shin ש-vs-ס: low-band decisive-gap guard (block noise that squeaked past floor)

stt_judge: straggler grace window — keep 3 engines, don't block on a hung one

main: background self-warm thread keeps the model hot 24/7

stt_judge: replace Azure with Google Gemini (handles short clips, phonetic prompt)

stt_judge: recognise Latin transliteration (Shh/shhh -> ש, s -> ס)

shin ש-vs-ס: lower presence floor 0.02->0.003 (fix isolated-sound false negatives)

stt_judge: replace Deepgram with Azure Speech (good on short clips)

Hybrid routing: STT for sound/syllable, math for words

stt_judge: require >=2 agreeing definitive votes (no single-model decisions)

stt_judge: force literal Deepgram transcription (purity)

Add STT 'pocket' for MVP ש/ס & ק/ת detection (toggle, math preserved)

ש scoring: replace brittle s_band heuristic with model-driven ש-vs-ס

k_sound: raise burst-centroid threshold 2500->3400 for child voices

Fix sh_sound crash (dict.score) + lenient s_band for isolated stage-1 ש

evaluate: never leak Python exceptions to the client; log traceback instead

ק competitive_align: compare the differing consonant, not the whole-word mean

data_store: stamp model_version into every evaluation record

wetness: default to WavLM-large layer 4 (best separator); add layer-sweep tool

wetness_detector: store PCA as raw numpy for cross-version pickle safety

Add WavLM wetness/lateral-lisp detector — ש-syllable pipeline only

SH syllable: log temporal/envelope features for wet-detection probe

SH syllable: add CTC-based CH-substitution gate

SH syllable: extend extraction window from 120ms to 200ms

SH syllable: log full feature set for wetness calibration

K syllable: top-choice CTC with clinical-pedagogy accept rules

K syllables: switch to multi-way CTC competitive alignment

K syllable: remove broken decay gate (was rejecting all correct K)

K-syllable pipeline: 4-gate acoustic validation (per-vowel + decay + voicing)

Generalize ק incorrect feedback — no longer assumes ת substitution

ק competitive_align: require decisive gap >= 0.10 to accept CORRECT

Add silence / no-signal gate to all scoring pipelines

warmup: eager Supabase init + dummy score pass + GET / and POST /warmup endpoints

Drop sub3k gate from ש pipeline; raise s_band threshold to 0.45

Add CV-syllable scoring pipelines (sh_syllable_*, k_syllable_*)

Add 8 CV syllables: 4 shin (sha/she/shi/shu) + 4 kuf (ka/ke/ki/ku)

Add 3 K words: akavish, sakin, mashrokit_kuf

Add mishkafayim_kuf — second classification of משקפיים for ק testing

Add 3 kaf words: kova, kadur, kelev

Revert audio static mount — narration not needed right now

Bump cache-bust to force HF Space rebuild for audio mount

Mount audio_data/ at /api/v1/audio for Unity narration playback

Add 9 shin words: mashrokit, karish, marshmelo, machshev, dvash, shtayim, nachash, sheva, mivreshet_shinayim

Add shin word: hipushit (חיפושית, ladybug)

Add shin word: yanshuf (ינשוף, owl)

Add shin word: deshe (דשא, grass)

Add shin word: galshan (גלשן, surfboard)

Unify shin scoring: apply 4-gate system to all ש words

Add 5 new shin words: shin, shofar, mishkafayim, shatiach, sharsheret

Add s_band + sub3k gates for sh_sound (15/15 calibration accuracy)

Log full feature set for sh_sound calibration