docs: the receipt for everything gig prep deliberately parked

Eleven items, each with the acceptance test it is waiting on and the reason it
was left rather than done. Ordered by what it costs to carry, not by effort.

The two with a real deadline are at the top: retiring tools/gig-up.sh, which is
only still present because nothing is deleted before its replacement has run for
real AND because it is the transcription check's reference; and folding
gig_gate's scaffolding into gig-preflight, which was refused this week for
buying tidiness rather than behaviour.

Written now rather than after the gig because the reasons are what rot. The list
of files is recoverable from git; 'we left this because a corpus rewrite with
Pulsar open destroyed a rehearsal edit once' is not.

docs/2026-09-25-post-gig-cleanup.md

+# After Thursday: what the gig prep parked
+
+**Status:** outline, written 2026-09-22. Nothing here is for before the gig.
+
+Every item below was found and deliberately left. The rule that produced the
+list: *nothing is deleted until the thing that replaces it has run for real*, and
+*a gig week is not when you refactor the boot path*. So this is the receipt for
+those decisions, with the acceptance test each one is waiting on.
+
+Ordered by what it costs to leave, not by effort.
+
+## 1. Retire the bash gate
+
+`tools/gig-up.sh` (620 lines) is superseded by `tools/check-gig.py` and carries a
+banner saying so. It is kept for two reasons, and **both expire on Thursday
+night**: nothing is deleted before its replacement has run on a real launch, and
+it is the reference for the transcription check
+(`test_every_bash_gate_check_survived_the_port` reads ITS check names and
+requires each to appear in the new probe table).
+
+*Do:* `git rm tools/gig-up.sh`. The test skips itself when the file is gone —
+deliberately, with a note that the probe table becomes the only source of truth
+at that moment. Re-read that note before deleting; it is the last chance to
+notice a check nobody ported.
+
+*Acceptance:* suite green with the skip, `grep -rn 'tools/gig-up' ` clean apart
+from dated snapshots (`MORNING.md`, `LAST_THIRD.md`, `TASKS_DUMP.md` are history
+and stay wrong on purpose).
+
+## 2. Fold the gate's scaffolding into gig-preflight
+
+`gig_gate.py` holds the verdict vocabulary; `gig-preflight.py` still has its own
+copy of `add()`, the OK/WARN/FAIL constants and the report renderer. The two
+agree today — which is *why* the machine layer absorbed with no rewrite — and
+that agreement is currently held by a test rather than by a shared import.
+
+*Do:* have `gig-preflight.py` import `gig_gate`, delete its duplicate ~60 lines,
+keep its standalone CLI contract (`rig-doctor.py`, `RUNBOOK.md` and its own
+docstring all reference it directly).
+
+*Acceptance:* `test_preflight_json_still_speaks_the_shared_record_shape` passes
+unchanged, and `gig-preflight.py --json` output is byte-identical modulo live
+values.
+
+*Why not now:* it edits a file the boot path calls, to buy tidiness rather than
+behaviour. That is exactly the trade a gig week should refuse.
+
+## 3. The duplicate FIXME lines
+
+90 files carry 114 duplicated `FIXME(#54/#94)` comment lines — the migrator
+stacked a fresh note on every pass before it learned to dedupe. Cosmetic: the
+tool is idempotent now, so the count is frozen.
+
+*Do:* one pass removing exact-duplicate consecutive FIXME lines. **With Pulsar
+closed** — a corpus rewrite with his editor open destroyed a rehearsal edit once
+already, and separately had a save from a stale buffer silently undo a
+migration.
+
+*Acceptance:* `silent-eval --seeded` verdicts diffed per file against
+`docs/2026-09-22-catalog-playability.tsv`: zero changes, the same way the four
+migrations were proven.
+
+## 4. The 36 PV011 bus-slot collisions
+
+Outside the 17-track set, so they cannot affect Thursday. Inside it, zero.
+
+*Do:* `pvlint --setlist` over the whole corpus, resolve per track. Expect real
+musical decisions, not mechanical ones — two tracks wanting the same crush bus is
+a choice about what they sound like together, which is why this was not
+auto-fixed.
+
+## 5. One BPM parser
+
+`lcxl3-driver.parse_bpm` and `setlist_samples.declared_bpm` are the same regex
+against the same `setcps (BPM/60/4)` form. The driver's copy is load-bearing
+during a set (the tempo readout), which is why it was left alone with a comment
+saying it collapses post-gig.
+
+*Acceptance:* `test_one_tidal_parser.py`'s `KNOWN_DEBT` shrinks by one entry —
+that list only shrinks, and paying an entry off is required to be recorded.
+
+## 6. Enable Ardour OSC, and close the live-fader blind spot
+
+`use-osc="0"` in `~/.config/ardour8/config`; port 3819 is already set. So **no
+tool can read the LIVE faders** — `check-mix` reads the last SAVED session, which
+is why the gate states that limit in those words every single run.
+
+This is the root cause of a recurring class: the D-row faders are MIDI-learned to
+CC 77-84, the saved file and the physical desk drift independently, and an orbit
+at −inf was caught three sessions running with a different set of tracks each
+time. With OSC on, `ardour faders` becomes a live question instead of an archival
+one.
+
+*Do:* enable OSC, teach `check-mix.py` to prefer the live read and fall back to
+the session file, and keep the caveat text for the fallback path.
+
+*Acceptance:* move a fader on the desk without saving; `check-mix` reports the
+new value. That is the leg that breaks, so it is the leg to test.
+
+*Risk:* OSC is a new listener on a box that performs. Enable it on a calm
+evening and leave it on through a full rehearsal before trusting it at a venue.
+
+## 7. `ete_a_mauerpark` d9
+
+Its `mask` shares `^58` with d6's gate. Structural, not a rename: d9 has no
+button slot in the authored grid, so the fix is gSel work, not a migration. Left
+because the track plays correctly — the two controls simply move together.
+
+*Acceptance:* `tools/button-column-audit.py` clean for that file, with d9's mask
+on a slot no other orbit claims.
+
+## 8. midiviz: the rest of the state wire
+
+Steps 3–5 of `docs/2026-09-22-midiviz-state-wire.md`. Step 2 shipped: the driver
+publishes `surface-state.json`, and midiviz draws the three family filters as
+zones with the detent at 64.
+
+- **Step 3 — the full 48-cell layer.** The publisher already sends all 32
+  analogue values; `_paint_state` filters to `family_filter` with a one-line
+  guard. Widening it needs dirty-cell repaint so the layer stays free at rest:
+  `seq` says whether anything moved, the value diff says which cells.
+- **Step 4 — orbit activity from `eval-events.jsonl`.** Already being written,
+  one object per ctrl+enter, and the driver already reads it. Tailing it tells a
+  viewer which orbits are live in the loaded track, which answers the real stage
+  question: *I turned the knob and nothing happened — is the pattern silent, or
+  is the knob not wired?* Surface state answers the second half; this answers the
+  first.
+- **Step 5 — only if still wanted:** beat-level activity needs an OSC feed from
+  SuperDirt. The only genuinely new component in any of this, and the only one
+  that touches the audio process. Weigh it against everything else on this page.
+
+Also inherited from the theme work: **`--theme` is one-shot**, so a converge
+brings `dark` back. Persist the choice rather than passing it.
+
+## 9. The two failing tests
+
+Both pre-existing, both real, neither mine:
+
+- `test_one_tidal_parser::test_no_new_file_parses_tidal_itself` flags
+  `tools/fold-orbits.py:96,100`. Those regexes match **PipeWire node names**
+  (`node.name = "..."`), not Tidal syntax, so this is a false positive in the
+  ratchet. The escape hatch the ratchet itself documents is a `tidal-parse-ok`
+  comment on the line — *not* a `KNOWN_DEBT` entry, since that list only shrinks.
+  Read the ratchet's own note before deciding: a false positive it cannot see is
+  exactly how a real one would hide.
+- `test_gearbox::test_label_falls_back_to_the_raw_key` — the dash fallback
+  changed in `6fb0f98` and the test was not updated. Decide which is right; the
+  test is the older statement of intent.
+
+## 10. Two setlists called opal2026
+
+`setlist_opal2026.txt` exists at the repo root (Sep 6) **and** in `armada/`
+(Sep 17). The root copy is the one the launcher used to generate `preload.scd`
+from on every launch, which is how Thursday's set came within one button-press of
+being warmed against the wrong gig. It is now only a last-resort fallback.
+
+*Do:* keep one. `armada/` is where setlists live (`setlist_thu24.txt` is there),
+so the root copy goes — but check nothing else reads it first, and note that
+deleting it makes the launcher's fallback branch unreachable, which is arguably
+the point.
+
+## 11. `preload.scd` is generated, gitignored, and mode 0600
+
+Its own history is the argument for looking at it again: a generated,
+gitignored, never-compared boot file *debuted as a crackle at the venue*. It is
+now compared — `check-preload` runs in the gate and in the launcher's converge —
+but nothing in the repo records what it contained on any given night.
+
+*Consider:* logging the bank count and a hash into the gig-log session file at
+boot. Cheap, and it turns "was it warm?" into a question with an answer after
+the fact. Weigh it against the ethos: an artifact nobody compares against
+reality is worse than none, and this one would be compared by exactly one
+future question.
+
+---
+
+## Not on this list
+
+**The 2h rehearsal, the crush-bus A/B, `sept1`'s controller and
+`ete_a_mauerpark`'s d9 effects.** Those are gig work, they need his ears, and
+they belong in `TODO_GIG.md` where they already are.