diff --git a/context/gauges-hud.md b/context/gauges-hud.md
index 3ecb848..aaba082 100644
--- a/context/gauges-hud.md
+++ b/context/gauges-hud.md
@@ -699,6 +699,47 @@ a batch of gauges per gauge turn (a visit is only a rate-mask check unless due).
 within ~1s, bg cost 2-4 ms/frame.  Diagnostics: `BT_PERF=1` → `[perf] … bgTasks/gaugeTurns/
 sweeps/active` 1 Hz (APPMGR.cpp); `[score] panel DRAW` edge log (btl4gau3.cpp, BT_SCORE_LOG).
 
+## The COOLANT-LEAK ALARM audio chain — fully built; the shared-Stop is AUTHENTIC (#99, 2026-08-01) [T1 disasm-verified]
+
+The audible leak warning is **implemented and works end to end**; an earlier note
+claiming it was "genuinely unbuilt" was wrong (asserted without checking).
+
+**The chain**, traced live (all probes under `BT_ATTRBIND_LOG`):
+`HeatSink::coolantActive` (published as the `ReportLeak` attribute) changes 0->1
+-> `AttributeWatcherOf<Logical>::Execute` sees it (`[watchpoll] CHANGE`)
+-> `AudioMatchOf<Logical>::SendNotificationOfChange` fires (`[matchfire] val=1 -> ctl 1`)
+-> `AudioControlSequence::ReceiveControl(StartAudioControlID)` -> `StartSequence`
+-> an `AudioIdleWatcher` ticks it each frame (`IdleAudioControlID` -> `RunSequence`)
+-> `[seqsend]` emits the authored pattern: two chirps then an ~8s sustained tone, looped.
+
+**The authored wiring (BTL4.RES): 19 subsystems, ONE shared alarm sequence.**
+`Condenser1-6, GeneratorA-D, Myomers, PPC_1/2, ERMLaser_1-3, SRM6_1/2, Avionics`
+each bind TWO `AudioLogicalTrigger`s to their own `ReportLeak` — match 1 -> ctl 1
+(Start), match 0 -> ctl 2 (Stop) — and every one of them drives the SAME sequence.
+
+**Consequence, and it is AUTHENTIC:** any single subsystem clearing its leak sends
+an unconditional Stop, silencing the alarm even while others still leak; the
+still-leaking ones never re-assert because their flag has not *changed*.  Reported
+from the field as a bug ("was sounding, resolved a leak, but there was another").
+**Do NOT 'fix' it** — verified in the arcade image, not merely inferred from the
+engine source lineage:
+
+| what | where |
+|---|---|
+| same MUNGA sources compiled in | `d:	esla_bt\munga\AUDSEQ.CPP` / `AUDWTHR.CPP` strings present |
+| object factory | `@00466184` `case 0x4b` -> alloc **0x5C** -> ctor `@0043ccec` |
+| vtable | `0x4ead48` |
+| `ReceiveControl` | `@0043d3a4` — switch on control IDs **1 / 2 / 9 / 12** = Start / Stop / Idle / Tempo (exactly our enum) |
+| Stop case | `@0043d3ca` -> unconditional `call StopSequence @0043d2d4` |
+| `StopSequence` | tests only its OWN `isRunning` (`+0x28`), then the Chase loop — **no refcount, no awareness of other leak sources** |
+
+So the arcade had the identical structure.  Silence in 4.11.674 was a different
+cause: the OpenAL source pool was exhausted in every player session (see the audio
+pooling fix) — an alarm that cannot acquire a source is silent.
+
+⚠ The bench cannot confirm audibility: it runs with no audio device (`live=0
+pooled=0`), so the control chain is verified but final playback is not.
+
 ## Key Relationships
 - Full history: `docs/GAUGE_COMPOSITE.md`; reticle recovery: `phases/phase-02-dpl2d-reticle.md`.
 - Uses: [[attribute-pointer]] + [[reconstruction-gotchas]]; reads [[subsystems]] state.