docs(board): 33 gigs, and the second supervisor nobody had told

Records the catalogue jump and, more usefully, why the midiviz complaint
survived its first fix: there were two things restarting it, and the latch only
answered one. The unit's Restart=always came from a real bug (an unplugged LCXL
exiting cleanly and taking the window with it for a session), so the fix had to
keep that and still let a person close the thing -- which is what a distinct
exit status is for.

Also writes down the tree's deliberate half-state, since a cold reader running
pytest in the shared checkout will hit an aborted collection that is fixed on
master and looks alarming otherwise.

TASKS_DUMP.md

@@ -2445,3 +2445,51 @@ with eyes on it.
 Worth keeping from the exchange: cross-checking caught one error in each
 direction. Neither session would have found its own
 ([[feedback_check_the_instrument_first]]).
+
+## Catalogue: 25 gigs -> 33, and midiviz actually closes now (2026-09-05, late)
+
+**Eight gigs drafted** from backlog.md's own anchors (`3a7c660` in Tidal,
+`201defc` in www). `gigs_total` 25 -> **33**, `tracks_total` 81, and
+`recorded` rose 40 -> **44** because the new gigs brought their recordings
+into the map. Not ear-verified, and every file says so in `durationBasis`; no
+`section`, no timecodes, `stage`/`bpm` null where the source has none.
+
+**13 unresolved names are IN the files** (`resolved: false`, `file: null`),
+not dropped — opal-2025 holds 6 of them and is the one worth a human pass.
+The three `confirm:`-flagged gigs are still withheld by design.
+
+**midiviz closes and STAYS closed** (`fbf3414`), and is deployed live. There
+were **two** supervisors, and convincing only one looked exactly like
+convincing neither:
+
+| supervisor | told by |
+|---|---|
+| `rig_units.ensure()` | the `$XDG_RUNTIME_DIR` latch |
+| **systemd `Restart=always`** | **`RestartPreventExitStatus=78`** |
+
+My earlier claim that "a clean exit is never restarted" was true of the OLD
+unit and stopped being true when the peer session switched it to
+`Restart=always` — correctly, since an unplugged LCXL used to exit cleanly and
+leave the window gone all session. midiviz now exits 78 for a deliberate close
+only, and a test asserts the code and the unit agree on that number.
+Tray: **Rig ▸ Close ▸**, opt-in per launcher so Ardour/Pulsar/SuperCollider can
+never appear there.
+
+Verified live, not just unit-tested: closed it, waited 9 s (two `RestartSec`
+windows), still inactive with no surviving process, and `rig --ensure` printed
+*"CLOSED by hand — leaving it alone"*. Latch then cleared, window back up.
+
+### The shared tree is in a deliberate half-state
+
+It is still on `claude/rig-streamline` @ `5468507` because the gigs agent was
+live-editing three files in it and switching branches under a working agent is
+the hazard this file keeps warning about. So only the FOUR files the services
+execute were deployed, via `git checkout master -- …`:
+`tools/bridge/midiviz.py`, `tools/bridge/launchers.py`, `perf-tray.py`,
+`tools/rig_units.py`. Everything else in the tree that "differs from master"
+differs only because that HEAD is 26 commits behind — including a suite-killer
+(`test_setlist.py`'s module-level `sys.exit`) that is fixed on master, so
+**running pytest in the shared tree still aborts collection**. Not a
+regression; run it on master.
+
+The full switch is still the two commands recorded above, and still PLN's call.