docs: board — session plan for the Opus shift: keystone is the v2 ghost still painting

Ground-truth pass found lcxl-leds-watch.service (the v2 LED watcher) still
running beside the v3 driver — the likely author of the paint-reset mystery —
and midi-autoconnect as the probable Midi-Thru→SC re-adder. Plan: kill the
tug-of-war first, then the A1 label off-by-one, then the generalize thread
PLN named (track-follow + live BPM from the cps knob), then the display
probe with his eyes on the device.

TASKS_DUMP.md

@@ -1014,3 +1014,35 @@ name on ROW 1 (per-control l1 currently EMPTY = wasted best row).
   (names: kick, percs, mel, all). Latch/held state already in self.values.
 - Fader labels → sample identity ("d8 jungle_breaks") still queued from
   earlier; "level [ard]" noise.
+
+## SESSION PLAN — 2026-08-30 (post-2nd-compact; Opus drives)
+
+Ground truth: branch claude/lcxl3-surface @ c462ce7 all pushed; driver pid 273247
+RUNNING with STALE rose_rouge map while PLN plays sept1.tidal; Ardour open direct
+on Tidal Live; all rig services active. NEW FINDING: v2-era
+`lcxl-leds-watch.service` still runs next to the v3 driver — prime suspect for
+the paint-reset mystery; `midi-autoconnect.service` is the likely Midi-Thru→SC
+re-adder. Stale artifacts: `.claude/RESUME.md` is the 08-21 checkpoint (old
+112-item board, several items since done — e.g. "open tidallive direct" ✓);
+`TODO.md` is the June Pulsar-highlighter thread, still real, unrelated.
+
+### Clusters
+| # | Cluster | Items | Note |
+|---|---------|-------|------|
+| S1* | Generalize driver | track-follow, live-BPM cps-knob (`range 60 180 "^29"`), lap reset on track change | PLN-named thread |
+| S2 | Display-semantics probe | bring-up-after-write, per-change bring-up 36h, arr-1 host-text %, name-on-row-1 | one probe, 4 bugs |
+| S3 | Eye-batch bugs | A1 label off-by-one (READ label_control first), stuck relative numeric, mute-combo summary, fader sample-identity labels | |
+| S4 | Edge layers / dawless | Track◀▶ setlist cursor, Page▲▼ pages, Solo/Arm, Mute/Select, SC OSC feed, v2 eval-from-controller | after S1 |
+| A | Ardour | fader-1 EAR TEST (PLN, 2 min), .map hardening, 12 gains at −inf again | |
+| C | Corpus | b5ad8b6 merge after play-through, 17 non-compiling tracks, pvlint 90 sweep | no gig scheduled |
+| M | Armada/comms | render-every-track, SC re-upload (6 artworks), kick rights → yaxu, CosmicFest metadata (PLN) | |
+| H | Housekeeping | v2-service tug-of-war triage (KEYSTONE), untracked-file triage, midimon transport, midiviz d13/14, bridge KWin rule | |
+
+### Recommended sequence (Opus)
+1. **H-keystone (~10 min):** decide paint ownership — RIG UP should stop
+   `lcxl-leds-watch` when the v3 driver runs; identify the Midi-Thru→SC re-adder.
+   Kills two standing mysteries and stops undefined paint behaviour under S1 work.
+2. **S3a (~5 min):** read `label_control`, fix the A1 off-by-one — same code S1 relabel touches.
+3. **S1 GENERALIZE (the thread):** track-follow + live BPM + lap reset; restart driver on sept1.
+4. **S2 probe** with PLN's eyes at the desk.
+Needs PLN async: Ardour fader-1 ear test; mute-group naming confirm; yaxu question; CosmicFest metadata.