# YazSes v2.1.0-dev.4 — Wave E ship-now tier (developer preview)

**Pre-release · not published** (publish workflows are pinned to `v0.x`/`v1.x`; v2 dev tags
do not ship). **v1.4.1 remains the stable release.** Every feature is **OFF by default**.

Opens **Wave E** — a fresh 2026 SoA round (10 candidate features distinct from the prior 24)
in `design/vision/v2-research/07-wave-e.md` + `design/adr/adr-v2-025..028`. The first four,
all on-device and pure-cored:

## New features (all off by default)

- **Hallucination Guard** (ADR-v2-025, `[hallucination]`) — drops Whisper's fabricated text on
  silence/noise (the phantom "Thank you.", "please subscribe", looped phrases) before it's
  typed. Whole-transcript-only ghost blacklist + repetition-loop detector + a segment-signal
  gate. Wired into the decode path; conservative.
- **Voice Snippets** (ADR-v2-026, `[snippets]`) — say a trigger ("insert my signature") to type
  a stored template. Distinct from `[macros]` (keystrokes).
- **Phonetic Corrector** (ADR-v2-027, `[phonetic]`) — fixes mis-heard proper nouns by *sound*
  against your vocabulary ("Cuber Netties" → "Kubernetes"), skipping known/exact words.
- **Multi-User Profiles** (ADR-v2-028, `[voiceprint] multi_profile`) — on a shared machine,
  loads each enrolled speaker's own vocab/hotkey/cleanup from their voiceprint. Distinct from
  Voice Guard (N-way routing vs binary gate); reuses the encrypted d-vector.

`yazses features` now lists 39 capabilities (new: `hallucination`, `snippets`, `phonetic`,
`multiprofile`).

## Wave E backlog (designed next / hardware-gated)
Spoken Math→LaTeX, Semantic Endpointing, Spoken Code Mode, Wake-Word Activation, Vocal-Strain
Guard, Voice Mouse Grid — several compose into a hands-free "zero-touch" accessibility bundle.

## Quality
994 tests green across Linux × macOS × Windows on Python 3.11 and 3.12. Every new feature's
pure core is unit-tested to 100%; heavy backends stay lazy behind optional extras, so the
default install and the v1 dictation path are unchanged.