CHANGELOG, TODO.d: Sprint 13, the recorder clock fix and what is still open

CHANGELOG.md

@@ -4,6 +4,17 @@ Sprint entries, newest first. Player-facing: what changed about *playing*, not
 about the code. Task IDs reference the L'Armada board; `n/a` where the work was
 unplanned (which, on a gig night, is most of it).
 
+## Sprint 13 — 2026-10-13 · one Quit, and the black box mends its own clock
+
+### Fixed
+- **A tray that vanished under a running rig.** Quit used to close only the tray, and nothing
+  brought it back: dark for 42 h while the rig played. Now `parvagues-sc` starts the tray, and
+  there is one Quit: rig up, it ends the session; rig down, it closes the tray.
+- **A recorder that kept a broken clock.** After an overnight suspend its input timestamps
+  stepped back ~2 h and ffmpeg never exited: 255,000 warnings, 100,000 s segment durations,
+  a 56 s hole in the file. The recorder now notices the broken clock and restarts itself
+  (alone, never the sound, at most every 2 minutes).
+
 ## Sprint 12 — 2026-10-11 · the rig says when its own signal path is dead
 
 Four faults on one afternoon, every unit `active` the whole time: the faders did

TODO.d/2026-10-13-recorder-after-suspend.md

-# The black box after a suspend: found, not yet fixed (2026-10-13)
+# The black box after a suspend: found and fixed (2026-10-13)
 
 Owner saw a sticky "rig: recording is down" at 10:00. Two separate facts, one of them
 a wrong first guess worth keeping out of the next investigation.
@@ -17,7 +17,22 @@ a wrong first guess worth keeping out of the next investigation.
   audio is probably intact; the clock under it is not, and a muxer on a clock that runs backward
   is a plausible source of the 56 s hole. Not proven.
 
-## What to do (measure before mechanism)
+## Resolved the same day
+
+- Restarted `gig-record` alone: the DTS lines stopped, health `ok`.
+- Durable fix in the recorder's own script (`guard_clock` in `tools/gig_record.sh`, tested by
+  `tools/tests/test_gig_record_clock_guard.py`): it watches the new log lines for
+  `Non-monotonic DTS` (zero in a healthy recorder) and restarts the recorder alone, at most
+  once per 120 s. Proven live: one injected line restarted the unit into a fresh segment.
+- Rejected, measured: `-use_wallclock_as_timestamps 1` makes the steady state worse (140
+  "backward in time" lines in 20 s).
+- No health-side grace added: the guard restarts within a second of a bad clock and the new
+  segment lands at once, so a grace would be a screw with no nut. Still open: the unexplained
+  56 s write gap at 09:59:43 (cause unknown, no resume involved). If it recurs, the evidence
+  file in `~/.local/share/parvagues/health-evidence/` is where to start.
+- Not done: nobody has closed the lid on the fixed recorder. A real suspend is the final proof.
+
+## Original plan (kept for the reasoning)
 
 1. Restart `gig-record` alone at a quiet moment: `systemctl --user restart gig-record`. It never
    touches the sound; it costs the second or two between segments, and it gives the recorder a