`live/collab/raph/piment_bresilien.tidal` is modified in the Tidal tree and is
**PLN's own edit** — his friend wants to collab on that track. Not mine, not to
be touched.
---
## AMENDMENT — 2026-09-05, the night the rig became supervised *(veridis_quo + rec lens session)*
### ✅ Shipped (all pushed to master; narrative in `armada/tasks/039-one-list-one-painter.md`)
| commit | what |
|---|---|
| `16c09aa` | **copycat/veridis_quo.tidal** — Daft Punk "Veridis Quo" algoraoke (Am, 116 BPM, 8-phrase organ cat, breaks stack); recognized from MariaAreadne's strudel "Vardis Quo" |
| `96f1322`+`3fdf5b9` | **lcxl3-driver.service** repo-owned unit (retry-forever, follow-mode, `Conflicts=lcxl-leds-watch` = one painter per board); `rig_units.py` gains lcxl3-driver + perf-tray rows + third boot policy `manual`; gear derives from THE list; 63 scripts' exec bits fixed IN THE INDEX |
| `da1c7c7` | OLED lap clock: lap 1 starts at first track LOAD (was process construction — showed 24:27 idle time once the driver became persistent) |
| `d1947ee`+`d3f58c6` | **OLED footer = REC clock** ("REC mm:ss", "?REC" on lens conflict, reverts when stale); repaint dedup on the SENT string + forced full repaint on the 20s reassert |
| `1667bb0`+`619bbae` | parvagues-rig.target regenerated; the pathspec-commit-skips-staged-mode-flips trap documented |
Also this session: the recurring saved--inf Ardour faders struck again (04-08+12, then 02+10 — raise & SAVE);
a ghci ballooned to **15.8 GB RSS** and the OOM killer took Pulsar+Ardour down (box has NO swap).
### 🔴 RESUME — end-to-end REC verification (the one thing left mid-flight)
Everything is deployed and running (`rig_units --status` = 9 units 0 problems), but the final
live proof needs Ardour + PLN:
1. Launch Ardour (any way — GUI or gig-up both resolve now), load Tidal Live.
2. Record ~20 s, stop. EXPECTED: OLED footer counts `REC 0:01…0:20` (no `?`), reverts to
`ParVagues` ≤4 s after stop; gig-log jsonl (`~/.local/share/parvagues/gig-log/gig-*.jsonl`)
shows ONE clean `rec:true→false` pair with monotonic `el`, no `conflict:true`.
3.**Check `arm`/`roll` fields are non-null** — that verifies the unproven guess that OSC
feedback bit 1 makes Ardour send `/rec_enable_toggle`. If still null: try bit 16
(master-section), noted in `OSC_FEEDBACK`'s comment in tools/gig-log.py.
4. If the OLED freezes again despite fresh records: the repaint fix (`d3f58c6`) treated
dedup + ghost-mirror divergence; next suspect is the device-side OLED write path (ACK
handling). The ghost mirror `/tmp/lcxl3-oled-ghost.txt` now mirrors REAL sends — trust it.
### 🔴 RESUME — check-mix.py has the argv-only session-resolution bug
GUI-launched Ardour (bare argv) makes check-mix resolve nothing. Port gig-log's 3-rung
ladder (`FilesLens._session_from_*`, tools/gig-log.py) — the sync comment there flags it.
### 🟡 Rig debts (small, named)
-`perf-tray.service` is the last unit not repo-owned (lives only in ~/.config/systemd/user).
-`tools/parvagues-protect.service` + `parvagues-samples.service` exist in the repo but are
neither installed nor in `rig_units.py` — decide: rig (add row + symlink) or retired (delete).
-`~/.local/bin/parvagues-gear.pre-repo.bak` — old copy, delete when the repo one has run a while.
- ghci 15.8 GB OOM: cause unknown (one-off?). Box has no swap; consider zram/swapfile, and
watch ghci RSS during long Pulsar sessions.
### 🟡 veridis_quo — next jam
First jam validated the rig, not yet the arrangement. Open ears-calls from the file:
superhammond register (`|+ note 12` comment), break density (`loopAt 2` vs `loopAt 1`),
and whether the 16-cycle organ cat breathes right against the slowed (×2) chord riff.
### 🟢 Follow-up unlocked by k:"rec" (unscheduled)
Bridge/HUD REC widget reading the same records — and C2's "rig health: Ardour rec-armed"
OLED row is now half-landed by the footer clock. PLN's ask that started it: *"record X has
track Y playing at time Z"* — the gig-log timeline now answers this (rec records interleave
with track/eval records).
### Superseded / corrections to earlier sections
- EPIC I's "surface driver" design (2026-08-28 amendment): **fully landed and now supervised**
— the driver is `lcxl3-driver.service` with `Conflicts=`, follow-mode paint, OLED labels.
- "PARKED — not to be committed by anyone but PLN": veridis_quo + rose_rouge edits were
validated by play tonight and committed (`e1b75e2`). Still parked: `piment_bresilien.tidal`,
`sunny_side_up.tidal` (pre-session hand edits, unvalidated), `backlog.md` rule unchanged.
- Global CLAUDE.md gained PLN's orchestration rule: main session delegates to explicit-model
