fix(logs): deep-review round 1 — say each failure once, and scrub it everywhere

- yt-dlp's text is scrubbed wherever it is logged, not only in the adapter:
  `formats.log_safe` takes out the URL, the `[extractor] <id>:` prefix and
  file paths, and the job-failed, refused, edit-failed and backfill lines use
  it. Done lines log the asset id, not the title-built path. The PII contract
  now bans `rel`/`rel_path`/`title`.
- A traceback goes with a download/edit failure only when it is our own code
  failing; for yt-dlp's DownloadError, our RuntimeErrors and OSErrors its foot
  is the unscrubbed message.
- yt-dlp's `error` maps to WARNING: it fires on attempts the fallback client
  set recovers from; the job's verdict is `download.job.failed`.
- The sink groups a bare `%s` pass-through by its text (numbers folded), and
  a process's last batch closes the repeat window, so a short-lived download
  child still reports what it dropped.
- Worker-side failures (stall, wall-clock, no asset, cached asset error) log
  `download.job.failed` with a cause; they left no line at all before. The
  crash, supervisor-error and non-zero-exit lines carry the code too.
- An edit failure is one coded line in `_fail_edit`, for every path.
- `logs.get_logger("__main__")` resolves to the `-m` module, so the worker
  and the download child keep their names.
- TRACE lines carrying user content are gated on `logs.consented_trace()`,
  not on the process level; a contract test pins it.
- The OAuth token refresh goes through `_upstream_failure`; transport errors
  carry reason `transport`. Subscribe recognises a duplicate by reason code
  only (the prose fallback was dead).
- Once-only guards for a stale sysconfig override and a refusing HLS session
  directory; item_detail splits Plex failures from our own parse bugs.
- The playlist breaker line names the playlist and the action; comments that
  were false (runner noqa, `_merge_vtt`, sink grouping) are corrected.

backend/tests/test_channel_subscribe_upstream.py

@@ -0,0 +1,69 @@
+"""Subscribing when YouTube says no — and when "no" means "you already are".
+
+The error no longer carries YouTube's body (it is the request quoted back, and it went into a log
+line and a 422), so "already following" is recognised by the reason code and nothing else.
+DB lane.
+"""
+
+import pytest
+from fastapi import HTTPException
+from sqlalchemy import select
+
+from app.models import Channel, Subscription
+from app.routes import channels
+from app.youtube.client import YouTubeError
+
+
+def _youtube_refusing(reason: str):
+    class _Client:
+        def __init__(self, db, user):
+            pass
+
+        def __enter__(self):
+            return self
+
+        def __exit__(self, *exc):
+            return False
+
+        def insert_subscription(self, channel_id):
+            raise YouTubeError(f"POST subscriptions -> 400: {reason}", reason=reason)
+
+        def get_channels(self, ids):
+            return []
+
+    return _Client
+
+
+@pytest.fixture
+def stub_channel(db):
+    db.add(Channel(id="UCdup", title="Already followed"))
+    db.commit()
+    return "UCdup"
+
+
+@pytest.mark.parametrize("reason", ["subscriptionDuplicate", "quotaExceeded,subscriptionDuplicate"])
+def test_already_following_on_youtube_is_recorded_not_refused(
+    db, make_user, monkeypatch, stub_channel, reason
+):
+    user = make_user()
+    monkeypatch.setattr(channels, "YouTubeClient", _youtube_refusing(reason))
+
+    channels.subscribe(stub_channel, user=user, db=db)
+
+    sub = db.execute(
+        select(Subscription).where(
+            Subscription.user_id == user.id, Subscription.channel_id == stub_channel
+        )
+    ).scalar_one()
+    assert sub is not None
+
+
+@pytest.mark.parametrize("reason", ["quotaExceeded", "<none>", "<unparseable>"])
+def test_any_other_refusal_is_answered_with_a_422(db, make_user, monkeypatch, stub_channel, reason):
+    user = make_user()
+    monkeypatch.setattr(channels, "YouTubeClient", _youtube_refusing(reason))
+
+    with pytest.raises(HTTPException) as caught:
+        channels.subscribe(stub_channel, user=user, db=db)
+
+    assert caught.value.status_code == 422

backend/tests/test_failure_lines.py

@@ -0,0 +1,220 @@
+"""E22.5 review round 1: a failure is reported once, with its code, and a repeating one is not a
+flood.
+
+Each case here is a place where the first sweep got the line wrong in one of two directions: a
+failure that left no line at all (a download the watchdog killed), or one that left too many (the
+same edit failure twice, a warning per segment request, per settings read).
+"""
+
+import json
+import logging
+from pathlib import Path
+from types import SimpleNamespace
+
+import pytest
+import yt_dlp
+
+from app import logs, sysconfig, worker
+from app.downloads import pipeline
+from app.plex import stream as plex_stream
+
+
+class _Sink(logging.Handler):
+    def __init__(self):
+        super().__init__(level=logging.NOTSET)
+        self.lines: list[str] = []
+        self.setFormatter(logs.JsonFormatter())
+        self.addFilter(logs.ContextFilter())
+
+    def emit(self, record):
+        self.lines.append(self.format(record))
+
+    def objects(self) -> list[dict]:
+        return [json.loads(line) for line in self.lines]
+
+    def events(self, code: str) -> list[dict]:
+        return [o for o in self.objects() if o.get("event") == code]
+
+
+@pytest.fixture
+def sink():
+    logger = logging.getLogger("siftlode")
+    handler = _Sink()
+    previous = logger.level
+    logger.addHandler(handler)
+    logger.setLevel(logs.TRACE)
+    try:
+        yield handler
+    finally:
+        logger.removeHandler(handler)
+        logger.setLevel(previous)
+
+
+class _NoDb:
+    """A session that finds nothing and commits nothing — these cases are about the line."""
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, *exc):
+        return False
+
+    def get(self, *a, **k):
+        return None
+
+    def commit(self):
+        pass
+
+
+@pytest.fixture
+def finalized(monkeypatch):
+    """Whether `finalize_if_running` fires: True unless a case says a route got there first."""
+    state = {"fires": True}
+    monkeypatch.setattr(worker, "SessionLocal", _NoDb)
+    monkeypatch.setattr(pipeline, "SessionLocal", _NoDb)
+    fire = lambda db, job_id, **fields: state["fires"]  # noqa: E731
+    monkeypatch.setattr(worker.service, "finalize_if_running", fire)
+    return state
+
+
+# --- the worker's own failures ---------------------------------------------------------------
+@pytest.mark.parametrize("cause", ["stalled", "timeout"])
+def test_a_download_the_watchdog_kills_leaves_a_coded_line(sink, finalized, cause):
+    """The child is killed, so it never reports; before this the worker did not either, and a
+    stalled download failed with no line anywhere."""
+    worker._fail_running(42, error="stopped", error_code="network", cause=cause)
+
+    (line,) = sink.events("download.job.failed")
+    assert line["level"] == "WARNING"
+    assert line["attrs"] == {"job_id": 42, "error_code": "network", "cause": cause}
+
+
+def test_a_job_a_route_already_stopped_is_not_reported_as_failed(sink, finalized):
+    finalized["fires"] = False
+    worker._fail_running(42, error="stopped", error_code="network", cause="stalled")
+
+    assert sink.events("download.job.failed") == []
+
+
+def test_a_caller_with_its_own_line_is_not_reported_twice(sink, finalized):
+    worker._fail_running(42, error="Worker error", error_code=None)
+
+    assert sink.events("download.job.failed") == []
+
+
+# --- an edit failure, once ---------------------------------------------------------------------
+def test_an_edit_failure_is_one_line_with_its_code_and_our_own_stack(sink, finalized):
+    try:
+        raise TypeError("a spec shape nobody expected")
+    except TypeError as exc:
+        pipeline._fail_edit(9, 3, str(exc), exc)
+
+    (line,) = sink.events("edit.job.failed")
+    assert line["attrs"] == {"job_id": 9, "asset_id": 3, "error_type": "TypeError"}
+    assert "TypeError" in line["exc"]
+    assert len([o for o in sink.objects() if o["level"] in ("WARNING", "ERROR")]) == 1
+
+
+def test_an_edit_that_ffmpeg_refused_names_no_file_and_carries_no_stack(sink, finalized):
+    """ffmpeg's stderr tail names its input, whose path is the channel and the title. The message
+    is scrubbed; the stack is left off, because its last line is the same text unscrubbed."""
+    msg = "ffmpeg failed (1): /downloads/Some_Channel/Season_2024/A_Private_Title_a3.mp4: Invalid data"
+    pipeline._fail_edit(9, 3, msg, RuntimeError(msg))
+
+    (line,) = sink.events("edit.job.failed")
+    assert "A_Private_Title" not in json.dumps(line)
+    assert "exc" not in line
+
+
+def test_an_edit_whose_source_is_gone_is_found_by_the_same_code(sink, finalized):
+    pipeline._fail_edit(9, 3, "The source file is missing.")
+
+    (line,) = sink.events("edit.job.failed")
+    assert line["attrs"] == {"job_id": 9, "asset_id": 3}
+
+
+# --- a download failure, scrubbed ---------------------------------------------------------------
+def test_a_failed_download_line_does_not_put_back_what_the_adapter_took_out(sink):
+    """yt-dlp logs its error through the adapter (scrubbed) and then raises the same text, which
+    the pipeline logs again. Unscrubbed, that second line is the URL the first one hid."""
+    msg = "ERROR: Unsupported URL: https://example.org/private/path?id=123"
+    pipeline._log_download_failed(5, msg, "unknown", yt_dlp.utils.DownloadError(msg))
+
+    (line,) = sink.events("download.job.failed")
+    assert "private/path" not in json.dumps(line)
+    assert "example.org/…" in line["msg"]
+    # Unclassified, but yt-dlp's own error: its stack is the `except`, its foot the raw text.
+    assert "exc" not in line
+
+
+def test_an_unclassified_failure_of_our_own_keeps_its_stack(sink):
+    try:
+        raise KeyError("requested_formats")
+    except KeyError as exc:
+        pipeline._log_download_failed(5, str(exc), "unknown", exc)
+
+    (line,) = sink.events("download.job.failed")
+    assert "KeyError" in line["exc"]
+    assert line["attrs"]["error_type"] == "KeyError"
+
+
+def test_the_metadata_backfill_names_the_asset_and_not_the_download(sink, monkeypatch):
+    from app.downloads import adapters, backfill
+
+    url = "https://www.youtube.com/watch?v=dQw4w9WgXcQ"
+
+    class _Refusing:
+        def __init__(self, opts):
+            pass
+
+        def __enter__(self):
+            return self
+
+        def __exit__(self, *exc):
+            return False
+
+        def extract_info(self, target, download):
+            raise yt_dlp.utils.DownloadError(
+                f"ERROR: [youtube] dQw4w9WgXcQ: Video unavailable ({target})"
+            )
+
+    monkeypatch.setattr(backfill.yt_dlp, "YoutubeDL", _Refusing)
+    assert backfill._extract_meta(url, adapters.YouTubeAdapter(), 11) is None
+
+    lines = [o for o in sink.objects() if "meta extract failed" in o["msg"]]
+    assert lines and all("dQw4w9WgXcQ" not in json.dumps(o) for o in lines)
+
+
+# --- a repeating failure, once ------------------------------------------------------------------
+def test_a_stale_setting_is_reported_once_not_on_every_read(sink, monkeypatch):
+    spec = next(s for s in sysconfig.SPECS if s.type == "int")
+    monkeypatch.setattr(sysconfig, "_raw_override", lambda db, key: "not-a-number")
+    monkeypatch.setattr(sysconfig, "_coerce_warned", set())
+
+    for _ in range(3):
+        sysconfig.get(None, spec.key)
+
+    assert len(sink.events("sysconfig.override_uncoercible")) == 1
+
+
+class _RefusingDir:
+    """A session directory on a disk that has gone away: every probe raises."""
+
+    def __truediv__(self, name):
+        return SimpleNamespace(exists=self._refuse)
+
+    @staticmethod
+    def _refuse():
+        raise PermissionError("the disk went read-only")
+
+
+def test_a_refusing_session_directory_is_reported_once_per_session(sink):
+    """Every segment request probes, and hls.js retries each 404 — a line per probe is a flood."""
+    session = plex_stream.HlsSession("k", Path("."), None, 0.0, ())
+    session.dir = _RefusingDir()
+
+    for seq in (3, 4, 5):
+        assert not plex_stream.artifact_may_arrive(session, plex_stream.segment_name(seq, None))
+
+    (line,) = sink.events("plex.stream.segment_probe_failed")
+    assert line["attrs"]["reason"] == "PermissionError"

backend/tests/test_log_pii.py

@@ -87,15 +87,62 @@ def test_an_upstream_failure_logs_the_reason_code_and_not_the_body(sink):
     assert err.reason == "quotaExceeded"
 
 
-def test_the_body_is_kept_for_the_one_user_who_consented_to_it(sink):
+@pytest.fixture
+def consented():
+    """The request scope of a user who turned Teljes (TRACE) on for themselves."""
+    scope = logs._RequestScope()
+    scope.user_id, scope.level = 7, logs.TRACE
+    token = logs._scope.set(scope)
+    try:
+        yield
+    finally:
+        logs._scope.reset(token)
+
+
+def test_the_body_is_kept_for_the_one_user_who_consented_to_it(sink, consented):
     _upstream_failure("GET", "search", _youtube_403(TERM))
 
-    # TRACE is emitted only while the user has turned Teljes on for their own session, which is
-    # exactly the consent the privacy policy describes. The diagnostic is not lost, it is gated.
+    # The diagnostic is not lost, it is gated — on the consent the privacy policy describes.
     (trace,) = sink.at("TRACE")
     assert TERM in trace["attrs"]["body"]
 
 
+def test_a_process_running_at_trace_is_not_anyones_consent(sink):
+    """`sink` runs the whole tree at TRACE — what an admin setting `log_level` to TRACE does. That
+    buys more detail, not every user's search terms quoted back (`logs._traced`)."""
+    _upstream_failure("GET", "search", _youtube_403(TERM))
+
+    assert sink.at("TRACE") == []
+    assert TERM not in json.dumps(sink.objects())
+
+
+def test_a_scope_with_no_elevation_is_not_consent():
+    scope = logs._RequestScope()
+    scope.user_id = 7
+    token = logs._scope.set(scope)
+    try:
+        assert not logs.consented_trace()
+    finally:
+        logs._scope.reset(token)
+    assert not logs.consented_trace()  # a scheduler thread: no request at all
+
+
+def test_a_refused_token_refresh_names_the_oauth_reason_and_not_its_prose(sink):
+    """The token endpoint's error is `{"error": "<code>", "error_description": <prose>}` — not the
+    Data API's shape. It goes through the same door, so it gets the same treatment."""
+    resp = httpx.Response(
+        400,
+        json={"error": "invalid_grant", "error_description": "Token has been expired or revoked."},
+        request=httpx.Request("POST", "https://oauth2.googleapis.com/token"),
+    )
+    err = _upstream_failure("POST", "oauth2/token", resp)
+
+    (line,) = sink.at("WARNING")
+    assert line["attrs"]["reason"] == "invalid_grant"
+    assert err.reason == "invalid_grant"
+    assert "revoked" not in str(err) and "revoked" not in json.dumps(line)
+
+
 def test_an_unparseable_body_still_names_the_status(sink):
     """A proxy answering HTML has no `reason` to take — the line must still be worth reading, and
     must still not fall back to printing the body."""
@@ -153,9 +200,16 @@ _BANNED_NAMES = frozenset(
         "source_url",
         "source_webpage_url",
         "password",
+        # A download's path is built from its channel and its title (`storage.rel_path`), so it
+        # names what someone downloaded exactly as the URL does. Log the asset id.
+        "rel",
+        "rel_path",
+        "title",
     }
 )
-_BANNED_ATTRS = frozenset({"text", "content", "email", "url", "query"})
+_BANNED_ATTRS = frozenset(
+    {"text", "content", "email", "url", "query", "rel_path", "title", "display_name"}
+)
 _ESCAPE = "log-pii-ok"
 _LOG_METHODS = frozenset(
     {"debug", "info", "warning", "warn", "error", "exception", "critical", "log"}
@@ -211,3 +265,37 @@ def test_no_log_call_passes_what_a_person_typed():
         f"the argument (or the line directly above the call) `# {_ESCAPE}: <why it is "
         f"safe>` — {offenders}"
     )
+
+
+def _guarded_by_consent(node: ast.AST, parents: dict[ast.AST, ast.AST]) -> bool:
+    while node in parents:
+        node = parents[node]
+        if isinstance(node, ast.If):
+            for sub in ast.walk(node.test):
+                if isinstance(sub, ast.Attribute) and sub.attr == "consented_trace":
+                    return True
+    return False
+
+
+def test_every_escaped_trace_line_is_gated_on_the_users_own_consent():
+    """`log-pii-ok` on a TRACE line means "this is what the user's Teljes consent is for". The TRACE
+    level alone does not hold that promise: an admin can run the whole process at TRACE, and a
+    scheduler thread has no user at all. So the line must sit under `logs.consented_trace()`."""
+    root = pathlib.Path(__file__).resolve().parents[1] / "app"
+    offenders = []
+    for path in sorted(root.rglob("*.py")):
+        src = path.read_text(encoding="utf-8")
+        lines = src.splitlines()
+        tree = ast.parse(src)
+        parents = {child: node for node in ast.walk(tree) for child in ast.iter_child_nodes(node)}
+        for node in ast.walk(tree):
+            if not _is_log_call(node) or not _offending(node):
+                continue
+            first = node.args[0] if node.args else None
+            at_trace = isinstance(first, ast.Attribute) and first.attr == "TRACE"
+            span = " ".join(lines[max(0, node.lineno - 2) : (node.end_lineno or node.lineno)])
+            if at_trace and _ESCAPE in span and not _guarded_by_consent(node, parents):
+                offenders.append(f"{path.relative_to(root)}:{node.lineno}")
+    assert offenders == [], (
+        f"a TRACE line with user content is not under consented_trace(): {offenders}"
+    )

backend/tests/test_logger_names.py

@@ -13,6 +13,8 @@ the rule, because the rule is only worth having if a hand-rolled name fails the
 import ast
 import logging
 import pathlib
+import sys
+import types
 
 import pytest
 
@@ -29,8 +31,8 @@ _ESCAPE = "logger-name-ok"
         ("app.plex.stream", "siftlode.plex.stream"),
         ("app.routes.channels", "siftlode.routes.channels"),
         ("app.worker", "siftlode.worker"),
-        # Not a module of ours (a script, a doctest): it lands on the root of our tree rather than
-        # inventing `siftlode.__main__`, which nothing would think to filter on.
+        # A script that is not a module of ours (pytest itself, here): it lands on the root of our
+        # tree rather than inventing `siftlode.__main__`, which nothing would think to filter on.
         ("__main__", "siftlode"),
     ],
 )
@@ -38,6 +40,32 @@ def test_the_name_follows_the_module(module, expected):
     assert logs.get_logger(module).name == expected
 
 
+@pytest.mark.parametrize(
+    ("spec_name", "expected"),
+    [
+        # How the worker and the download child actually run: `python -m app.worker`, and the
+        # worker's own `[sys.executable, "-m", "app.downloads.runner", ...]`. `__name__` is
+        # `__main__` in both, and taken at its word both processes logged as the bare root.
+        ("app.worker", "siftlode.worker"),
+        ("app.downloads.runner", "siftlode.downloads.runner"),
+    ],
+)
+def test_a_module_run_with_dash_m_keeps_its_own_name(monkeypatch, spec_name, expected):
+    main = types.ModuleType("__main__")
+    main.__spec__ = types.SimpleNamespace(name=spec_name)
+    monkeypatch.setitem(sys.modules, "__main__", main)
+
+    assert logs.get_logger("__main__").name == expected
+
+
+def test_a_plain_script_has_no_spec_and_stays_the_root(monkeypatch):
+    main = types.ModuleType("__main__")
+    main.__spec__ = None
+    monkeypatch.setitem(sys.modules, "__main__", main)
+
+    assert logs.get_logger("__main__").name == "siftlode"
+
+
 def test_every_derived_logger_stays_under_the_one_tree():
     """`configure_logging` hangs the level and the handler off the root, and the tests attach their
     sink to `siftlode` — a name outside that tree would be invisible to both."""
@@ -62,11 +90,25 @@ def test_no_module_hand_rolls_a_logger_name():
             if not isinstance(node, ast.Call):
                 continue
             func = node.func
-            if not (isinstance(func, ast.Attribute) and func.attr == "getLogger"):
-                continue
-            # A no-argument `logging.getLogger()` is the ROOT logger — level plumbing, not a
-            # module's own logger, and nothing to name.
-            if not node.args:
+            called = (
+                func.attr
+                if isinstance(func, ast.Attribute)
+                else func.id
+                if isinstance(func, ast.Name)
+                else ""
+            )
+            if called == "get_logger":
+                # The derivation itself is only as good as its argument: a literal here does not
+                # name a module, it collapses to the root and pools modules together again.
+                arg = node.args[0] if len(node.args) == 1 else None
+                if isinstance(arg, ast.Name) and arg.id == "__name__" and not node.keywords:
+                    continue
+            elif called == "getLogger":
+                # A no-argument `logging.getLogger()` is the ROOT logger — level plumbing, not a
+                # module's own logger, and nothing to name. A `name=` keyword is still a name.
+                if not node.args and not node.keywords:
+                    continue
+            else:
                 continue
             start = max(0, node.lineno - 2)  # the call's lines, plus the one above it
             span = " ".join(lines[start : (node.end_lineno or node.lineno)])

backend/tests/test_logsink.py

@@ -90,6 +90,40 @@ def test_a_flood_of_one_message_is_stored_up_to_the_limit_and_then_summarised(st
     assert notice.attrs["template"] == "RSS poll skipped for channel %s: HTTP 404"
 
 
+def test_a_pass_through_template_groups_by_its_text_not_by_the_bare_placeholder(stored):
+    """`"%s"` is how the yt-dlp adapter hands over yt-dlp's own line, and as a key it says nothing:
+    every distinct warning would share one bucket and the thirty-first — whatever it said — would be
+    dropped as a "repeat" of the first. The text groups instead, numbers folded, so a retry storm is
+    still one repeat and the warning that explains the failure still gets through."""
+    sink, log, rows = stored
+    limit = logsink.REPEATS_PER_MINUTE
+    for i in range(limit + 10):
+        log.warning("%s", f"[download] Got error: timed out. Retrying fragment {i} (1/10)...")
+    log.warning("%s", "[youtube] …: Requested format is not available")
+    stored_rows = rows()
+    assert len(stored_rows) == limit + 1
+    assert stored_rows[-1].msg == "[youtube] …: Requested format is not available"
+    sink._repeats.started -= logsink.REPEAT_WINDOW_SECONDS
+    notice = rows()[-1]
+    assert notice.attrs["suppressed"] == 10
+    assert notice.attrs["template"].startswith(
+        "[download] Got error: timed out. Retrying fragment #"
+    )
+
+
+def test_a_process_that_ends_inside_the_minute_still_says_what_it_dropped(stored):
+    """Every download child lives well under the limiter's minute. Its last batch closes the window
+    early, or the repeats it dropped would go unreported."""
+    sink, log, rows = stored
+    for i in range(logsink.REPEATS_PER_MINUTE + 3):
+        log.warning("RSS poll skipped for channel %s: HTTP 404", f"UC{i}")
+    rows()
+    sink.write_pending(final=True)
+    notice = rows()[-1]
+    assert notice.event == "log.repeats_suppressed"
+    assert notice.attrs["suppressed"] == 3
+
+
 def test_request_lines_are_never_treated_as_a_flood(stored):
     _sink, log, rows = stored
     for _ in range(logsink.REPEATS_PER_MINUTE + 5):

backend/tests/test_ytdlp_logging.py

@@ -59,7 +59,10 @@ def sink():
         ("debug", "TRACE"),
         ("info", "DEBUG"),
         ("warning", "WARNING"),
-        ("error", "ERROR"),
+        # yt-dlp calls `error` from `trouble()` just before it raises, on every failed ATTEMPT —
+        # the pipeline's fallback client set then often succeeds. The job's verdict is the
+        # pipeline's `download.job.failed`, not this.
+        ("error", "WARNING"),
     ],
 )
 def test_each_ytdlp_level_lands_where_it_belongs(sink, method, level):
@@ -83,6 +86,45 @@ def test_the_source_url_does_not_travel_with_the_explanation(sink):
     assert "Requested format is not available" in line["msg"]
 
 
+@pytest.mark.parametrize(
+    ("line", "kept", "gone"),
+    [
+        # The shape nearly every yt-dlp warning and error has: no URL, the id as a prefix.
+        (
+            "ERROR: [youtube] dQw4w9WgXcQ: Video unavailable",
+            "ERROR: [youtube] …: Video unavailable",
+            "dQw4w9WgXcQ",
+        ),
+        # The generic extractor's "id" is the URL's own basename.
+        (
+            "WARNING: [generic] my-private-holiday-clip: Falling back on generic information",
+            "WARNING: [generic] …: Falling back on generic information",
+            "my-private-holiday-clip",
+        ),
+        # `Destination` is yt-dlp's word, not an id; the path after it is built from the title.
+        (
+            "[download] Destination: /downloads/Some_Channel/Season_2024/A_Private_Title_a7.mp4",
+            "[download] Destination: …",
+            "A_Private_Title",
+        ),
+    ],
+)
+def test_what_names_the_download_comes_out_of_real_ytdlp_shapes(sink, line, kept, gone):
+    formats._YDL_LOG.warning(line)
+
+    (obj,) = sink.objects()
+    assert obj["msg"] == kept
+    assert gone not in json.dumps(obj)
+
+
+def test_a_download_error_is_scrubbed_the_same_way_wherever_it_is_logged():
+    """yt-dlp raises the text it has just logged, and the pipeline logs the exception again one
+    line later (`download.job.failed`). The same scrub, or that line puts the URL back."""
+    text = "ERROR: Unsupported URL: https://example.org/private/path?id=123"
+
+    assert formats.log_safe(text) == "ERROR: Unsupported URL: example.org/…"
+
+
 def test_a_line_with_no_url_is_left_alone(sink):
     formats._YDL_LOG.warning("nsig extraction failed: Some formats may be missing")