archilyzer, branch HEAD

Merge r20/whisper-fix (Clear failure list keeps its log after the list empties) into r20/integration

2026-10-10T07:10:44Z

commit 974fa4e5dd1cc68677bc91f731e9e909e68c6b09 parent 1c92a17816b8a5e883e47936b3c68fb6d11493fb Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 03:10:44 -0400 Merge r20/whisper-fix (Clear failure list keeps its log after the list empties) into r20/integration Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 20 — the whisper red, as it went (Clear failure list unmounted its own log)

2026-10-10T07:10:13Z

commit ba8abad51eabeae65cd963646bcbb63beba562c6 parent 184d1c298517d557d0e740653f5d88708c5b3a59 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 03:10:13 -0400 plans: release 20 — the whisper red, as it went (Clear failure list unmounted its own log) Co-Authored-By: Claude Opus 5.5 (1M context)

channels: "Clear failure list" keeps its log after the list empties

2026-10-10T07:10:13Z

commit 184d1c298517d557d0e740653f5d88708c5b3a59 parent 1c92a17816b8a5e883e47936b3c68fb6d11493fb Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 03:10:13 -0400 channels: "Clear failure list" keeps its log after the list empties The section rendered its clear controls only while the failed list was non-empty, so the refresh StreamActionLog fires at the end of its own run unmounted the log with "Cleared N entries" in it. A `ran` flag set on click keeps the block for the page's life, the button disabled at (0) — the rule in FACTS.md, "A run log lives in the panel's React state", and DiagnosticsStage's pattern. whisper.spec.ts:184 failed 3/3 under next start. Co-Authored-By: Claude Opus 5.5 (1M context)

Merge main (cf162d7c: umtool song asr-via-ops batching) into r20/integration

2026-10-10T07:01:18Z

commit 1c92a17816b8a5e883e47936b3c68fb6d11493fb parent ea1e503cadc761a7e5788f44689c863480c7b823 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 03:01:18 -0400 Merge main (cf162d7c: umtool song asr-via-ops batching) into r20/integration Co-Authored-By: Claude Opus 5.5 (1M context)

Merge r20/a6-a9 (release 19 A6–A9: archive.org imports and remote listings, cues without an index, publish edges, MCP archival writes) into r20/integration

2026-10-10T07:01:18Z

commit ea1e503cadc761a7e5788f44689c863480c7b823 parent 458e992307d0303c995846ba73fc8fee2a36ae1b Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 03:01:18 -0400 Merge r20/a6-a9 (release 19 A6–A9: archive.org imports and remote listings, cues without an index, publish edges, MCP archival writes) into r20/integration Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 19 A6–A9, as shipped (Track A of the overnight batch)

2026-10-10T06:59:26Z

commit 4d36f8e7222a9c4940ee2f1f31cc50624d58c7a8 parent dcf45f12e40614b846e2cc335d15823058fc342c Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 02:59:26 -0400 plans: release 19 A6–A9, as shipped (Track A of the overnight batch) The four slices' records under Track A, tonight's gates appended to the Track A gates, and the editor changelog's [Unreleased] bullets. Co-Authored-By: Claude Opus 5.5 (1M context)

mcp: archival tools through the editor — get_job, enqueue, channel_coverage, notes (release 19 A9)

2026-10-10T06:50:47Z

commit dcf45f12e40614b846e2cc335d15823058fc342c parent 56a52ba8c959611cae4f09c8b7fa5dcad44bf37f Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 02:50:47 -0400 mcp: archival tools through the editor — get_job, enqueue, channel_coverage, notes (release 19 A9) On the fetch_clip pattern (editor URL + WORKER_TOKEN; the editor writes), each tool one existing route: get_job reads GET /api/ops/job/ with its log tail; enqueue posts sync, download-missing, retry-bucket (chosen ids), transcribe-bucket, fetch-posts or import-video and answers the job id; channel_coverage reads the new GET /api/ops/coverage (`pnpm ops get coverage `), which dates every held video off disk with coverageDate — the recorded date when the channel has a title rule (release 20 D1), else the upload date — and reports per year and month and every gap past gapDays. notes reads umtool (UMTOOL_URL): list from /api/browse/decisions, read from /api/notes/context; a reply answers with the `umtool notes reply` command, since umtool's HTTP route records every write as the operator's. Settings, storage and deletes stay CLI/ops-only. The plans' instructions gain the archive-work step; README tool table, OPERATING, ENVIRONMENT.md (UMTOOL_URL now read by the MCP) and COMMANDS.md follow. Co-Authored-By: Claude Opus 5.5 (1M context)

Merge r20/integration (release 20 D1 recorded dates) into r20/a6-a9

2026-10-10T06:39:21Z

commit 56a52ba8c959611cae4f09c8b7fa5dcad44bf37f parent b4322ab1cd3b4f67e0ed1ec7a4d4ccec5a1c2218 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 02:39:21 -0400 Merge r20/integration (release 20 D1 recorded dates) into r20/a6-a9 Co-Authored-By: Claude Opus 5.5 (1M context)

publish: the ops build verb takes allowMissingMedia; `publish build --out ` (release 19 A8)

2026-10-10T06:39:16Z

commit b4322ab1cd3b4f67e0ed1ec7a4d4ccec5a1c2218 parent 2baea15e823029add6e25dc281eaaba8575c09c9 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 02:39:16 -0400 publish: the ops build verb takes allowMissingMedia; `publish build --out ` (release 19 A8) POST /api/ops/publish {verb: "build", allowMissingMedia: true} carries the flag onto the queued build stage (`--allow-missing-media`, as `build site` already could), local builds only — the docker runner refuses it. `archilyzer publish build` takes --allow-missing-media too, and --out : after the build (or the no-op of a fresh one) the site's bundle is laid out in for a server that is not Pages — hard links into the bundle where the filesystem allows (a build installs a new bundle, never edits one), else copies; the destination is emptied first and refused, before any build, where a local deploy's would be. Not a deploy: nothing recorded. The publish lane over `pnpm ops lane` shipped in A4. Co-Authored-By: Claude Opus 5.5 (1M context)

ops: cues without an index — build-cues, get transcript, and the MCP's get_transcript falls back to the editor (release 19 A7)

2026-10-10T06:33:53Z

commit 2baea15e823029add6e25dc281eaaba8575c09c9 parent 6bd5722378cd6cf08623a8962f7a9582865c9269 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 02:33:53 -0400 ops: cues without an index — build-cues, get transcript, and the MCP's get_transcript falls back to the editor (release 19 A7) build-cues {slug, ids?, force?} writes transcript.cues.json through the digest card's Normalize action, narrowed to held ids (a stray refused, named) on the channel's queue. GET /api/ops/transcript?id[&slug] (`pnpm ops get transcript [--slug]`) reads one video's cues off disk with no index and writes nothing: a fresh cues.json, else what normalize would write (buildNormalizedTranscript, factored out of normalizeTranscript), else the English VTT alone. The MCP's get_transcript, for a video its archive does not hold, asks that route through the configured editor (the fetch_clip pattern) and marks the answer as read off the editor's disk, with no moment links. Co-Authored-By: Claude Opus 5.5 (1M context)

ops: a help paragraph for each of the twelve actions that had none

2026-10-10T06:24:03Z

commit 6bd5722378cd6cf08623a8962f7a9582865c9269 parent 0baa0fa1e03f3442c4e848ae3e20c2b63f84ea09 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 02:24:03 -0400 ops: a help paragraph for each of the twelve actions that had none channel-priority, sync, download-missing, metadata-scan, import-video, feed-metadata, refresh-report, relocate-back, evict-clips, tags, tag-videos and keep-videos now open a paragraph in usage(), so COMMANDS.md's rows carry their bodies (release 19 C2's found-and-left; relocate and lane gained theirs in A4). COMMANDS.md regenerated. Co-Authored-By: Claude Opus 5.5 (1M context)

ops: archive.org imports take many items or a search; Odysee/BitChute remote listings (release 19 A6)

2026-10-10T06:22:49Z

commit 0baa0fa1e03f3442c4e848ae3e20c2b63f84ea09 parent 8ccd3dc569453445937df267bcc0f76c0a1ba98a Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 02:22:49 -0400 ops: archive.org imports take many items or a search; Odysee/BitChute remote listings (release 19 A6) import-archive-org takes {item} as before, {items: [...]} or {query, limit?} (archive.org's advanced search, one request through the polite client) as one job: the download gap and failure count shared across the batch, a jittered pause between items, records already held (on disk or in the saved-video store) skipped, files archive.org marks not for download listed as RESTRICTED by a dry run and skipped by an import, a 401/403 skipping the rest of its item and three such items in a row stopping the job with the platform backed off. The log ends with a summary: line. remote-listing {slug} (and `pnpm ops get remote-listing `, waited on, the result on stdout) lists an Odysee or BitChute channel with one flat-playlist read on the platform's queue, refused while the platform is held or cooling down, after its import floor, backing it off on a 429, and answers what it lists that is not held and what is held but not listed. Nothing is written. Co-Authored-By: Claude Opus 5.5 (1M context)

Merge r20/d2-r20 part 2 (release 20 D3 verified closed, D1 recorded dates; D2 Twitch ids held on r20/twitch-ids) into r20/integration

2026-10-10T06:05:52Z

commit 458e992307d0303c995846ba73fc8fee2a36ae1b parent 8ccd3dc569453445937df267bcc0f76c0a1ba98a Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 02:05:52 -0400 Merge r20/d2-r20 part 2 (release 20 D3 verified closed, D1 recorded dates; D2 Twitch ids held on r20/twitch-ids) into r20/integration Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 20 D2 held for an operator ruling — the two options, the readers that miss the join; changelog and STATE corrected

2026-10-10T06:05:07Z

commit 24ce786e21fd861d128338b8b9089ad15f6d685e parent a0accf89f61b6f636e5c4b57ddd8f84f7c014be9 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 02:05:07 -0400 plans: release 20 D2 held for an operator ruling — the two options, the readers that miss the join; changelog and STATE corrected Co-Authored-By: Claude Opus 5.5 (1M context)

Merge r20/e2e-speed (e2e speed S1–S3: start mode by default with a build stamp, a timing record per run, the slow timers behind test knobs) into r20/integration

2026-10-10T06:02:29Z

commit 8ccd3dc569453445937df267bcc0f76c0a1ba98a parent d77b126fc1a8eb9333c998d5aedbc8fdee9a70cb Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 02:02:29 -0400 Merge r20/e2e-speed (e2e speed S1–S3: start mode by default with a build stamp, a timing record per run, the slow timers behind test knobs) into r20/integration Co-Authored-By: Claude Opus 5.5 (1M context) # Conflicts: # editor/CHANGELOG.md

common: revert release 20 D2 — held for an operator ruling (it changes published Twitch URLs); kept on r20/twitch-ids

2026-10-10T06:01:48Z

commit a0accf89f61b6f636e5c4b57ddd8f84f7c014be9 parent c084840a709ae5cac07384acab33eba9cb09e978 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 02:01:48 -0400 common: revert release 20 D2 — held for an operator ruling (it changes published Twitch URLs); kept on r20/twitch-ids This reverts commit 21c44a80. Co-Authored-By: Claude Opus 5.5 (1M context)

plans: e2e speed S1–S3, as shipped (Track E)

2026-10-10T06:00:53Z

commit 505340fd59f5fea3ccf5adfe0f033ad43296b0c5 parent b8c2811d82500d2cd7c78c13a0c41a8e3aa1cc6b Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 02:00:53 -0400 plans: e2e speed S1–S3, as shipped (Track E) The record under plans/e2e-speed.md "As it went", FACTS amended (start mode is the default and its build is stamped), and the editor changelog's [Unreleased] bullet. Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 20 D3, D1 and D2, as shipped (the caption-track bug closed, recorded dates, one Twitch id)

2026-10-10T05:59:50Z

commit c084840a709ae5cac07384acab33eba9cb09e978 parent 21c44a807157a99a4e56de7ccb37f6b92a2ae44f Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 01:59:50 -0400 plans: release 20 D3, D1 and D2, as shipped (the caption-track bug closed, recorded dates, one Twitch id) Co-Authored-By: Claude Opus 5.5 (1M context)

umtool song: asr-via-ops batches every window into one ops transcribe job

2026-10-10T05:56:52Z

commit cf162d7cbf289f194f87418af2404faad2e663a0 parent 7f91177c75088089fdf8556b05ca9af26d5a5c89 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 01:56:52 -0400 umtool song: asr-via-ops batches every window into one ops transcribe job One window per job waited ~30 min each for a worker behind the auto lane, so the MK plans' 841 windows would have taken days. The windows are now cut at 16 kHz and laid end to end with 2 s of silence between them, transcribed in one "words": true job (ASR_WORKER names the worker), and each word mapped back to its source by the window it falls in. Co-Authored-By: Claude Opus 5.5 (1M context)

e2e: the editor config names the stamp script's dist dir it mirrors

2026-10-10T05:45:45Z

commit b8c2811d82500d2cd7c78c13a0c41a8e3aa1cc6b parent 0bdab05ed797d9f7076e06e2f271669835bbecf3 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 01:45:45 -0400 e2e: the editor config names the stamp script's dist dir it mirrors Co-Authored-By: Claude Opus 5.5 (1M context)

e2e: the four slow timers on a test clock (e2e speed S3)

2026-10-10T05:31:20Z

commit 0bdab05ed797d9f7076e06e2f271669835bbecf3 parent 7517df525fa5343fbc6d3e08034e20e7b10fd560 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 01:31:20 -0400 e2e: the four slow timers on a test clock (e2e speed S3) Two are product timers and take a test-only duration from the editor's test server (E2E_SERVER_ENV), declared in envVars.ts: - E2E_BACKOFF_BASE_MS (20 s; 60 s in production): the first rate-limit cooldown, read in platformBackoff.ts's nextBackoff. The cap and the hold arithmetic keep the real constants. pacing.spec still watches it lapse. - E2E_CLIP_WINDOW_GAP_MS (2 s; 30–45 s): the pause between two clip-window fetches, read in fetchWindows.ts. fetch-window.spec now asserts the one pause the batch owes, from the job log. Two were never product timers: publish-lane.spec's 45 s and channel-storage .spec's 25 s were the stuck-job holder's own release timer, sized for a cold dev server. The route takes `release=` (finished detached, as releaseAfterMs is), and the specs let the holder go once they have seen what it holds. The stamp also stops counting the changelogs, read at run time only. Before/after, start mode, per test: pacing:195 66 s -> 22 s; publish-lane:78 50 s -> 6 s; fetch-window:337 34 s -> 3 s; channel-storage:872 28 s -> 3 s. Co-Authored-By: Claude Opus 5.5 (1M context)

common: release 20 D2 — one Twitch id: records carry the canonical their directory does, not the native v

2026-10-10T05:11:15Z

commit 21c44a807157a99a4e56de7ccb37f6b92a2ae44f parent 0f614b3a43376d07d562101aae4c83424749ccd3 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 01:11:15 -0400 common: release 20 D2 — one Twitch id: records carry the canonical their directory does, not the native v canonicalTwitchVideoId (lib/videoId.ts) is the one normalization: summarize uses it for a Twitch record, readNormalizedTranscript applies it to a transcript.cues.json written before (normalizeSummaryId), and the index re-processes once every record it holds under a v-id ("Twitch ids v1", the platform-labels shape), re-keying it. The Twitch player is still handed v. Co-Authored-By: Claude Opus 5.5 (1M context)

common: release 20 D1 — recorded dates for VOD-mirror channels, read off the title by a per-channel rule

2026-10-10T05:02:30Z

commit 0f614b3a43376d07d562101aae4c83424749ccd3 parent c3890608162a0f674ecb611034f333fc35200d28 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 01:02:30 -0400 common: release 20 D1 — recorded dates for VOD-mirror channels, read off the title by a per-channel rule A channel's config.json gains `recordedDate: {titlePattern}` (a case-insensitive regex with named groups year, month, day; month a number or an English month name; checked like the download filter's patterns). The index derives each record's `recordedDate` (YYYYMMDD) from its title — a real day no later than the upload date, else none — and re-derives a channel's records once when its rule changes (`recordedDateRule:` in the index meta). `coverageDate` reads it before the upload date. The Configure form and `ops channel-config` set it (`recordedDateTitlePattern`); CHANNEL.md regenerated. Omitted everywhere a channel has no rule, so no page changes until one is set. Co-Authored-By: Claude Opus 5.5 (1M context)

e2e: a timing record per run, and per-spec deltas against the branch's last (e2e speed S2)

2026-10-10T05:01:29Z

commit 7517df525fa5343fbc6d3e08034e20e7b10fd560 parent b6a859055fd100d0925a1c9895e764d77e567e83 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 01:01:29 -0400 e2e: a timing record per run, and per-spec deltas against the branch's last (e2e speed S2) Every suite's config writes Playwright's json report beside list (test-results*/timings.json). scripts/e2e-timings.mjs totals it per spec file, prints the files slowest first with each one's change against the last run on the same branch that ran it (the wall time against the last run of the same spec files), and records the run in the git common dir (e2e-timings// .json, the last 30), untracked and shared by every worktree. Co-Authored-By: Claude Opus 5.5 (1M context)

common: release 20 D3 — the en → 0 cues caption bug is closed; an index test for the fall-through, recorded in FACTS and STATE

2026-10-10T04:46:29Z

commit c3890608162a0f674ecb611034f333fc35200d28 parent dcc6501471c9d7f0cc6c65b79cd9df408732b774 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 00:46:29 -0400 common: release 20 D3 — the en → 0 cues caption bug is closed; an index test for the fall-through, recorded in FACTS and STATE An empty en-orig before an en with text, and an empty en before an en-US with text, through the real buildIndex. FACTS gains "The caption-track rule" and a superseded note on the old "fixture must carry timing tags" line; STATE closes the bug and names what is left (the stats cache does not see the caption-track pass's re-reads). Co-Authored-By: Claude Opus 5.5 (1M context)

Merge r20/d2-r20 part 1 (release 21 D2: clip windows cut from a saved container) into r20/integration

2026-10-10T04:43:54Z

commit d77b126fc1a8eb9333c998d5aedbc8fdee9a70cb parent 7f91177c75088089fdf8556b05ca9af26d5a5c89 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 00:43:54 -0400 Merge r20/d2-r20 part 1 (release 21 D2: clip windows cut from a saved container) into r20/integration Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 21 D2, as shipped (the saved-container tier for clip windows)

2026-10-10T04:43:16Z

commit dcc6501471c9d7f0cc6c65b79cd9df408732b774 parent fc175654d01b10966ba45fa5933e39f5b6c3b742 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 00:43:16 -0400 plans: release 21 D2, as shipped (the saved-container tier for clip windows) Co-Authored-By: Claude Opus 5.5 (1M context)

channels: release 21 D2 — a clip window of a saved video is cut from its container, not fetched

2026-10-10T04:42:46Z

commit fc175654d01b10966ba45fa5933e39f5b6c3b742 parent 7f91177c75088089fdf8556b05ca9af26d5a5c89 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 00:42:46 -0400 channels: release 21 D2 — a clip window of a saved video is cut from its container, not fetched fetchWindowAction asks the saved-video store after the clip-window cache and the in-flight jobs, before the network: a container that covers the window is cut locally (evidenceClip-server's evidenceCutArgs) into data//clips/ with the fetched sidecar shape plus `source: "saved-video"`, and answered `cached: true`. No pointer, or a container that ends before the window, goes to the network as before; a pointer whose container cannot be read (missing, an unmounted or unanswering store drive, a store mid-move, unreadable by ffprobe) is refused with the media guard's (or the store guard's) sentence. The fetch-windows batch puts covered windows in their own group, platform `saved-video` on `clips:saved-video` (no platform's hold, cooldown or gap); the controller cuts them with no request and no pause, and fails an unreadable one as `unreachable`. The video page marks a cut window. The MCP plan says a locally held video is cut without a fetch. Co-Authored-By: Claude Opus 5.5 (1M context)

e2e: start mode by default, with a build stamp that cannot be stale (e2e speed S1)

2026-10-10T04:13:49Z

commit b6a859055fd100d0925a1c9895e764d77e567e83 parent 7f91177c75088089fdf8556b05ca9af26d5a5c89 Author: I Mean I'm Just Saying Date: Sat, 10 Oct 2026 00:13:49 -0400 e2e: start mode by default, with a build stamp that cannot be stale (e2e speed S1) The editor and umtool suites run against `next start` unless E2E_MODE=dev. Before any server starts, scripts/e2e-stamp.mjs fingerprints what the build reads (index blob ids, and the working-tree bytes of every dirty or untracked file, over the package, common/ and the lockfile, minus specs and unit tests) and rebuilds through the heavy slot under the 5 GB cap when the stamp differs, saying why. The build goes to its own directory (editor .next/e2e via E2E_NEXT_DIST_DIR, umtool .next-e2e-start), never the .next a running app serves from. CI (the sharded image) keeps its own build. The export (default, hub) and homepage suites accept the switch and stay on next dev, saying so: they are static exports whose specs rewrite fixtures mid-run. The editor suite's export server stays next dev for the same reason. resetData retries its fixture copy on EEXIST (release 19's channel-work.spec:208). Co-Authored-By: Claude Opus 5.5 (1M context)

plans: overnight 2026-10-10 — the next batch run from the primary checkout (e2e speed, B2, A6–A9, D2 + R20, the TISM pilot)

2026-10-10T03:25:12Z

commit 7f91177c75088089fdf8556b05ca9af26d5a5c89 parent 6db33734b791dcb771046ecf0e47d56cdf37fa64 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 23:25:12 -0400 plans: overnight 2026-10-10 — the next batch run from the primary checkout (e2e speed, B2, A6–A9, D2 + R20, the TISM pilot) Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 19's rollout (editor, umtool, source mirror + homepage; sites unchanged)

2026-10-10T02:05:44Z

commit 6db33734b791dcb771046ecf0e47d56cdf37fa64 parent eee10ca9b043d5635f22f6f45732d636fc6795bb Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 22:05:44 -0400 plans: release 19's rollout (editor, umtool, source mirror + homepage; sites unchanged) Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 19's privacy line names no deployment hash

2026-10-10T01:15:52Z

commit eee10ca9b043d5635f22f6f45732d636fc6795bb parent ae08e4de35b4df2eaee9909c5fdeb615386669ce Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 21:15:52 -0400 plans: release 19's privacy line names no deployment hash Co-Authored-By: Claude Opus 5.5 (1M context)

plans: releases 19 and 21 — the landing's final gates (report 46/46, umtool 21/21 after release 22's fix round)

2026-10-10T01:15:39Z

commit ae08e4de35b4df2eaee9909c5fdeb615386669ce parent 194a70a2aeca8d5a3a4b6937c58eabda8a9fddda Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 21:15:39 -0400 plans: releases 19 and 21 — the landing's final gates (report 46/46, umtool 21/21 after release 22's fix round) Co-Authored-By: Claude Opus 5.5 (1M context)

reports: near the end, the place the URL names is the part being read while it is on screen

2026-10-10T01:15:19Z

commit 194a70a2aeca8d5a3a4b6937c58eabda8a9fddda parent da4a15b51190437f3f97c179e42cc07dc0845cfd Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 21:15:19 -0400 reports: near the end, the place the URL names is the part being read while it is on screen The last places of a report cannot scroll up to the reading line, so a switch after a jump to the last claim opened the slide before it. Both readers now prefer the URL's place when it is on screen below the line. The export changelog's sticky-switch bullet folds into release 22's (both unreleased). Co-Authored-By: Claude Opus 5.5 (1M context)

Merge r19/shapes-fix (release 22's fix round: the sticky view switch, the part being read, umtool's slide URLs, two spec races) into r19/integration

2026-10-10T01:11:35Z

commit da4a15b51190437f3f97c179e42cc07dc0845cfd parent 2c850a8dc7dfd95f8e9ace36b916250c974680c2 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 21:11:35 -0400 Merge r19/shapes-fix (release 22's fix round: the sticky view switch, the part being read, umtool's slide URLs, two spec races) into r19/integration Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 22's fix round — the sticky switch, each first-run failure's cause and fix, final counts

2026-10-10T01:11:08Z

commit 8e78a76aaa2e94c1965f0a972c63a3171a13a7e8 parent af992bdc9c20dd37767e70413a20035d553b405c Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 21:11:08 -0400 plans: release 22's fix round — the sticky switch, each first-run failure's cause and fix, final counts Co-Authored-By: Claude Opus 5.5 (1M context)

reports: the part being read is the last place at the top of the reading area (under the sticky bars), not above the top third

2026-10-10T01:09:01Z

commit af992bdc9c20dd37767e70413a20035d553b405c parent 01a66959299e7308c3fd6bba1e219b7e90e2c245 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 21:09:01 -0400 reports: the part being read is the last place at the top of the reading area (under the sticky bars), not above the top third A jump to a section with a short lead, then a switch, opened its first claim's slide: the claim's top was already above the top-third line. The line is now where a jump lands a place (its scroll-margin-top, under the site's header and the switch) plus 32 px, on the export site and in umtool. overview.spec waits for ArrowRight's focus (moved on the next frame) before Enter, as it does after every other key. Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 19's landing as it went (gates, the two fixes, start-mode e2e 24 min); e2e-speed planned

2026-10-10T01:07:01Z

commit 2c850a8dc7dfd95f8e9ace36b916250c974680c2 parent afc918a01e571703f50a8f12ee3335a9ea39f5d6 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 21:07:01 -0400 plans: release 19's landing as it went (gates, the two fixes, start-mode e2e 24 min); e2e-speed planned Co-Authored-By: Claude Opus 5.5 (1M context)

reports: the view switch is sticky — under the site's header on the export site, atop the reading column in umtool

2026-10-10T00:55:00Z

commit 01a66959299e7308c3fd6bba1e219b7e90e2c245 parent d1dfeed4415f0a5a819faafb91010e2db0ffd191 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 20:55:00 -0400 reports: the view switch is sticky — under the site's header on the export site, atop the reading column in umtool The operator's ruling: switching keeps the place, for a real reader, so Article | Slides | Overview stays in reach while the article scrolls. On the export site it leaves the report's

for a strip right after it, `sticky top-14` within the article's box (the site header is 56 px + its border); in umtool it sticks to the top of the reading column. Nothing scrolls on a click any more, so the part in view is the part read. Every anchor in the article lands below the bars (`[&_[id]]:scroll-mt-32` on the export article, `scroll-mt-16` on umtool's blocks and claims), and anchorInView's line is never above a landed anchor's scroll-margin-top. Specs: the switch is `[data-report-views] [data-view-switch]`; after the jump to a claim the switch is in the viewport and the claim below it. Co-Authored-By: Claude Opus 5.5 (1M context)

umtool e2e: the evidence walk's first j waits for the keys to be live (a press before hydration was lost)

2026-10-10T00:54:53Z

commit d1dfeed4415f0a5a819faafb91010e2db0ffd191 parent 2f4615cc30c07f120b45bb3b068a9aa0b467cc59 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 20:54:53 -0400 umtool e2e: the evidence walk's first j waits for the keys to be live (a press before hydration was lost) The trace shows the press landing ~30 ms before the walk hydrated (its evidence fetch, an effect, starts after it); nothing in the walk changed. The press now repeats until the walk moves — j at the last citation stays there, so a repeat is safe. Co-Authored-By: Claude Opus 5.5 (1M context)

umtool: a slides URL that names a place becomes that place's slide (#s-), as on the export site

2026-10-10T00:54:48Z

commit 2f4615cc30c07f120b45bb3b068a9aa0b467cc59 parent afc918a01e571703f50a8f12ee3335a9ea39f5d6 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 20:54:48 -0400 umtool: a slides URL that names a place becomes that place's slide (#s-), as on the export site ArticleShapes now rewrites `?rv=slides#` (a switch from the text, a link to a section) to the slide's own hash, replacing the entry, as ReportReader does; it used to stay `#report-head` / `#later`. Co-Authored-By: Claude Opus 5.5 (1M context)

export: the phone overview spec reads both boxes in one frame (the focused pair smooth-scrolls between two reads)

2026-10-10T00:46:50Z

commit afc918a01e571703f50a8f12ee3335a9ea39f5d6 parent 59b7d7bcaf091d80aedbbbebc95568a5c4021f70 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 20:46:50 -0400 export: the phone overview spec reads both boxes in one frame (the focused pair smooth-scrolls between two reads) Co-Authored-By: Claude Opus 5.5 (1M context)

r19 gates: the X fakes read --post-range; the slides parity test lives beside Slide.tsx

2026-10-10T00:31:19Z

commit 59b7d7bcaf091d80aedbbbebc95568a5c4021f70 parent ec7018490b752bfb2696e9c93e5dff560cb97644 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 20:31:19 -0400 r19 gates: the X fakes read --post-range; the slides parity test lives beside Slide.tsx fetchOlderPosts' inline fake and the editor e2e fake-gallery-dl still capped on --range after ec701849, so the older-walk limit test wrote 2, not 1. The exportSlides test imported components/ from lib/, a new back-edge the architecture guard refuses; it moves to components/report/. Co-Authored-By: Claude Opus 5.5 (1M context)

social: an X fetch's limit caps posts (--post-range), not files — --range never stopped a metadata-only walk

2026-10-09T23:43:39Z

commit ec7018490b752bfb2696e9c93e5dff560cb97644 parent b56921eda36d3ea632d9c2c446c9677745492d1f Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 19:43:39 -0400 social: an X fetch's limit caps posts (--post-range), not files — --range never stopped a metadata-only walk Co-Authored-By: Claude Opus 5.5 (1M context)

Merge reports/timeline-feeds (a living report's dated entries: the Timeline region, a slide per entry, RSS + JSON feeds) into r19/integration

2026-10-09T22:41:17Z

commit b56921eda36d3ea632d9c2c446c9677745492d1f parent 137a7ae7f8c5b843e3fb0eb601e764d9f3047042 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 18:41:17 -0400 Merge reports/timeline-feeds (a living report's dated entries: the Timeline region, a slide per entry, RSS + JSON feeds) into r19/integration Co-Authored-By: Claude Opus 5.5 (1M context)

export: the report fixture keeps a timeline; timeline.spec; serve-out applies _headers

2026-10-09T22:37:20Z

commit 3249397137d83010f00c82fe2c8b4cf575f1e548 parent 21f0421be4dce83ac18500225306ebe0a161bd18 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 18:37:20 -0400 export: the report fixture keeps a timeline; timeline.spec; serve-out applies _headers The fixture report gains two entries (2026-10-02, updated 2026-10-04, and 2026-10-04), each citing only c01, so no citation number and no header date moves; its view names the feeds on the fixture site's URL and fixtures/report-site/write.ts rewrote page.json and c01's moment. The e2e stage writes feed.xml and feed.json as compose would (contract.ts). scripts/serve-out.mjs applies the bundle's _headers as Pages does (literal, splat and :placeholder paths), so a spec sees the feeds' media types and CORS. e2e-report/timeline.spec.ts: the Timeline's place, order, dates and anchors, Subscribe and the alternates, both feeds served and their items, an entry's slide, a phone. slides.spec and overview.spec renumbered for the two entry slides (11 → 13). Co-Authored-By: Claude Opus 5.5 (1M context)

export: the timeline's page test, and a changelog line

2026-10-09T22:11:23Z

commit 21f0421be4dce83ac18500225306ebe0a161bd18 parent 5036cb4985d34c0381074fb66f000393b7f2d466 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 18:11:23 -0400 export: the timeline's page test, and a changelog line Co-Authored-By: Claude Opus 5.5 (1M context)

mcp: get_report reads a report's timeline, newest first; `section` may name an entry

2026-10-09T21:53:36Z

commit 5036cb4985d34c0381074fb66f000393b7f2d466 parent 192a13e1aff217f861194186ad7f3e3795553bef Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 17:53:36 -0400 mcp: get_report reads a report's timeline, newest first; `section` may name an entry Co-Authored-By: Claude Opus 5.5 (1M context)

export: a report's Timeline, and Subscribe: RSS · JSON when it has feeds

2026-10-09T21:53:36Z

commit 192a13e1aff217f861194186ad7f3e3795553bef parent 509770004976fe4fd124e5a5e35d28ca2cb5ec00 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 17:53:36 -0400 export: a report's Timeline, and Subscribe: RSS · JSON when it has feeds The Timeline sits after In brief, before what the check found: dated blocks newest first on a rail, the date first and in the accent, linking to the entry's own anchor. The header links the feeds and the page's metadata names them (, absolute). A moment cited in an entry says so under "Cited in". Co-Authored-By: Claude Opus 5.5 (1M context)

report: a timeline's feeds — feed.xml (RSS 2.0) and feed.json (JSON Feed 1.1)

2026-10-09T21:53:36Z

commit 509770004976fe4fd124e5a5e35d28ca2cb5ec00 parent 43cb0d43b242b693741e0ec1594e4050064088bf Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 17:53:36 -0400 report: a timeline's feeds — feed.xml (RSS 2.0) and feed.json (JSON Feed 1.1) lib/report/feeds.ts renders both, pure: one item per entry newest first, its permalink the page at its anchor, its body as the HTML export renders it with every link absolute (a citation's marker to its reference on the page). Compose writes them beside the page for a report with entries on a site with a public siteUrl, and names them in the view (absolute URLs); a site with none publishes no feed. _headers serves them as application/rss+xml and application/feed+json, readable cross-origin; the bundle audit knows the files as report data. Co-Authored-By: Claude Opus 5.5 (1M context)

report: a timeline — dated entries, newest first, in the view, the slides and the exports

2026-10-09T21:53:35Z

commit 43cb0d43b242b693741e0ec1594e4050064088bf parent a8a5806f075b23620d1d8568390b120f47f4fb4e Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 17:53:35 -0400 report: a timeline — dated entries, newest first, in the view, the slides and the exports report.json gains `entries` ({id, date, title, body, updated?}): ids share the section/claim anchor namespace and refuse the page's own anchors; dates are report dates, an entry's `updated` not before its `date`; every cite: link in a body resolves. lib/report/entries.ts is the one order (newest first, the same instant in file order) the numbering, the view, the slides and the exports read. The view's `updated` counts the newest entry (the index card with it). One slide per entry after In brief. report.html / report.md carry a Timeline before the sections. The change summary names entries added, removed and edited. REPORT.md regenerated. Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 18 — the rollout as it went

2026-10-09T21:49:47Z

commit 137a7ae7f8c5b843e3fb0eb601e764d9f3047042 parent a8a5806f075b23620d1d8568390b120f47f4fb4e Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 17:49:47 -0400 plans: release 18 — the rollout as it went Co-Authored-By: Claude Opus 5.5 (1M context)

Merge hotfix/no-session-ids (release 15's rollout line trimmed)

2026-10-09T21:13:53Z

commit a8a5806f075b23620d1d8568390b120f47f4fb4e parent 05148db6b454255b77d31e8116a804106f6b35da Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 17:13:53 -0400 Merge hotfix/no-session-ids (release 15's rollout line trimmed) Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 15's rollout line trimmed

2026-10-09T21:13:46Z

commit ec0903b37a17d1b8647672290cd7cb5d062cab35 parent 4134b9740227f4e96a258390deb9ac2376605dea Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 17:13:46 -0400 plans: release 15's rollout line trimmed Co-Authored-By: Claude Opus 5.5 (1M context)

docs: COMMANDS.md regenerated for tracks A, D and E

2026-10-09T18:53:05Z

commit 05148db6b454255b77d31e8116a804106f6b35da parent 320b844af53698185305ad5a1218670672cc499f Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:53:05 -0400 docs: COMMANDS.md regenerated for tracks A, D and E Co-Authored-By: Claude Opus 5.5 (1M context)

Merge track E (release 22 E1–E6: slide fields, the slides view and isometric overview, umtool's switch, slides exports) into r19/integration

2026-10-09T18:52:53Z

commit 320b844af53698185305ad5a1218670672cc499f parent 657bf290795b2eece02e63b0e876da3c64c8c5d9 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:52:53 -0400 Merge track E (release 22 E1–E6: slide fields, the slides view and isometric overview, umtool's switch, slides exports) into r19/integration Co-Authored-By: Claude Opus 5.5 (1M context)

Merge track D (release 21 D1 attach-media, D3 playable copies + torrents, D4b the last-resort seeder) into r19/integration

2026-10-09T18:52:47Z

commit 657bf290795b2eece02e63b0e876da3c64c8c5d9 parent f5d2a6e654fa9c7a569642f0c610d52533dc7cdd Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:52:47 -0400 Merge track D (release 21 D1 attach-media, D3 playable copies + torrents, D4b the last-resort seeder) into r19/integration Co-Authored-By: Claude Opus 5.5 (1M context)

Merge track A (release 19 A1–A5: ops finds its token, jobs over ops, read side + gated auto-queue control, archival writes, fetch-window queue + dedupe) into r19/integration

2026-10-09T18:52:44Z

commit f5d2a6e654fa9c7a569642f0c610d52533dc7cdd parent 1bc6e241fde7b4726d8de5e136b1c4924e110e29 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:52:44 -0400 Merge track A (release 19 A1–A5: ops finds its token, jobs over ops, read side + gated auto-queue control, archival writes, fetch-window queue + dedupe) into r19/integration Co-Authored-By: Claude Opus 5.5 (1M context)

Merge hotfix/no-session-ids (the rule above the generated block)

2026-10-09T18:52:32Z

commit 1bc6e241fde7b4726d8de5e136b1c4924e110e29 parent d6fc5bf74844fd476c9ee03cee313309db380dfe Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:52:32 -0400 Merge hotfix/no-session-ids (the rule above the generated block) Co-Authored-By: Claude Opus 5.5 (1M context)

agents: the session-link rule sits above the generated Next.js block

2026-10-09T18:52:28Z

commit 4134b9740227f4e96a258390deb9ac2376605dea parent 5c3045c088a11cac5258121d924eeb90bf91c69d Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:52:28 -0400 agents: the session-link rule sits above the generated Next.js block Co-Authored-By: Claude Opus 5.5 (1M context)

Merge hotfix/no-session-ids (session trailers and links never ship) into r19/integration

2026-10-09T18:52:15Z

commit d6fc5bf74844fd476c9ee03cee313309db380dfe parent 016638429fcc7f477f00d94a990bafd5189492f9 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:52:15 -0400 Merge hotfix/no-session-ids (session trailers and links never ship) into r19/integration Co-Authored-By: Claude Opus 5.5 (1M context)

Merge r19/integration (the release 22 overview ruling) into Track E

2026-10-09T18:49:50Z

commit f5da2b230effaf8d4a6151a97c80be22d30467be parent 31c4cfd2d150e2f89d7d3c6ef1d5289c0ac79f40 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:49:50 -0400 Merge r19/integration (the release 22 overview ruling) into Track E Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 22 — Track E as shipped (E1–E5), the overview ruling confirmed; export changelog; PUBLISH.md's exports name the slides

2026-10-09T18:49:26Z

commit 31c4cfd2d150e2f89d7d3c6ef1d5289c0ac79f40 parent ce0dc5e1989551fb882224bf931f6283c520bb65 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:49:26 -0400 plans: release 22 — Track E as shipped (E1–E5), the overview ruling confirmed; export changelog; PUBLISH.md's exports name the slides Co-Authored-By: Claude Opus 5.5 (1M context)

common, export: slides.html and slides.pdf join a report's exports (release 22 E5)

2026-10-09T18:47:37Z

commit ce0dc5e1989551fb882224bf931f6283c520bb65 parent 330ffef1294d490daa25ecc4d9a02b6550f6c2de Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:47:37 -0400 common, export: slides.html and slides.pdf join a report's exports (release 22 E5) `slides` (slides.html) and `slides-pdf` (slides.pdf) join REPORT_EXPORT_FORMATS, so the downloads line, compose's publishing, the cited audit's allowlist, the revision commit's exports.json and the CLI's and the ops job's `formats` take them with no change of their own. slides.html is one self-contained file like report.html (lib/report/exportSlidesHtml.ts): the site's slide stylesheet inline, the light base in the site's accent, every picture a data: URI, a few lines of script to fit and step, citations numbered to a reference list under the deck, each slide's way back to its place on the site, and the sources slide naming which document this is. Its markup mirrors Slide.tsx class for class (a test holds them equal): the export is built in the editor's server bundle, where React's server renderer is not imported. slides.pdf prints it one 1280 × 720 page per slide (the printer takes a page size), skipped with a note where there is no browser, as report.pdf is. The evidence pack carries slides.html with its media; a clip's poster is inlined or packed when this host has it, never a problem when it has not. A report whose `slides.hide` says so exports neither. Co-Authored-By: Claude Opus 5.5 (1M context)

umtool: an article read as slides and in the overview, beside its notes; the source tab lints its slides (release 22 E4)

2026-10-09T18:43:03Z

commit 330ffef1294d490daa25ecc4d9a02b6550f6c2de parent 8932ad0dd46a349cdb457f9abffa21202105b2e3 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:43:03 -0400 umtool: an article read as slides and in the overview, beside its notes; the source tab lints its slides (release 22 E4) The article page's reader gets the published page's switch (Article | Slides | Overview, `?rv=` and the hash, the shared components over the lenient articleView), so a draft is read in every shape before it ships. The text stays mounted (hidden) under the other views, its note marks with it; the notes rail stays beside them. A citation on a slide opens its evidence in the rail; each slide counts the open notes on its part of the article (a section's on its slide, the title's on the title slide, a citation's on each slide showing it); Esc or "Read this in the article" goes back to the text at the slide's place. The source tab lists what would make the slides read badly (slideLint, and the validator's slide problems), each linked to its slide. The lenient view now carries the slide fields and a claim's sentence as its id. Co-Authored-By: Claude Opus 5.5 (1M context)

common, export: the isometric overview — each part of the article beside its slide (release 22 E3)

2026-10-09T18:42:52Z

commit 8932ad0dd46a349cdb457f9abffa21202105b2e3 parent bd910cbe90fb2d025a84edde75abc33b87bdd106 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:42:52 -0400 common, export: the isometric overview — each part of the article beside its slide (release 22 E3) ReportOverview lays the article and its slides side by side on one plane, tilted isometrically (rotateX(55deg) rotateZ(-45deg), the plane's box sized to the tilt so it scrolls whole): the article's parts as a column of page blocks, each slide beside its part as a 16:9 card, a line between them. The pair in force is lit and lifted; pointing at a pair lights both; a block opens the article there, a card the slides. ↑ ↓ (Home, End) walk the pairs, ← → the block and the card, Enter opens, Esc goes back to the article at the pair; only the pair in force is in the tab order. The tilt is never the only way to read it: the same ordered list lies flat (two columns; stacked in a narrow box) for reduced motion, a narrow box, a keyboard arrival, or the "Flat" toggle (remembered in this browser). The switch gains Overview (`?rv=iso`, the pair's anchor in the hash); the scroller alone moves to the pair, never the article behind. Co-Authored-By: Claude Opus 5.5 (1M context)

common, export: the article as slides — the shared slides, the view switch, the reader (release 22 E2)

2026-10-09T18:42:00Z

commit bd910cbe90fb2d025a84edde75abc33b87bdd106 parent 711ae6f0a3861df3093e236940d2b8bc062345fe Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:42:00 -0400 common, export: the article as slides — the shared slides, the view switch, the reader (release 22 E2) common/components/report/slides/: Slide (one slide per kind and layout, its left edge the colour of what it says — a claim's verdict, else the accent — and "Read this in the article" to its anchor), ReportSlides (one 16:9 stage letterboxed in its box, the canvas scaled as one piece, reflowed on a narrow portrait screen; ← → PgUp PgDn Home End, Esc, swipe; a live "Slide n of N" label; a rail), ViewSwitch (Article | Slides, a group of aria-pressed toggles), SlidesHost (what differs by host: markdown, a citation's marker, the way back), useReaderPlace (the `rv` param and the hash, pushed on a switch, replaced on a step). common/styles/slides.css draws them from the site's tokens (accent, light and dark). The export site's report header carries the switch (ReportReader): Slides lays the slides over the page, built in the browser from page.json; the article stays the static page. The place is kept both ways — the part of the article in view opens its slide; Esc or "Read this in the article" lands on the slide's anchor, scrolled to and focused. The page gives every place a slide stands for an id (#report-head, #in-brief, #report-end). Co-Authored-By: Claude Opus 5.5 (1M context)

source: Claude session trailers and links never ship — built-in rules strip them from the mirror's whole history, the audit refuses any left; AGENTS.md says never add them

2026-10-09T18:16:53Z

commit 5c3045c088a11cac5258121d924eeb90bf91c69d parent f625ca7d76a5f2ada902c6fe432eff394a7df21f Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:16:53 -0400 source: Claude session trailers and links never ship — built-in rules strip them from the mirror's whole history, the audit refuses any left; AGENTS.md says never add them Operator, 2026-10-09. Two built-in scrub rules run after the operator's: the session trailer line goes from every published commit message and file, and a bare session link becomes "[session link removed]"; the gate denies the link in any case. The strings are assembled from parts, so the published source holds neither. AGENTS.md opens with the rule (it overrides any reminder asking for the trailer); the implementer rules drop the trailer; the mirror decisions record the change. The release-18 rules line no longer spells the denylist's guarded suffix. The private repository's history is not rewritten (standing decision). Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 21 Track D record — D1 attach-media, D3 prepare-playable, D4b the seeder of last resort, as shipped, with the gates and what is owed to the rollout

2026-10-09T18:07:07Z

commit daadb0befa068115ad0682881f926b8cb03c6f5a parent 2ba8a09e4581e93d26eee593c8faa22d9e936905 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:07:07 -0400 plans: release 21 Track D record — D1 attach-media, D3 prepare-playable, D4b the seeder of last resort, as shipped, with the gates and what is owed to the rollout Co-Authored-By: Claude Opus 5.5 (1M context)

common: attachMediaPlan test titles and ids are synthetic throughout

2026-10-09T18:06:57Z

commit 2ba8a09e4581e93d26eee593c8faa22d9e936905 parent d9ce2e6c664ef703fa554d00930eff1a25d773c4 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:06:57 -0400 common: attachMediaPlan test titles and ids are synthetic throughout Co-Authored-By: Claude Opus 5.5 (1M context)

common: each prepare-playable remux takes the heavy slot (queue-lock --heavy, per video); COMMANDS.md regenerated with the media attach, media playable, seed and tracker rows and the two ops actions

2026-10-09T18:06:19Z

commit d9ce2e6c664ef703fa554d00930eff1a25d773c4 parent a835e96885ddbf6a9817328d914c056f0c5641bd Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:06:19 -0400 common: each prepare-playable remux takes the heavy slot (queue-lock --heavy, per video); COMMANDS.md regenerated with the media attach, media playable, seed and tracker rows and the two ops actions Co-Authored-By: Claude Opus 5.5 (1M context)

common: report slides — the slide fields, their rules, and buildReportSlides (release 22 E1)

2026-10-09T18:06:17Z

commit 711ae6f0a3861df3093e236940d2b8bc062345fe parent e504009d8bc78f96f9a392095822c9845f055a8f Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:06:17 -0400 common: report slides — the slide fields, their rules, and buildReportSlides (release 22 E1) report.json gains `slides` ({hide, title, points, closing}) and a section's or claim's `slide` ({title, points, cite, layout, hide}); version stays 1. The validator holds them to slideRules.ts (lines one line, at most 5 points of at most 140 visible characters, a `cite` of the slide's own section or claim, points citing only what the article cites, an evidence layout with something to show) and refuses a section or claim id that is one of the page's own anchors. lib/report/slides.ts builds the deck from the page view — title, In brief, What the check found, each section and claim, sources — each slide anchored to an id the article page carries; with the reader's place in the URL (`rv`, `#s-`) and a lint for umtool. The view carries the fields through (compose and umtool's lenient view alike); REPORT.md regenerated; the demo fixture gains them. Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 19 Track A records — A1 to A5 as shipped, with the gates

2026-10-09T18:04:47Z

commit bf0355624493ca065fa5a483dd4ff2b71f7f2921 parent 47dc212014f5ed779c09adeeb7e62001135a3e15 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:04:47 -0400 plans: release 19 Track A records — A1 to A5 as shipped, with the gates Co-Authored-By: Claude Opus 5.5 (1M context)

Merge worktree-mk-game-hooks (MK um-song hooks from the original arcade samples; song data under ~/reports; sources linked from the saved-video store)

2026-10-09T18:04:24Z

commit f625ca7d76a5f2ada902c6fe432eff394a7df21f parent a837bf8560a1050e680c9e5930b325d309f735cf Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:04:24 -0400 Merge worktree-mk-game-hooks (MK um-song hooks from the original arcade samples; song data under ~/reports; sources linked from the saved-video store) Co-Authored-By: Claude Opus 5.5 (1M context)

Merge r19/integration into Track D (tracks B and C: the docs cli and COMMANDS.md, the heavy slot, reports check/verify-quotes/attach-video, release 22's plan) before the Track D records; the changelog keeps both sides, Track D first

2026-10-09T18:02:16Z

commit a835e96885ddbf6a9817328d914c056f0c5641bd parent 2484c402c66b17b2bae34b5ea57d05af8b6f885e Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:02:16 -0400 Merge r19/integration into Track D (tracks B and C: the docs cli and COMMANDS.md, the heavy slot, reports check/verify-quotes/attach-video, release 22's plan) before the Track D records; the changelog keeps both sides, Track D first Co-Authored-By: Claude Opus 5.5 (1M context)

common: the home seeder of last resort — archilyzer seed (WebTorrent 3, TCP + WebRTC through node-datachannel) seeds a torrent only while no other seeder has it; archilyzer tracker; settings.seeder; the docker profile seeder behind a WireGuard namespace; doctor compares egress

2026-10-09T18:01:50Z

commit 2484c402c66b17b2bae34b5ea57d05af8b6f885e parent 57ae8c2a8abe38f01747471d9ff8c5892a526f5b Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 14:01:50 -0400 common: the home seeder of last resort — archilyzer seed (WebTorrent 3, TCP + WebRTC through node-datachannel) seeds a torrent only while no other seeder has it; archilyzer tracker; settings.seeder; the docker profile seeder behind a WireGuard namespace; doctor compares egress D4b of release 21. lib/lastResortSeeder.ts is the pure state machine (standby after other seeders on every poll for standbyAfterSeconds; back at once for a waiting leecher with no other source, or after the same window alone; silent trackers in standby resume; a transition restarts both clocks), 13 unit tests against a scripted scrape. controller/seeder.ts observes (each tracker's scrape, merged by max; its wires counted by peer) and applies (standby = removed from the client, data kept; seeding = added back with skipVerify after a size check); no DHT, LSD, UPnP or web seeds; bindInterface held. The tracker is bittorrent-tracker HTTP + WS, no UDP, tracking only the manifests' infohashes. Localhost integration tests: TCP seed → standby → resume, and a WebRTC leecher completing through the WS tracker. docker-compose.seeder.yml: gluetun v3.41.3 (firewall always on), the seeder and tracker in its namespace, SEEDER_WG_CONF required; a config test pins the construction. pnpm-workspace allows node-datachannel's prebuilt binary and nothing else new. Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 22 — the isometric overview is the operator's meaning (confirmed)

2026-10-09T17:53:09Z

commit 016638429fcc7f477f00d94a990bafd5189492f9 parent e504009d8bc78f96f9a392095822c9845f055a8f Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 13:53:09 -0400 plans: release 22 — the isometric overview is the operator's meaning (confirmed) Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 22 planned — one article, two shapes: slides from the same report.json, a reader switch, the isometric overview, slides exports

2026-10-09T17:50:32Z

commit e504009d8bc78f96f9a392095822c9845f055a8f parent 1d2c642eb95328a122175c94e02b8ad3d8e6af7d Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 13:50:32 -0400 plans: release 22 planned — one article, two shapes: slides from the same report.json, a reader switch, the isometric overview, slides exports Ruling 2026-10-09: articles may be built a bit differently (report.json gains slide fields); the isometric overview pairs each section with its slide. Co-Authored-By: Claude Opus 5.5 (1M context)

docs: COMMANDS.md regenerated for track B's reports and heavy rows

2026-10-09T17:46:01Z

commit 1d2c642eb95328a122175c94e02b8ad3d8e6af7d parent 927e559f62e24b56736990616b6232245dc94cae Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 13:46:01 -0400 docs: COMMANDS.md regenerated for track B's reports and heavy rows Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 18 — As it went: Wave 0, main merged in again, the gates and the full suite on a837bf85

2026-10-09T17:31:00Z

commit 927e559f62e24b56736990616b6232245dc94cae parent db37837b04ce66384f3d55b3ffdd47cbc2384b53 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 13:31:00 -0400 plans: release 18 — As it went: Wave 0, main merged in again, the gates and the full suite on a837bf85 Co-Authored-By: Claude Opus 5.5 (1M context)

common: stageRun test cancels once the lock wait is announced, not after a fixed 50 ms (raced the announcement under load)

2026-10-09T17:30:37Z

commit db37837b04ce66384f3d55b3ffdd47cbc2384b53 parent e6d2da4cd7b8aea1ab6df372c248f6b33bf2cebd Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 13:30:37 -0400 common: stageRun test cancels once the lock wait is announced, not after a fixed 50 ms (raced the announcement under load) Co-Authored-By: Claude Opus 5.5 (1M context)

Merge track B (B6 test economy, B1 the heavy slot, B4 ops transcribe timing + priority, B3 reports check/verify-quotes/attach-video, B5 umtool debts) into r19/integration

2026-10-09T17:30:37Z

commit e6d2da4cd7b8aea1ab6df372c248f6b33bf2cebd parent 166a1ac691b412199b8b4c387f0ef1b7a56293ac Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 13:30:37 -0400 Merge track B (B6 test economy, B1 the heavy slot, B4 ops transcribe timing + priority, B3 reports check/verify-quotes/attach-video, B5 umtool debts) into r19/integration B2 (report-to-video robustness) did not ship: writes to umtool/report-to-video/ were refused by the permission system; the partial work is outside the tree. Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 19 Track B records — B6, B1, B4, B3, B5 as shipped, the gates, and B2 not shipped (writes to report-to-video refused)

2026-10-09T17:26:12Z

commit cf5e57ac081ceaa0ae8cff3f58c81827773e3061 parent 70a0123f1e3c66f9d924e3f065dc6c1c6dc7b433 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 13:26:12 -0400 plans: release 19 Track B records — B6, B1, B4, B3, B5 as shipped, the gates, and B2 not shipped (writes to report-to-video refused) Co-Authored-By: Claude Opus 5.5 (1M context)

common: prepare-playable — each saved container remuxed losslessly into a browser-playable mp4 (+faststart) or webm, one single-file torrent per copy, sha-keyed so a re-run is a no-op

2026-10-09T17:24:58Z

commit 57ae8c2a8abe38f01747471d9ff8c5892a526f5b parent 6d9cb23d1cee35c03e999dc33224207f4570ae86 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 13:24:58 -0400 common: prepare-playable — each saved container remuxed losslessly into a browser-playable mp4 (+faststart) or webm, one single-file torrent per copy, sha-keyed so a re-run is a no-op D3 of release 21. lib/playableMedia.ts is the decision table (h264/vp9/av1 + aac/opus/mp3 → mp4; vp9/vp8/av1 + opus/vorbis in webm → webm; anything else listed, never re-encoded), first video + first audio only, metadata and chapters dropped, +bitexact. create-torrent 6.1.3 + parse-torrent 11.0.24: name ., 256 KiB–1 MiB pieces, no urlList, no comment, no created by, no private flag, fixed creation date, announce from a parameter. Copies, torrents and playable.json under playable// beside the saved-video store. Job kind prepare-playable (drainable, needsMedia, ingest, one queue machine-wide), ops route + action, pnpm ops row, archilyzer media playable. Co-Authored-By: Claude Opus 5.5 (1M context)

Merge track C (C2 OPERATING.md + docs cli, C4 homepage docs, C5 FACTS, the track's record) into r19/integration

2026-10-09T17:13:38Z

commit 166a1ac691b412199b8b4c387f0ef1b7a56293ac parent 93e59abdd569b113a1728de0c31b50f4681da298 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 13:13:38 -0400 Merge track C (C2 OPERATING.md + docs cli, C4 homepage docs, C5 FACTS, the track's record) into r19/integration Co-Authored-By: Claude Opus 5.5 (1M context)

Merge r19/integration into Track B (the release 19-21 plans, track C's doc fixes, the jobKinds test fix) before the Track B records

2026-10-09T17:08:56Z

commit 70a0123f1e3c66f9d924e3f065dc6c1c6dc7b433 parent ef01fcd6406aa333ce294f1cfe5c236547664eb7 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 13:08:56 -0400 Merge r19/integration into Track B (the release 19-21 plans, track C's doc fixes, the jobKinds test fix) before the Track B records Co-Authored-By: Claude Opus 5.5 (1M context)

umtool: the two-section Note spec waits for its mark before reading notes.json; the edit-guard test takes its kind from the registry

2026-10-09T17:08:39Z

commit ef01fcd6406aa333ce294f1cfe5c236547664eb7 parent af054e224d5cf9021fe1696411e7c60fd0e6bf55 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 13:08:39 -0400 umtool: the two-section Note spec waits for its mark before reading notes.json; the edit-guard test takes its kind from the registry From the B5 umtool e2e run: the new article-notes test read notes.json before the save had landed (the existing tests wait for the mark first, and now this one does), and projects.spec's registry guard caught a kind id spelled out in lib/report/edit-guard.test.mjs — it now asks the registry for a kind that takes notes. Two comments name REPORTS_DIR instead of a home path. Co-Authored-By: Claude Opus 5.5 (1M context)

common: attach-media — a channel's held videos get their media from a local archive (directory, zip read in place, 7z) into the saved-video store, with local-archive provenance

2026-10-09T17:08:08Z

commit 6d9cb23d1cee35c03e999dc33224207f4570ae86 parent 93e59abdd569b113a1728de0c31b50f4681da298 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 13:08:08 -0400 common: attach-media — a channel's held videos get their media from a local archive (directory, zip read in place, 7z) into the saved-video store, with local-archive provenance D1 of release 21. The id from the folder's "()" or the file's yt-dlp suffix, or explicit items; written through persistSourceVideo (a staging file on the store's disk, then a rename), keepReason pin, origin {kind: "local-archive", archive, entry, sha256, attachedAt}; refuses over a pointer unless replace; createRecords writes metadata.info.json through the metadata history (writer "local-archive"); [LOST] folders listed, never attached. Job kind attach-media (drainable, needsMedia, ingest), ops route + action, pnpm ops row, archilyzer media attach. Co-Authored-By: Claude Opus 5.5 (1M context)

editor: the ops route tests' corpus helper opts its fixture paths out of tracing

2026-10-09T16:46:32Z

commit 47dc212014f5ed779c09adeeb7e62001135a3e15 parent cb56ff9d4982a9579592f6448592819c76ddc2a9 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 12:46:32 -0400 editor: the ops route tests' corpus helper opts its fixture paths out of tracing next-build-trace.test.mjs scans every module under editor/app that is not a test file, and _testCorpus.ts (A2b) joins paths off import.meta.dirname; its calls carry /* turbopackIgnore: true */ as the check asks. Test-only: no route imports it. Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 19's Track C record (C1-C5, as shipped)

2026-10-09T16:45:13Z

commit 8614ab772b4a76f56f47e8c206d5bb4fef13dbb1 parent 6b98abb7391aa04ff7d06032addc5fa1a7abcbd3 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 12:45:13 -0400 plans: release 19's Track C record (C1-C5, as shipped) Co-Authored-By: Claude Opus 5.5 (1M context)

Merge r19/integration into track C (release 21's plan, the jobKinds test fix) before the record commit

2026-10-09T16:40:18Z

commit 6b98abb7391aa04ff7d06032addc5fa1a7abcbd3 parent b7b88a56bb42860a3ab031f40dbbd63f9f109086 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 12:40:18 -0400 Merge r19/integration into track C (release 21's plan, the jobKinds test fix) before the record commit Co-Authored-By: Claude Opus 5.5 (1M context)

homepage: the Running an archive doc says how to run one from a shell or an agent

2026-10-09T16:40:05Z

commit b7b88a56bb42860a3ab031f40dbbd63f9f109086 parent a21a7592179d5b31871e52b16fbf27c5e41aaa71 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 12:40:05 -0400 homepage: the Running an archive doc says how to run one from a shell or an agent A section, "From a shell or an agent": `pnpm ops` (the running editor's actions over HTTP), `pnpm archilyzer` (the core's command line) and the MCP server, three example commands, and links to the source's OPERATING.md and COMMANDS.md. Gates: `pnpm --filter homepage test` 23/23; `build:nodata` ok (45 s, 5 GB scope). Co-Authored-By: Claude Opus 5.5 (1M context)

plans: release 21 planned — playable archives (local media for held videos, clip windows cut locally, per-video torrents, a home seeder of last resort behind a VPN; pilot TISM on jeralyzer-private)

2026-10-09T16:39:24Z

commit 93e59abdd569b113a1728de0c31b50f4681da298 parent 1a9e2aa1e99a58fc6d5f7ba2993587120231d53f Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 12:39:24 -0400 plans: release 21 planned — playable archives (local media for held videos, clip windows cut locally, per-video torrents, a home seeder of last resort behind a VPN; pilot TISM on jeralyzer-private) Rulings 2026-10-09: pilot on jeralyzer-private first; no web seed and no R2 — a home seeder only, behind a VPN, seeding only when no other seeder exists; ASR for the three new videos only. PLAN.md's release index names it. Co-Authored-By: Claude Opus 5.5 (1M context)

umtool: its own e2e ports per worktree, a hermetic mix.spec, `umtool window` notes its edits, two-block Note, SITES_DIR from the CLI's checkout, /sites is "articles"

2026-10-09T16:12:27Z

commit af054e224d5cf9021fe1696411e7c60fd0e6bf55 parent 19e0f6b76ec44d4e137752be52f3c31a9f61f0bb Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 12:12:27 -0400 umtool: its own e2e ports per worktree, a hermetic mix.spec, `umtool window` notes its edits, two-block Note, SITES_DIR from the CLI's checkout, /sites is "articles" Six debts from the articles + notes merge: - Per-worktree e2e ports. umtool's `e2e` script now runs through the worktree injector (`worktree.mjs run`), so the documented `pnpm --filter umtool run e2e` gets the worktree's own ports. And the injector's index was the FIRST worktree root containing the cwd: every `.claude/worktrees/` is nested in the main checkout, so all of them got offset 0 — the main checkout's ports — for every suite. It is the most specific root now (indexForPath, tested). - mix.spec's order dependence (FACTS: `:166`/`:201` red in every full run, green alone). Specs before it fetch windows through the editor stub into the fixture corpus's clips/, which the bench folds in with a project's clips-raw. The file now moves the corpus's clip windows aside for its tests and puts them back after; and the render-scratch test matches the option label exactly instead of as a substring. - `umtool window` goes through the routes' edit guard: on a generated manifest each change is an `edit` note for the generator. The guard is lib/report/edit-guard.mjs now (plain JS, so the CLI runs the same one); guard.ts only types it. - A selection that runs across two sections gets its Note button: it is anchored in the section it starts in, to that section's end (else the one it ends in, from its start). A spec covers it. - The CLI finds SITES_DIR (and CHANNELS_DIR) from its OWN checkout: REPO_ROOT is the repo of `/umtool/bin/.mjs` when that is the entry script, so `umtool notes --all` run from a ~/reports workspace reads the corpus's sites. The app keeps the cwd walk. umtool/lib/*.test.mjs joins test:scripts. - The nav entry and crumbs for /sites read "articles" (lowercase, like every umtool nav entry): the editor's /sites is the sites themselves. The URL is unchanged; sites.spec asserts the new name and that no "sites" link is left. Co-Authored-By: Claude Opus 5.5 (1M context)

Merge branch 'r19/integration' into worktree-agent-a103bef2ee4055362

2026-10-09T16:09:47Z

commit cb56ff9d4982a9579592f6448592819c76ddc2a9 parent c968ddf6c36ff18bdc434d87e8bb69487a74d648 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 12:09:47 -0400 Merge branch 'r19/integration' into worktree-agent-a103bef2ee4055362

editor: clip windows get their own queue per platform and are never queued twice

2026-10-09T16:09:23Z

commit c968ddf6c36ff18bdc434d87e8bb69487a74d648 parent d6afa774ba81b366cc9f5cdb3e2b27f68789a121 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 12:09:23 -0400 editor: clip windows get their own queue per platform and are never queued twice Release 19 slice A5. A fetch-window job ran on the platform's download queue, and a queue runs one job at a time — a tier orders only the waiting jobs — so a few seconds of video waited behind a multi-hour persist-videos or sync. clipWindowQueueKey (lib/queueKeys.ts) puts the single fetch (fetchWindowAction) and the batches (fetchWindowsAction) on `clips:`: one window at a time per platform, the batches' clip-window gap kept, the platform's hold, cooldown and backoff still shared. A repeated ask is answered with the job already fetching it: jobs/windowJobs.ts' windowInFlight finds a queued or running fetch-window whose window contains the asked one, or a fetch-windows batch with such an item, same channel, video and height cap, off the job's replay spec. The single route answers that jobId with `existing: true`; the batch lists such windows in `inFlight`, starts no job for them, and the ops route adds their jobs to `jobIds` so --wait follows the windows asked for. Retry is not deduplicated. umtool/report-to-video/fetch-via-editor.mjs (its own wait and 10-minute poll timeout) is not touched: that directory belongs to another session. Co-Authored-By: Claude Opus 5.5 (1M context)

reports: `reports check`, `reports verify-quotes` and `reports attach-video` on the CLI

2026-10-09T16:03:42Z

commit 19e0f6b76ec44d4e137752be52f3c31a9f61f0bb parent 762642b24ae0f1efcbd96b6b4824412745bd062a Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 12:03:42 -0400 reports: `reports check`, `reports verify-quotes` and `reports attach-video` on the CLI Three report jobs that meant a build, or a by-hand script, before: - `archilyzer reports check [--reports a,b] [--allow-missing-media]` is the reports stage of compose with nothing written — resolveSiteReports: validation, every cited quote against its record, prepared media present and current, the report video under the publish limit — and exits 1 with compose's own problem lines. --reports checks drafts not yet in site.json. - `archilyzer reports verify-quotes [--json]` runs compose's quote check over every video, audio and post quote of one report: the best score and track, and the `en-orig` track's score where the record has one. A quote that matches a served `en` rewrite but not en-orig (the words as spoken) is `en-orig-drift`, though compose, which takes the best track, would pass it. The span check is now ONE function, checkSpanQuote, which compose calls too (it was inline in resolveSiteReports); its drift sentence likewise (quoteDriftMessage). - `archilyzer reports attach-video [--poster] [--caption]` (publish/reportVideo.ts): an H.264 mp4 under 24 MiB is remuxed (+faststart); anything else is encoded (H.264/AAC, bitrate from the duration inside 96% of the limit, narrower as it falls) and encoded again smaller if a pass overshoots; too long to be watchable is refused with its length. video.mp4 and a poster (given, kept, or a frame) land beside report.json, then `video` is set — only if the video's own fields validate. Tests: bin/reports-check.test.ts, 10 (a temp corpus with a rewritten `en` track; two of them encode real 3 s clips with ffmpeg, skipped without it). composeReports' three test files unchanged and green (19). Co-Authored-By: Claude Opus 5.5 (1M context)

ops: archival writes — settings, publish lane, platform holds, workers, one video, cleanup, relocate dryRun; archilyzer storage report

2026-10-09T16:01:32Z

commit d6afa774ba81b366cc9f5cdb3e2b27f68789a121 parent 5aaeec9382a2492196141dbdbe2a4948a212f663 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 12:01:32 -0400 ops: archival writes — settings, publish lane, platform holds, workers, one video, cleanup, relocate dryRun; archilyzer storage report Release 19 slice A4. Each route is the existing action: - POST /api/ops/settings {patch}: saveSettings, refusing an unknown key, a key another writer owns (channelPriority, autoQueue, workers, storage) and a value the schema would coerce (settings/patch.ts compares every leaf sent with the schema's reading of the merged settings — the schema clamps rather than throwing, so { ok: true } would otherwise answer an unsaved value). - lane: the publish lane (savePublishSettingsAction with the stored fields, the publish hold, start/stop/drainPublishLaneAction); the four ingest lanes' drain now goes through drainAutoQueueAction. - clear-platform-hold {platform}: clearPlatformHoldAction. - POST /api/ops/workers {op, ids}: enable/disable/drainWorkerAction per id. - transcribe-one {slug, id, file?}: transcribeOneAction / whisperVideoAction. - delete-file {slug, id, file}: deleteVideoFileAction (removeMediaFile). - do-not-clean {slug, id, keep?}: toggleDoNotCleanAction, set not toggled. - POST /api/ops/cleanup {slug, sweep}: the channel page's four sweeps. - relocate {dryRun}: previewRelocationAction per slug (a missing channel is the bulk path's "channel not found"). `archilyzer storage report [...] [--json]` (common/bin/storage-report.ts): each channel's tierable media, text and clip bytes off its report and where its media is (corpus disk, a location, a custom path, legacy), with totals per place; unknown is never 0; no drive is touched. Co-Authored-By: Claude Opus 5.5 (1M context)

transcribe-file: durationMs is the engine's time (waitedMs the queue); a short file waits in the urgent tier

2026-10-09T15:56:03Z

commit 762642b24ae0f1efcbd96b6b4824412745bd062a parent 2c13c9da898c003820e64b604594a206fd4ad286 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 11:56:03 -0400 transcribe-file: durationMs is the engine's time (waitedMs the queue); a short file waits in the urgent tier `pnpm ops transcribe`'s durationMs ran from the job's start, so it counted the ffmpeg cut and the wait for a free worker — on a box whose GPU worker was busy with the lane, minutes that read as a slow engine. It now runs from the moment a worker takes the job (onWorker; the last attempt's, after a transport retry) to the transcript, and the new `waitedMs` is the queue. The job's log says both. The pool already parks the lane's units (background) behind a manual acquire, so a file transcription was ahead of the lane — but FIFO behind a manual batch's next video. A cut of up to 15 minutes of audio (URGENT_MAX_AUDIO_SEC, measured from the cut WAV's size before the acquire) now asks the pool's existing "urgent" tier, ahead of every parked waiter; a longer file keeps "foreground". The tier orders waiters only: nothing running is interrupted. transcribeWithWorker takes a `tier` and passes it to pool.acquire; no other caller changes. Tests: transcribeFile 21 → 24 (the tier from the cut's length; waitedMs vs durationMs with the GPU slot held; a file job parked LAST takes the freed slot before a parked manual and a parked lane acquire — and fails with the tier forced to "foreground"). Co-Authored-By: Claude Opus 5.5 (1M context)

ops: the read side — settings, storage, sites, workers, lanes, scheduler, cleanup; the lane control route needs the token

2026-10-09T15:52:43Z

commit 5aaeec9382a2492196141dbdbe2a4948a212f663 parent a35e901abbf219ff0ce44600f0d8e9464c530ec4 Author: I Mean I'm Just Saying Date: Fri, 9 Oct 2026 11:52:43 -0400 ops: the read side — settings, storage, sites, workers, lanes, scheduler, cleanup; the lane control route needs the token Release 19 slice A3. GET /api/ops/{settings[?key], storage, sites, workers, auto-queue, scheduler, cleanup/} answer what each page already builds (getSettings, buildStorage, listSites, buildWorkersPayload, buildAutoQueueStatusPayload, buildSchedulerStatusPayload, and a new loadCleanupRow — loadCleanupSummary's row for one channel), behind the token, through one readRoute door (_read.ts) that refuses unknown query keys and redacts every secret-named string (a remote worker's token). The CLI reads them as `get settings []`, `get storage|sites|workers|auto-queue|scheduler`, `get cleanup `. /api/auto-queue/control was the one route that starts and stops lanes with no gate. It answers opsAuth now. Its one UI caller, the /operations lane header's Start/Drain/Stop, calls startAutoQueueAction / stopAutoQueueAction and a new drainAutoQueueAction instead (server actions carry Next's origin check; a page holds no token). The six e2e specs that drive the route send the test token. Co-Authored-By: Claude Opus 5.5 (1M context)