's own follows, blocks and dismissals; a
naming that failed is tried again, and the list says it is not complete.
- The length lookups have a daily share of their own: a large history no
longer leaves the export's Watch later file waiting a day.
- Only a channel of the user's own history can be put out of the list.
- The dialog's videos have a query root of their own: the list's
invalidations no longer re-ask (and re-explore) under an open dialog.
The Channel manager lists the channels the Takeout history shows watched
often but not followed, the most watched first: twenty at a time, the
channel's About on hover, its name to its page, Follow, and Not interested
with an Undo. A row's count opens the videos it counts, ready to play one
or all. Before a history is uploaded it points to the upload; while the
import still asks YouTube it says more channels may join.
The import asks YouTube how long the watched videos of the candidate
channels are (five or more watched, none the user follows, blocked, put
out or owns), keeps the length on the Takeout row and names the channels
that make the list: five videos of five minutes or more. The shorts job
asks what an import could not. A channel's counted videos come in as an
exploration's when its list is opened, ready to play; a channel can be put
out of the list and back.
- An ad-hoc request made while its target's run is under way is the run after it: the scheduler
dropped a second fire of a running id, so a second Takeout file was never imported and its key
stayed queued until a restart.
- Un-marking or resetting a video is remembered (video_unmarks): an import of an older YouTube view
no longer marks it watched again; un-marks take the user's Takeout lock.
- One user's failure in the Shorts job's Takeout pass is theirs: the job keeps its own result, the
others are settled, and a failed lookup spends none of the daily share.
- The upload reads the user's id before giving the read connection back.
- A watched row waits on a discovery video (search or explore) until the catalog keeps it, so a
mark never pins it against the cleanup jobs.
- The history is decoded a window at a time, a Watch later file over 2 MiB is refused unread, and
two uploads at once across all users (takeout_server_busy).
- A Watch later list goes in as one presence read and one insert.
- Uploads: one at a time per user and ten an hour, each refused only after its body arrived;
the history read an entry at a time (32 MiB cap), a Watch later list capped at YouTube's 5000,
a user's kept videos at 200k; any unreadable shape is a 422, not a 500.
- Quota: Watch later lookups stop at the backfill reserve and at a per-user daily share; what is
not asked about waits, known videos still go in, and the shorts job retries the rest.
- Staging and settling share one per-user advisory lock (no deadlock); a run settles only the
rows it read, so a row staged meanwhile is never called gone.
- A failed run's notice names a translated failure, never the upstream's text.
- The status says when an import is queued; the page watches it to its end even when left, then
reads the feed, its counts and the playlists again.
- Settings: one live region, the zone kept (unavailable) while a run goes, focus handed back,
a missed drop kept from the browser, an oversize file refused before it is sent.
- Scheduler and run lists show Takeout runs by the importing user.
The import's notice reaches a user who may not see the Scheduler, an
admin page, yet its link opened it. It now opens Settings › Account,
where the import's status is; any other run still opens the Scheduler.
The job id is held to the server's by the backend-ids contract. The
notice's Hungarian names the list as the app does elsewhere.
A "YouTube history" section under YouTube access: the watch history and
the Watch later csv are dropped or picked (one request each, each file's
outcome said on its own), and the section then says what the import did
- only the counts it produced - or that it is running, failed or was cut
off by a restart. While it runs it asks again until it ends; a failed
re-read shows the failure, not the previous answer.
The drop zone the Files upload had inline is now ui/DropZone, used by
both. The e2e seed clears the account's kept Takeout videos.
A watched video the catalog did not know at import time waits in `takeout_items`. The Shorts job now
settles those rows right after its probe (the only place a video's Short-ness becomes known), every
user's, a commit each: `reader.takeout.apply_waiting`, its count in the job's result.
`GET /api/takeout` is what the settings page reads: the user's last import run (`jobruns.last_run`)
and the kept videos by kind and outcome, "waiting" for one not known yet.
The YouTube API reads neither a user's watch history nor their Watch later list, so the Takeout
export is the only way in. `POST /api/takeout` takes one file of it (watch-history.json or
Watch later-videos.csv, told apart by content), keeps its videos in `takeout_items` and queues the
user's `takeout_import` run, whose outcome reaches their inbox.
The run marks watched the known videos the Shorts probe has cleared, at the Takeout time; a Short
is not marked, and a state the user set here (watched, hidden, or changed after that time) wins, by
the upsert's own condition. A video the catalog does not know yet waits with no outcome. Watch later
videos are fetched if unknown and appended oldest first.
An ad-hoc run can now ask for a completion notice (`run_ad_hoc(..., notify=True)`).
During a deploy the gateway answers 502 (or nothing answers at all) until the app is back. The
player took the first such answer for a broken stream: it said "Playback couldn't start" over a
picture still playing from its buffer, stopped loading, and never asked again, so the episode froze
where the buffer ended with the error still up. Now a 502/503/504 or no answer is the server being
away: the buffer keeps playing undisturbed, the session is asked for again where the viewer is
(2, 4, 8, then every 15 s), and the error is said only after three minutes away. A seek or another
item replaces a pending retry. Measured on localdev: a 45 s API stop now resumes at the viewer's
position, where it froze with the error before.
Gates the folder move had narrowed or left open: the MenuCheck contract scans the whole app again (it
saw only components/); local/layers refuses an import of an entry (main.tsx, the /watch page) from
any layer, and reads a backtick import() as the path it is; the /watch contract forbids the shell,
the modules and the features by folder (its one path to App had gone stale); chunkGuard's own test
gets the vitest focus/skip rules. The root tests' exemption from local/layers is in the gate
register, with its three imports.
Placement: Plex's chip builder moves into lib beside the feed's, the folder names into
features/folders, and three files only one page uses into that page. ui/FilterRail takes its reset
from useRail instead of a page id. The channel page reads the feed's view and builds "Show in feed"
itself, and the feed's scroll key is the feed's hook, so the shell holds no feed logic.
The floating header re-measures when one page's pills replace another's in the same commit (the
claim count stayed equal, so only the observer caught it). A Plex view applied and edited now stays
across a visit to another page, as the feed's does: pinned by a test. docs/development.md says what a
page with a rail adds in lib.
CLAUDE.md gains the header-content, module-store and saved-views rows. docs/development.md says what a
new page is and which tests hold it. components/ui/README.md names every file in the directory, and a
contract test keeps it that way: seventeen had slipped in unlisted. copyItems' comment still claimed it
was a feature file.
A module's definition names its store (feed: filters, view, saved views; Plex: browse state, saved
views; channels, files, playlists their provider), and ModuleStores mounts them at the root in
registry order instead of a hand-nested tree in main.tsx. A contract test holds that every store's
read side is in reach and that no store's static imports leave its own module, so the order cannot
matter.
Saved views were one provider that switched between the feed and Plex by the active page and reset
on each switch. The logic is now one hook (useSavedViewsStore) with an instance per module; the
shared chip, popover and modal read the views their page or rail hands down. A feed view applied
and edited now survives a visit to Plex. Plex's views are still asked for only on its page, and a
module's manage modal still closes when its page is left. The feed's URL-override check applies
only on the feed page, where its filter params live.
The Channels store's tab type moves from the page into the store's context.
A page used to be listed three times (lib/modules for its label, icon and gate; shell/moduleRegistry for
its page and rail; lib/pageMeta for its title), and App held every rail's layout and collapse flag.
Now each module's `module.ts` says it all — label, icon, gate, its lazy page and rail, its width —
handed out through its index.ts, and features/navigation/moduleRegistry lists them in nav order for
the rail, the palette, the tab title, the reachable set and the shell's render. A contract test holds
the registry to the page ids and each definition to its own folder, with no static import of a page.
Every rail is lazy now, loaded together with its page so a pinned rail never arrives after it and
shoves the content aside; the feed's and the playlists' rails took @dnd-kit off the boot path (boot
428.5 → 393.9 KiB gzip, the App chunk 122 → 32 KiB). A rail reads its own layout and collapse flag
(lib/useRail) and writes a layout change to the `me` cache too, since it now mounts anew on each
visit. App's session effects, its one-shot entry-URL handlers and the admin nudge move into shell
hooks; App is layout. The registry's three layer suppressions are gone: none is left.
The header's pills now arrive by portal after the shell has mounted, so the shell's one synchronous
measurement saw an empty header, and the ResizeObserver brought `--hdr-h` up a frame or more later:
meanwhile the content sat under the pills (the feed toolbar jumped 55px on a tablet; e2e caught it).
The measurement re-runs before paint whenever what fills the header changes.
The header used to pick its pills by `page === …` and import five module controls, and Plex's
Discover/Browse toggle needed a context bridge to reach the sub-view that lives in PlexBrowse.
Now the header is an empty band: PageShell hands its root box out as a slot, a page portals its
own pills into it (PageHeader), and the palette's Jump-to pill shows where no page does
(PageHeaderDefault, rendered inside the page's Suspense so a loading page shows no stand-in).
Plex's toggles take the sub-view as a prop; PlexViewBridge and its context are gone, and with
them the header's five layer suppressions.
No visible change: every header pill's box measured identical against the previous build on eight
pages, in Hungarian and English.
The guard's logic was a src/lib file the config project also listed. The app project, which
references the config project, then refuses that source on a clean build (TS6305) — while every
local run reused the referenced project's cached output and stayed green; only the Docker image
build failed. The logic is back inside chunkGuard.ts, its test beside it at the frontend root
(vitest now also collects `*.test.ts` there), and a gate contract fails when the config project's
include reaches into src/ again. The guard was re-proved on a build with a leaking re-export.
`local/layers` (eslint.config.js) resolves every import to its file and refuses one that goes up
(lib/ → components/ui/ → features/ → modules/ → shell/) or that reaches into another module's own
files. A module's index.ts is its public face: the stores, mounted at the root and read app-wide,
which features/ may read too; a page is still opened lazily by import(). Its contract test probes
every edge, and three mutants of the rule (off, no face check, a door for everyone) each turn it red.
Six modules got that face; the shared channel UI (about, tags, tag manager) moved to
features/channel, the filter chips and the saved-views context to features/filters, and the prefs
store's controller type moved next to the store instead of importing it from its own page. The
header's per-module controls and the eagerly loaded rails keep eight inline suppressions, each
naming why: they go when the header gets slots and every rail loads lazily.
`chunkGuard` fails the production build when a lazily imported module lands in a boot chunk (the
entry, the shell, what they import) — e.g. a page re-exported from an index.ts, which makes its
import() a silent no-op. Proved on a build with that mutant; today's build has none.
A pure move (git mv, history kept): 333 files leave components/, which now holds only the platform
layer (ui/) and the legal pages. The shell is the frame (App, header, nav rail, palette, boot); each
page lives in modules/<page>/ with its rail, store and tests; what several pages share sits in
features/<name>/; the shared-link page is watch/. Domain-free helpers went straight to ui/
(ViewSwitcher, FilterRail, the confirm dialog, the language switcher, the copy rows).
Every relative specifier was rewritten (imports, dynamic imports, vi.mock, new URL), and the
contract tests that scan sources now scan the whole app source, not only components/ — each glob
was counted before and after so none went green over fewer files. The locale-format rule matched
`/src/components/` only, so the move would have taken ~300 files out from under its i18n-singleton
check; it now covers every folder that holds components, with a test per folder.
Every address-bar writer glued `pathname + query`, so a path of `//evil.com/x` named another host:
history refused it (SecurityError) and the first nav mirror threw the app into its crash screen. One
builder, `hrefHere`, now makes the address from the URL itself and stays on this origin.
- /auth/google/connect is a navigation and carries no headers: the tab names the account it
was built for in `expect`, checked by the same comparison as the request header
(built_for_another_account); a mismatch lands on /?account=changed
- the link marker names the OAuth state of its own flow: an older flow coming back no longer
spends the marker of the connect still open (which then returned as a plain sign-in)
- a 401 account_changed reloads the tab with a one-time flag, and the reloaded tab says it
now shows the account signed in
- a refused switch to a suspended or not-yet-approved account says so in the signed-in
app's words, naming the address a request was recorded for
- the prefs merge locks FOR NO KEY UPDATE, so a transaction that only added a row referring
to the account does not hold the save up; a test pins the lock
- giving up a lost page closes a channel page open over it
- e2e README names the chromium-only specs
A tab names the account it was claimed for on every request, beacon and upload
(X-Siftlode-Expect-Account); current_user and /auth/logout refuse a mismatch with
401 account_changed and keep the session, so a tab left showing the account
signed in before reloads instead of writing to the new one.
/auth/login clears a connect marker left by an abandoned connect, and a link
completes only on a session that still holds. The prefs merge reads the row
under a lock. A tab's sessionStorage records its own owner. A page the account
loses is given up, address bar and remembered page included. The panel layouts
are preferences; the storage contract also refuses a key handed to the Storage
API or to usePref outside its kind. A refused sign-in as someone else and a
failed sign-out say so. The chromium-only e2e specs are in the gate register,
the member walk waits for its requests, and devices.spec checks the save itself.
CLAUDE.md's per-account state row becomes the browser storage rule: a key
in storage.ts with its kind, a preference through usePref. The
self-hosting guide says to set a password even when signing in with
Google, so a broken Google client does not lock the account out.
Performance mode's hint says it applies on this device only.
storage.ts's registry gives each key a kind (pref, device, view, cache,
public); a key without one does not compile. A contract test refuses a key
spelled out elsewhere, an account preference read or written straight from
storage instead of through usePref, and an e2e seed of a key nobody reads.
The keys the next account's wipe keeps are the public ones.
The seven preferences kept only in the browser move onto usePref: start on
Home, the onboarding dismissal, the release banner's last version, the
channel manager's layout, the Plex playlist layout, the playlist rail's sort
and channel discovery's sort. usePref reads an older bare-string copy, and a
value only this browser held is saved to the server once, so the account's
other devices get it.
Performance mode is the device's: it is neither read from the account nor
saved to it.
The per-tab account pin, the X-Siftlode-Account header and ?account= go: the session cookie alone
says who is asking. localStorage keys lose their account suffix; the browser records whose state it
holds (siftlode.owner), BootGate hands it to the account the server answered as, and a different
account starts from a clean slate (the public /watch keys stay). The first boot moves the account's
old per-account copies onto the plain keys and drops other accounts' copies. Until the claim, and
after another tab claimed the browser for someone else, the account helpers read and write nothing.
The account panel lists no other accounts: "Add another account" becomes "Sign in as someone else"
(Google's account chooser; the pick replaces this account, a cancel keeps it), shown where Google
sign-in is configured.
/auth/link and /auth/upgrade become /auth/google/connect?access=identity|read|write: one handler, one
table of what each access asks Google for, the same callback markers. An unknown access is refused
(422) instead of quietly meaning read, and an instance without Google answers ?link=error for a
grant too, not a silent return to the app. The frontend builds the URL in lib/googleConnect, and a
contract test holds it to the backend route and its access values.
peter