docs(notes): epilog session ear notes, fans and the KeyStep G key

armada/tide-table/performance_notes.md

@@ -830,3 +830,30 @@ video (`REC-20260928-215506-screen.mp4`; the master sits at +3.80 s in it).
   Serif, Take89 (2026-05-22), so Montreuil26 is the stem source for that pair.
 - Naming: **"indeed the real name now is Mafia"** — "Aria Sans Serif" (Montreuil26
   stem dirs) is the old name.
+
+## 2026-11-01 — epilog session on XPS24 (decommissioned laptop), fans
+- **"rose_rouge plays just fine"** on the freshly relaunched rig (SuperDirt -> Ardour ->
+  Pulsar, LCXL3 painting track metadata, Ardour faders following). No sound complaint.
+- **"we're venting high atm ... this doesn't seem to be silent lol"** — fans at ~2400 rpm
+  (thermal-hush's gentle cap) with the package at 67-70C while `performance` mode, browsers
+  and the rig were all up; CPU avg only ~13%, so the heat is the cap-and-cycle, not load.
+- **~13:42 "no sound"** (even Pulsar sample-browser clicks and desktop sounds, HDMI into
+  speakers): the PipeWire graph clock stalled right after a USB touch on the KeyStep 32
+  (kernel: usb 1-3.3 disconnect/re-enumerate 13:41:51). SuperDirt stayed up; a 1.5 s test
+  file never finished playing. `systemctl --user restart wireplumber` alone cleared it
+  (health.json back to ok, tone played). HDMI 1 volume came back at 70% (was 94%).
+
+## 2026-11-01 — Arturia KeyStep 32: the G key drops presses (hardware, not software)
+- **"the middle G ... does nothing"**, intermittent, about 1 in 3 registers. Read the raw ALSA
+  sequencer port (`aseqdump -p 20:0`, before SuperCollider/Tidal) over four captures, then a
+  controlled run of 10 presses per note, C3 up to B3 and back.
+- Every neighbour key arrived 10/10 with normal velocities (74-117) and clean note-on/off pairs.
+  **G3 (note 55): first ascent F#3 -> G#3 skipped it entirely (0/10); the way back down gave
+  4/10 (PLN heard 2-3/10).** Every G3 that did arrive was velocity 120-127; an earlier capture
+  also gave one G3 at velocity 1. The same capture later holds 22 more G3 note-ons (runs of 10
+  and 12) after PLN started hard-pressing the key: all velocity 120-127. (Correction: this
+  entry first said "5 in ~30", read while the capture was still recording.)
+- Reading: a flaky contact under that key (only the hardest hits trigger both contacts), at the
+  source: USB stayed up the whole time, no stuck or doubled notes. Not confirmed by opening it.
+- Next: hard-press the key 30-40 times, then dismount the key and clean the contact if it sticks.
+  Gig workaround: play the note an octave away or transpose in the patch.