log 049: four green lights and a dead lens

Captured at task close, not reconstructed. The learning worth keeping is
that a green light is a claim about liveness and liveness was never the
question — and that a test in this repo had pinned the MIDI re-resolve
design so firmly that 'cheap by construction: an idle surface costs
nothing at all' went un-re-examined until an idle surface cost a gig.

TODO.d/2026-09-25-post-gig-belt-and-braces.md

@@ -34,3 +34,20 @@ Each line pays off a `TODO_GIG` "NOT TONIGHT" entry or a loss measured in
 - [ ] **7. Propose, PLN's call:** a `Mix` track inside Ardour (input =
       `Master/audio_out 1+2`, armed) so the mix lands in the same take at stem
       grade. Solves "one orbit unarmed", not "forgot REC".
+
+---
+
+## Done 2026-09-25 — and the three things only PLN can do
+
+Commits `ce66317` · `6a94b47` · `187b17c` · `83ced8f` · `dd851bc`.
+Log: `armada/tasks/049-four-green-lights-and-a-dead-lens.md`.
+
+- [ ] **Arm `Tidal 08` in Ardour, then Ctrl+S.** The gate says NO-GO until you
+      do. Same pass: raise the nine faders sitting at −∞ and save, so the
+      session boots the mix you play.
+- [ ] **`systemctl --user restart gig-log`.** The rebind only helps a reader
+      started after it. The current one is still holding the dead subscription
+      from the 24th — `tools/gig-log.py status` now says so in words.
+- [ ] **Install the black box unit:** `tools/rig-install.sh` (or
+      `systemctl --user link ~/Work/Sound/Tidal/tools/gig-record.service`).
+      Until then the Bridge shows `black box — unit not-found`, which is honest.

armada/tasks/049-four-green-lights-and-a-dead-lens.md

+---
+log: 049
+title: "Four green lights and a dead lens"
+date: 2026-09-25
+task: "TODO.d/2026-09-25-post-gig-belt-and-braces.md, 1-5 of 7"
+tags: [tooling, rig, gig, surface]
+shareable: true
+---
+
+## Cap (what & why)
+The 2026-09-24 set went well and the rig held — 5 xruns in 94 minutes, zero
+throttle. What it lost, it lost silently: 17m44s of audio to a late REC, one
+whole orbit to a record button nobody had ever looked at, and every control
+move of the night to a MIDI lens that was reading a dead wire. PLN, the morning
+after: *"we should have belt sus and logs validation before a gig and a better
+'tableau de bord' to confirm all is armed"* — and then, *"too bad these TODOs
+were not done 20 days ago. second best time is now"*.
+
+## Manœuvre (how)
+Five changes, one commit each, every one of them measured against the gig's own
+logs before a line was written. Two turned out to be deletions.
+
+## Prise (findings / artifacts)
+- **The `^42` double-press has a mechanism.** PLN: *"moving from track X to Y, i
+  had to press states like ^42 twice to 'apply their state'"*. `carry_values`
+  deliberately sends no CC for buttons (a 0 into a `midiOn` gate is
+  `someCyclesBy 0` — the stream stops), and it also cleared the driver's latch.
+  Both halves cannot be right: nothing is sent, so Tidal's bus keeps 127, while
+  the driver now believes 0. The next press computes `not False` = ON and
+  re-sends 127 into a bus already at 127. Inaudible — and the press after it
+  closes the gate he was opening. The journal logged **37 of these across 14 of
+  the set's 15 track changes**. Fixed by deleting the clear.
+- **`Tidal 08` was never armed**, alone among twelve. `check-mix.py` now audits
+  the record button and `check-gig.py` says NO-GO on it, naming the track.
+- **gig-log's MIDI lens: two independent faults.** The re-resolve lived *inside*
+  `for line in self._proc.stdout`, so it only ran after a line arrived — a
+  silent port blocks that iterator forever, and `aseqdump` never exits. Now a
+  timed `select`. And the liveness test itself was wrong: 38 hours later the
+  port still resolved to the same name at the same address, because
+  lcxl3-driver had restarted and the kernel reused the client number. So ask
+  the kernel's own wiring table. Verified live: client 133 was `Connecting To:
+  129:0, 14:0` and gig-log's aseqdump, client 135, was not in the list.
+- **The black box had never run.** `tools/gig_record.sh` has existed since OPAL
+  and is item 9 of the RUNBOOK. No unit, no row in `rig_units.py`, no
+  `recordings/` directory. It also recorded the wrong thing: the default sink,
+  which was the laptop card while Master was hand-linked to the UMC. It now
+  follows `ardour:Master/audio_out 1`.
+- **A GATE panel on the Bridge**, rendering `check-gig --json` — 43 checks with
+  their fixes, not a second opinion.
+- **Nothing checked that `rig_units.SERVICES` and `rig-install.sh`'s `UNIT_SRC`
+  name the same units.** Three tests now do.
+
+## Sel (the shareable learning)
+**A green light is a claim about liveness, and liveness was never the question.**
+Every dot on the dashboard was green that night. SuperDirt up, driver up,
+gig-log up, Ardour up, log growing at 1 Hz. And: one orbit recording nothing,
+one lens reading a wire that had been cut nine hours earlier, one safety net
+that had never existed. "Is it running" and "is it doing its job" are different
+questions, and only the first one is cheap.
+
+The second lesson is smaller and sharper. This repo's own test file asserted
+that the MIDI re-resolve *should* run inside the read loop, with the reasoning
+written out: *"cheap by construction: an idle surface costs nothing at all"*.
+That is true, and it is the bug. **"Costs nothing when idle" and "notices when
+idle" are the same code path, and only one of them can be true.** A test can
+pin a design decision so firmly that the decision stops being re-examined —
+so when the gig disagreed with the test, the test was what changed.
+
+And a third, from being wrong in the middle of it: a name is not an identity.
+The first fix rebound on "the port vanished" and "the address moved", and
+*neither* describes what happened. Only the kernel's subscription table did.
+The same morning, `pkill -f "server.py --port 8791"` matched its own command
+line and killed the shell that ran it — the rule against identifying processes
+by name, broken by the person writing the fix for identifying ports by name.
+
+## Hameçon (hook)
+Twelve tools audited this rig: faders, mutes, routing, ghost orbits, sample
+banks, preload coverage, thermals, xruns, clocks, units, symlinks, MIDI paths.
+None of them audited the record button.
+
+## Sillage (what it unlocks)
+- The gate now fails on the two things that cost the last gig, before a note.
+- Ardour's OSC arm lens is still lying (`arm:false` for 76 recorded minutes,
+  `conflict:true` on every record) — diagnose before anything reads it, and a
+  REC reminder must key on the FILES lens only.
+- Open proposal: a `Mix` track inside Ardour, input `Master/audio_out 1+2`,
+  armed — the mix at stem grade, in the same take. Solves an unarmed orbit; the
+  black box is what solves a forgotten REC.