diff --git a/docs/manual-qa-test-plan.md b/docs/manual-qa-test-plan.md
index 59fa4552..f363e87a 100644
--- a/docs/manual-qa-test-plan.md
+++ b/docs/manual-qa-test-plan.md
@@ -349,7 +349,7 @@ Additional cases from a code-grounded gap audit of this plan (8 read-only audits
 - **TC-DICT-08** Stop then immediately restart with old responses in flight → stale-session transcripts must NOT leak into the new session; in-session order preserved despite out-of-order HTTP (`use-streaming-dictation.ts:148-227`).
 - **TC-DICT-09** Multi-segment insertion: move the caret / delete content above mid-session → subsequent inserts stay contiguous and clamp into doc bounds, no crash on editor destroy (`dictation-group.tsx:13-68`).
 - **TC-DICT-11** Three gate states: dictation toggle off (mic hidden, 403 if forced) / STT model blank (503 "Voice dictation is not configured") / endpoint 401/404 (verbatim server message) (`ai-chat.controller.ts:220-269`).
-- **TC-DICT-12** [unit] `encodeWavPcm16` correctness — every streaming VAD segment is encoded here and POSTed as `audio/wav`, so a header/clipping bug silently degrades transcription across the whole feature with no visible error. Assert: the RIFF/WAVE/fmt/data chunk tags, channels=1, bitsPerSample=16, `byteRate = sampleRate*2`, `dataSize = samples.length*2`; sample-rate override (16000 default vs 8000/48000) written at offset 24; clipping `+1.0 → 32767`, `-1.0 → -32768`, out-of-range `±1.5` clamps to the rails with no wraparound at exactly −1.0; blob length `44 + samples.length*2` (`apps/client/src/features/dictation/utils/encode-wav.ts:3-32`; full spec in `docs/backlog/qa-plan-unit-test-candidates.md`).
+- **TC-DICT-12** [unit] `encodeWavPcm16` correctness — every streaming VAD segment is encoded here and POSTed as `audio/wav`, so a header/clipping bug silently degrades transcription across the whole feature with no visible error. Assert: the RIFF/WAVE/fmt/data chunk tags, channels=1, bitsPerSample=16, `byteRate = sampleRate*2`, `dataSize = samples.length*2`; sample-rate override (16000 default vs 8000/48000) written at offset 24; clipping `+1.0 → 32767`, `-1.0 → -32768`, out-of-range `±1.5` clamps to the rails with no wraparound at exactly −1.0; blob length `44 + samples.length*2` (`apps/client/src/features/dictation/utils/encode-wav.ts:3-32`; full unit-test spec tracked in Gitea issue #139).
 
 ### V.3 Editor & collaboration