The v2 research waves¶
The v2 feature line (adr-v2-000 through the 120s, design/adr/) was not designed from a single backlog — it came out of a repeating cycle: a state-of-the-art literature/product sweep, then a ranked list of candidate features, then ADRs for the ones worth building. This directory is that research trail, published so the ADRs have a visible "why" behind them instead of appearing pre-formed.
Every file here is a snapshot of the field on the date it was written, not a live roadmap. Where a candidate feature is discussed and never shipped, that is a research note, not a promise — run yazses features and check the linked ADRs for what is actually built today. Citations have not been re-run through research/verify_refs.py since the original sweep; treat arXiv IDs and figures as pointers to chase, not confirmed facts.
| File | Fed | Domain |
|---|---|---|
| 01-voice-hci.md | adr-v2-001 – 009 | Voice/speech HCI & dictation SoA |
| 02-agentic-os.md | adr-v2-004 – 010 | On-device LLM agents & agentic OS |
| 03-ar-wearables.md | adr-v2-010, 011, 013 | AR/VR/wearable input |
| 04-accessibility.md | adr-v2-009, 012, adr-015 | Accessibility & assistive input |
| 05-ambient.md | adr-v2-004, 005 | Ambient/context-aware computing & memory |
| 06-wave-d.md | adr-v2-014 – 024 | Wave D: translation, denoise, meeting scribe, RAG, silent speech |
| 07-wave-e.md | adr-v2-025 – 034 | Wave E: hallucination guard, math/code modes, wake-word |
| 08-wave-f.md | adr-v2-035 – 044 | Wave F: speaking coach, smart-paste, read-back voice |
| 09-wave-g.md | adr-v2-045 – 054 | Wave G: ITN, redaction, field-aware dictation, sign language |
| 10-wave-h.md | adr-v2-055 – 064 | Wave H: spreadsheet mode, temporal/unit conversion, self-repair |
| 11-wave-i.md | adr-v2-065 – 074 | Wave I: terminal safety gate, contextual biasing, diarized capture |
| 12-wave-j.md | adr-v2-075 – 084 | Wave J: confidence-gated re-ask, git choreographer, recording import |
| 13-wave-k.md | adr-v2-085 – 094 | Wave K: chorded shortcuts, undo/redo timeline, focus-class profiles |
| 14-wave-l.md | adr-v2-095 – 104 | Wave L: non-speech & prosodic interaction (Vocal Joystick, earcons) |
| 15-wave-m.md | adr-v2-105 – 114 | Wave M: minimal-bandwidth AAC & text intelligence (Vocal Morse) |
| 16-wave-n.md | adr-v2-115 – 124 | Wave N: structural editing, i18n, accessibility-output correctness |
| 17-diarized-recording-import.md | adr-v2-125, 126 | Wave O: diarized recording import (yazses transcribe) |