docs(log 040 + board): the wrong cut wore the cover art, and the resume points

Captain's log 040 and the pre-compact resume state.

The log's finding, in one line: OPAL-26 sounded broken on SoundCloud while every
rendered file was correct to the millisecond, because SoundCloud will not replace
audio behind a permalink and our uploader is idempotent — so every re-upload
under a fresh slug ADDED a copy. The whole Aug-12 first cut was still up beside
the six correct v4 ones. And the tell PLN was actually reacting to: eight of the
fourteen stale uploads wear cover art and not one of the correct six does. The
superseded audio looked finished. His ear was reading the artwork.

Two lens lessons in it, both recurring. 'Matches to the second' was a tolerance,
not a verdict — SoundCloud reproduces full_duration to the MILLISECOND, so at the
±1.5s a transcode seems to deserve, the Aug-12 and v4 cuts of /wap are the same
track, which is how six stale uploads passed a duration check in August. And roll
call cannot find a stray, because a check that only looks where it expects a
thing can never find that thing somewhere else.

Plus the git one, which cost four commits: this is a shared checkout, another
session moved HEAD between two of my commits, and 'git push origin master' from a
non-master HEAD is a silent no-op that exits 0. The PUSHED in my own log was
something I wrote myself.

The board now carries R1-R5 written for a reader with zero context: exact
commands with paths, the artifacts already on disk so nothing is regenerated,
the three decisions that are PLN's and not a machine's, and the shared-checkout
hazard spelled out before anyone commits into it again.

TASKS_DUMP.md

@@ -1883,3 +1883,145 @@ master's tracked copy). Nothing was reset or discarded in a tree shared with
 live sessions. **Before merging or rebasing that branch, expect these to
 double-apply** — `rm` the untracked ones and check out the modified ones from
 master rather than committing them twice.
+
+---
+
+## 🔴 RESUME POINTS — written 2026-09-05 pre-compact, for a session with ZERO context
+
+### R1 · Finish the OPAL-26 SoundCloud release  ⟵ *the only shipping thread*
+
+**State:** the account holds 6 correct v4 tracks, 14 stale Aug-12 uploads, and 7
+v4 tracks never uploaded. Artwork for all 101 arty tracks is archived. The delete
+manifest is generated and dry-run clean. **Blocked on PLN's `--go` only.**
+
+Artifacts (all already on disk, do not regenerate):
+- `~/Work/Sound/Prod/sc_pull_2026-09-05/catalog.json` — the account, 132 tracks
+- `~/Work/Sound/Prod/sc_pull_2026-09-05/artwork/` — 101 originals
+- `~/Work/Sound/Prod/sc_pull_2026-09-05/opal26_generation_verdicts.json`
+- `~/Work/Sound/Prod/sc_pull_2026-09-05/opal26_delete_manifest.json` — 14 entries
+- `~/Work/Sound/Prod/Opal26_master/release/opal26_plan.json` — 15 entries, exact
+
+**Step 1 — delete the 14 (NEEDS PLN'S APPROVAL FIRST):**
+```
+cd ~/Work/Sound/tidal-ears
+.venv/bin/python -m tidal_ears.cli master sc delete \
+  --manifest ~/Work/Sound/Prod/sc_pull_2026-09-05/opal26_delete_manifest.json
+# ^ dry run, expect "14 deletable, 0 refused". Add --go ONLY with his word.
+```
+⚠️ **The delete transport is UNPROVEN** — no DELETE has ever been issued from this
+codebase against api-v2, and `POST /tracks` 400s forever on that path. If `--go`
+returns 403 with a captcha body, wire `delete_manifested()`'s existing `deleter`
+callable to `scbrowser.Browser.api("DELETE", …)`; there is no `--transport` flag yet.
+
+**Step 2 — re-verify, then upload the 8 (7 missing + take-five once its slug frees):**
+```
+cd ~/Work/Sound/Tidal
+python3 armada/tide-table/sc_generation_lens.py \
+  ~/Work/Sound/Prod/sc_pull_2026-09-05/catalog.json \
+  --master ~/Work/Sound/Prod/Opal26_master \
+  --plan ~/Work/Sound/Prod/Opal26_master/release/opal26_plan.json \
+  --newest tracks_v4_streaming
+python3 armada/tide-table/check_permalinks.py \
+  ~/Work/Sound/Prod/Opal26_master/release/opal26_plan.json
+cd ~/Work/Sound/tidal-ears && .venv/bin/python -m tidal_ears.cli master sc form-album \
+  --plan ~/Work/Sound/Prod/Opal26_master/release/opal26_plan.json \
+  --sharing private --state ~/Work/Sound/Prod/Opal26_master/release/sc_state.json --go
+```
+**PASS `--state` THIS TIME.** It never has been, which is why the account was the
+only record. Re-run `sc pull` after, not just during — uploads have been accepted,
+verified, and then vanished.
+
+**Step 3 — re-attach the covers.** They live only on the tracks being deleted.
+Artwork can only be SET via `scwrite.py`'s `sc album` PUT path; `scform.py` has no
+artwork verb. Source images: `sc_pull_2026-09-05/artwork/`.
+
+**Step 4 — fill `audio:` in `www/content/lives/2026/opal-festival-2026.md`** (empty
+today), then the catalog's release links behind it.
+
+**Open decisions, both PLN's:**
+- **`/desire`** is the one manifest entry that is not a duplicate. Desire was cut
+  from the release so no v4 replacement exists — deleting it removes Desire from
+  SoundCloud entirely. The FLAC is safe locally (`tracks/`, `tracks_v3/`). Cut from
+  the release ≠ off the platform.
+- **`/opal26-remix`** (4273.548 s, unidentified) and **`/opal26_master_raw`**
+  (5207.454 s, the raw master) are deliberately NOT in the manifest.
+- **Suffix policy:** OPAL's uploads are unsuffixed; CosmicFest used
+  `--permalink-suffix`. Decide before uploading more, or the two gigs stay
+  inconsistent forever.
+
+### R2 · PLN's hands, 2 minutes, and it is currently FAILING
+
+```
+python3 tools/check-mix.py     # expect: 0 silent orbits
+```
+Right now it reports **Tidal 07 / 08 / 10 / 12 at -inf dB** in the saved
+`Tidal Live` session — the **third** recurrence (04-08+12, then 02+10, now these).
+Raise those four faders in Ardour **and SAVE**, then re-run until clean. Ardour
+meters are POST-fader, so a shut fader looks identical to Tidal emitting nothing.
+
+🔎 **Lead worth one look:** the unmerged branch `claude/rig-autoconnect-regression`
+(1 commit, `a4da75e`) is *"the two reconcilers that lived only in $HOME — five
+orbits reached no Ardour track"*. Five orbits reaching no track and four orbits at
+-inf may be the same wound. Check before merging or discarding that branch.
+
+### R3 · Catalog coverage — 14 gigs uncatalogued (unstarted, mid effort)
+
+`catalog_view.json` covers 25 gigs; `www/content/lives/` has **39** gig pages
+(2022:2 · 2023:4 · 2024:14 · 2025:14 · 2026:5). Missing: algolia-fdlm-2022,
+mephisteuf, 38c3-chaos-music-club, 38c3-house-of-tea, opal-festival-2024,
+toplap-solstice-2024, 39c3-house-of-tea, 39c3-toilet-rave, algorave-lyon, ensad,
+opal-festival-2025, toplap-fromscratch-2025, ete-surprise, le-vortex.
+The hard part is the track↔gig mapping, not the parsing — see
+`feedback_locate_matrix_method` (never trust a low-z locate, ear-gate it).
+Also still open from the reconciler itself: 3 DIVERGENT tracks (Blue Gold, Love
+First, Clameur) and 2 conflicts (Atari-ght, Ere de Jeu).
+
+### R4 · Two parsers disagree about what a gig track IS
+
+`tools/gig-metadata.py` vs `armada/tide-table/build_gig_tracksjson.py`. The latter
+drafts NEW pages and never computes the `styleDistribution` that `LiveEvent.js`
+renders; they also disagree outright on REVOLUTION's style (jazz vs nujazz). Until
+one owns the concept and the other calls it, use `gig-metadata.py --only-played`
+for a REGEN of an existing gig. `feedback_one_parser_per_concept`.
+
+### R5 · Small, named, unblocked
+
+- **`test_orphan_orbits.py` FAILS** (16 rows, expects 13) and
+  **`test_lcxl3_display.py` collects ZERO tests**. Both pre-existing. A test file
+  that cannot fail is not a passing test. The 16-vs-13 smells like the corpus grew
+  and the fixture did not — derive the expectation, don't freeze it.
+- **`mix/` overlaps tidal-ears' mastering CLI.** Committed (`a14aee7`) so the
+  overlap is visible in one tree rather than half-invisible in an untracked dir.
+  PLN's call which one is the spine.
+- **Three captain's logs claim `log: 025`**, which `025-fifteen-intros` owns, and
+  they use dated filenames against the README's `NNN-slug.md`. Their dates
+  (2026-08-14) land them BEFORE 039, so the next free numbers read out of order.
+- **`TODO.md` is deliberately untracked** — a paused 2026-06-22 perf/thermal board
+  would be a second answer to "what's next". Fold its live items in and delete, or
+  keep it outside git.
+- **www `feat/paradiso`**: 1 unmerged commit (`b130192` "Paradiso: v2").
+
+### ⚠️ SHARED-CHECKOUT HAZARD — read before committing anything in Sound/Tidal
+
+`/home/pln/Work/Sound/Tidal` is checked out on **`claude/rig-streamline`** (another
+session, 7 commits ahead of master, tip `c7ac174`). **Verify HEAD before every
+commit** (`git rev-parse --abbrev-ref HEAD`) and **verify the push moved the
+remote** (`git rev-list --left-right --count origin/master...master` must read
+`0 0`) — `git push origin master` from a non-master HEAD exits 0 and pushes
+nothing. Commit to master through a temporary worktree; never `git checkout` in
+that tree, other sessions and PLN have live uncommitted work in it.
+
+That branch also holds **uncommitted duplicates of work already on master** —
+`tools/check-mix.py`, `tools/gig-log.py` (modified), `tools/ardour_session.py` and
+`armada/tide-table/build_delete_manifest.py` (untracked). Before rebasing or
+merging it, `rm` the untracked ones and restore the modified ones from master, or
+they double-apply. Also uncommitted there and **not ours**:
+`copycat/veridis_quo.tidal`, `perf-tray.py`, `tools/rig_units.py` (the other
+session) and PLN's own `piment_bresilien` / `sunny_side_up` / `rose_rouge` /
+`tous_ensemble` `.tidal` edits.
+
+### 🧹 Branch hygiene done this session
+Pruned (merged, verified): `claude/button-column-migration`, `claude/lcxl3-surface`,
+`tidal-to-tracks-tooling`, and tidal-ears' `claude/sc-gated-delete`. Surviving with
+reason: `claude/rig-streamline` (active), `claude/rig-autoconnect-regression`
+(1 unmerged commit, see R2), www `feat/paradiso` (1 unmerged).

armada/tasks/040-the-wrong-cut-wore-the-cover-art.md

+---
+log: 040
+title: "The wrong cut wore the cover art"
+date: 2026-09-05
+task: "A3 (SoundCloud uploader) · catalog coherence · check-mix session resolution"
+tags: [soundcloud, release, opal2026, tooling, catalog, forensics, git]
+shareable: true
+---
+
+## Cap (what & why)
+
+PLN opened the day with a complaint and a question: *"on soundcloud i see our
+private cuts of OPAL and they are badly cut begin/end unaligned we need to
+reupload and save their images tho wheres out catalog at?"*
+
+Two jobs, then. Find out why the OPAL-26 release sounds wrong on SoundCloud, and
+report where the catalog actually stands. The board had a theory for the first
+one already — a note saying `/take-five-drops` had uploaded "24 s short" while
+every other track "matched to the second". That theory was wrong in both halves,
+and finding out how it was wrong is the whole log.
+
+## Manœuvre (how)
+
+**First, refuse to re-render.** The obvious move on "the cuts are bad" is to
+re-cut. Before touching `apply_boundaries.py` I measured what was on disk:
+every one of the 14 `tracks_v4_streaming/` FLACs matches `segments_v4.json` to
+the millisecond, CRIME's `pad_end_s: 2.0` rest and REVOLUTION's 5 s reverb tail
+included. The upload plan's titles, files and durations line up exactly. The ear
+file's mtime (19:33) postdating the renders (19:18) turned out to be the
+`release_signoff` record being written, not a late boundary edit — I checked that
+specifically because it looked like a smoking gun and would have justified a
+re-render. **The local pipeline was clean. The fault had to be downstream.**
+
+**Then read the account instead of reasoning about it.** `master sc pull --out
+… --artwork` — read-only, one pass, and the existing token was still live so no
+login was needed. 132 tracks, 10 playlists, **101 artwork originals** to
+`~/Work/Sound/Prod/sc_pull_2026-09-05/`. That single command was the artwork
+rescue PLN asked for, the diagnosis, *and* the state recovery: no
+`sc_state.json` had ever been written, so the account itself was the only record
+of what had landed.
+
+**Then build the lens, and get it wrong three times.** `sc_generation_lens.py`
+compares every upload's duration against every local render generation.
+Three defects in my own tool, each left in as a guard:
+
+1. **Filenames are not authoritative.** A title of `Perfect <3` uploads to
+   `/perfect-3`, so a filename-derived slug reported two live tracks as NOT
+   UPLOADED. Titles now come from the release plan — what the uploader used.
+2. **Streaming and club masters share every duration**, so offering both as
+   rival candidates marked all 14 AMBIGUOUS. Identical generations collapse into
+   one; they are one set of boundaries wearing two masters.
+3. **A duration collision is not provenance.** `/au-revoir-lord-toyota` (2020)
+   and `/saria-demo` (2025) matched OPAL cuts to 50 ms by pure chance on a
+   132-track account. Strays are now hard-gated on a date floor — nothing older
+   than the oldest render on disk can be a cut of it — and the discards are
+   printed, not silently dropped.
+
+**Then the second pass, which is the one that earned its keep.** Roll call —
+checking each *expected* permalink — structurally cannot see a stale cut hiding
+under a different slug. So the lens also scans the whole account by duration.
+
+## Prise (findings / artifacts)
+
+**SoundCloud accumulates generations.** It will not replace the audio behind a
+permalink (paid feature), and our uploader is idempotent by permalink, so every
+re-upload under a fresh slug *adds* a copy. Nothing ever replaces anything. The
+entire Aug-12 first cut of OPAL-26 was still up — all fifteen tracks — beside
+the six correct v4 ones:
+
+| | count | |
+|---|---|---|
+| correct v4 | 6 | `/wap` `/there-s-something-about-drums-3` `/you-my-sunshine` `/piment-bresilien` `/eh-ouais-je-funk` `/ghosts-in-the-toilets` |
+| stale | **14** | `/take-five-drops` + 13 under junk slugs: `/opal26-wap`, `/12-mafia-12`, `/ceci-nest-pas-une-bombe`, `/you-my-sunshine-5`, `/desire`, `/15-la-revolution-sera-samplee-15` … |
+| never uploaded | 7 | ceci-n-est-pas-une-bombe, am-i-doing-it-right, perfect-3, gimme-acid, mafia, vague-de-crime, revolution |
+
+**`/take-five-drops` was never "24 s short". It was 417.000 s — to the
+millisecond, the Aug-12 cut.** Not truncation. A different generation.
+
+**Duration is far sharper than anyone had assumed.** SoundCloud's
+`full_duration` reproduces the source *to the millisecond* (477.600 s uploaded →
+477.600 s reported). So the working tolerance is **0.05 s**, not the seconds a
+transcode seems to deserve. At ±1.5 s, the Aug-12 and v4 cuts of `/wap`
+(361.0 vs 361.6) and `/perfect` (315.0 vs 314.6) are the same track — which is
+precisely how six stale uploads passed a duration check in August.
+
+**Built:** `armada/tide-table/sc_generation_lens.py` (`fe8ec16`),
+`build_delete_manifest.py` (`bd3a680`), and `master sc delete --manifest` in
+tidal-ears (`ab3285b`) — the "explicit, separate call" `scwrite.py`'s refusal
+message had been promising since August. Dry run: **14 deletable, 0 refused**,
+audited independently.
+
+**Catalog** (`f1fa484`): `catalog.generated.json` went 73 → **81 tracks**,
+23 → **25 gigs**, now identical to the view with no silent drops, OPAL and
+CosmicFest both present. Root cause: `AS_OF = "2026-06-06"` was a hardcoded
+literal in `build_catalog_view.py`, so every rebuild re-stamped the same frozen
+date. Coverage gap that remains: 39 gig pages in www vs 25 gigs catalogued.
+
+**check-mix** (`27f2351`): the argv-only session resolution is fixed by
+extracting gig-log's 3-rung ladder into a shared `tools/ardour_session.py`.
+169 gig-log tests pass unchanged. It then failed on its first real run —
+**Tidal 07/08/10/12 at -inf dB in the saved session**, the third recurrence.
+
+## Sel (the shareable learning)
+
+**The wrong cut was the one wearing the cover art.** Eight of the fourteen stale
+uploads had artwork. Not one of the six correct v4 tracks did. So browsing his
+own account, the superseded audio looked *finished* and the correct audio looked
+like drafts. PLN's ear was reading the artwork, and the artwork was lying.
+
+Artwork is a completeness signal, and completeness signals attach to whatever
+someone last fussed over — which is usually the *older* thing, because it had
+more time. When you ask "is this release done?", read the audio, never the
+chrome around it.
+
+**Two lessons about lenses, both old friends.** The board's "matches to the
+second" was never a verdict, it was a tolerance — the instrument was the bug,
+again. And roll call cannot find a stray, because a check that only looks where
+it expects a thing can never discover a thing somewhere else. Absence needs
+proof.
+
+**And one about git.** Four of my commits landed on a concurrent session's
+branch, because this is a shared checkout and another session moved HEAD under
+me between two commits. I did not notice for an hour, because **`git push origin
+master` from a non-master HEAD is a silent no-op** — it pushes the local
+`master` ref, which was already current, prints nothing, exits 0. The "PUSHED"
+in my log was something I had written myself. Recovered by cherry-picking onto
+master through a temporary worktree, touching neither their branch nor the live
+tree. The global rule already said "use a worktree in shared repos"; I had
+applied it to delegated agents and quietly exempted the orchestrator, which is
+the session that commits most.
+
+## Hameçon (hook)
+
+*"Your ears were right and the audio was fine."* A release that sounded broken,
+where every rendered file was correct to the millisecond — because the platform
+had quietly kept every draft, and the drafts were the ones with the covers on.
+
+Runner-up, for the git audience: *the commit that reported success for pushing
+nothing.*
+
+## Sillage (what it unlocks)
+
+- The delete manifest is generated and dry-run clean; `--go` is PLN's call. Once
+  the 14 stale uploads go, the 7 missing v4 tracks can be uploaded and the
+  covers re-attached, which finally fills OPAL's empty `audio:` field on the
+  website and the release links behind it in the catalog.
+- `sc_generation_lens.py` is reusable and CosmicFest will need it — that release
+  went up under a suffix policy while OPAL's uploads are unsuffixed.
+- The check-mix fix makes PLN's parked two-minute fader check *runnable*, and it
+  went from unrunnable straight to failing. Assembling checks into one caller
+  revalidates them under fresh conditions.
+- The unmerged branch `claude/rig-autoconnect-regression` holds one commit —
+  *"the two reconcilers that lived only in $HOME — five orbits reached no Ardour
+  track"*. Five orbits reaching no track, and four orbits at -inf, are close
+  enough to be worth checking whether they are the same wound.