diff --git a/docs/AUDIO_FIDELITY.md b/docs/AUDIO_FIDELITY.md
index fd953ac..8b44d0f 100644
--- a/docs/AUDIO_FIDELITY.md
+++ b/docs/AUDIO_FIDELITY.md
@@ -428,6 +428,14 @@ in the mech reset (part_012.c:9446-9447) — a genuine regression, not dead cont
 inbound lock/launch on the player exists; DistanceToMissile = range to nearest inbound missile
 (far default when none). The authored watchers then reproduce the accelerating alarm unchanged.
 
+**ERA-WITNESS CONFIRMED (2026-08-13, Oracle, via the distributable soundboard):** Oracle found
+the sample himself on the board — `[1:74] IncomingAlarm01 n58` — and confirms it is the
+incoming-missile alarm, adding "**it was sped up in the pods**": independent field confirmation
+of both the sample identity AND the authored DistanceToMissile→tempo acceleration (the 100..800
+→ tempo 600..10 mapping above). Gitea #83 (comments 1781/1783). The reconstruction target is now
+fully specified except the binary's exact IncomingLock/DistanceToMissile update rule (what sets
++0x3FC/+0x400 — seeker-locked-only vs any-inbound, and nearest-missile selection).
+
 <a name="f5"></a>
 #### F5. FootStep is a contact LEVEL, not a clip-transition pulse (High, CONFIRMED)