Wave I — SoA research, 10 net-new features¶
Date: 2026 · Tier: design/ — public engineering research · Author: Mohsen Seyedkazemi Ardebili Companion: fed adr-v2-065 through adr-v2-074. See the waves index.
A snapshot of the field, kept as the research record behind the ADRs it fed. Feature status should be checked against
yazses featuresand the linked ADRs, not this note.
All on-device, off by default, distinct from the ~62 features existing before this wave (v2 + Waves D-H). Ranked strongest-first. Anchors web-verified at the time (2025-2026); items #5-#10 rest on engineering substrates (AT-SPI2, WM IPC, CSL) rather than a single paper — flagged honestly in the original sweep.
- Hard Contextual Biasing — decode-time trie/WFST hotword boosting so rare names/jargon actually win (vs soft
initial_prompt). Anchor: WCTC-Biasing (2506.01263), trie K-step (2509.09196), sherpa-onnx hotwords. Pure:build_hotword_trie+bias_logits/rescorer. Distinct from Context-Primed (soft prompt) + Personal Adapter (LoRA). → adr-v2-069. - Diarized Conversation Capture + Rename-by-Voice — attributed per-turn Markdown, rename "speaker two" → "Alice". Anchor: pyannote community-1 / pyannote.audio 4.0. Pure:
SpeakerLabelMap+apply_rename+render_attributed_markdown. Distinct from Meeting Scribe (bulk) + Multi-User Profiles (auth). → adr-v2-074. - Adaptive Latency Governor — load-aware decode policy + speculative decoding (a distil draft model matching identical output ~2× faster). Anchor: HF spec-decode, distil-large-v3.5 (2311.00430). Pure:
pick_policy(load, config). Distinct from Ghost-Ahead (pre-warm only). → adr-v2-073. - Per-Language Auto Model Switching — LID first ~1-2s → hot-swap a language-specialized model. Anchor: Whisper LID + per-language distil/LoRA. Pure:
route_language(lid, registry). Distinct from Code-Switch (within-utterance) + Polyglot (fixed pair). → adr-v2-072. - Voice-Driven Document-Wide Find-and-Replace — "replace every utilise with use". Anchor: AT-SPI2 EditableText. Pure:
parse_replace_command+to_editor_actions. Distinct from Spoken Edit (last utterance only). → adr-v2-068. - Terminal Command Safety Gate — hold
rm -rf/curl|shuntil spoken confirm when focus is a terminal. Pure:classify_command_riskruleset + confirm state machine. Inspects command semantics before injection. → adr-v2-065. - Structured-Markup Dictation — speak Markdown/org tables and lists. Pure:
render_markup+ structure grammar. Distinct from Spreadsheet (drives an app) + Code/Math. → adr-v2-067. - Voice Window/Workspace Management — "move window left half", "workspace 3". Anchor: wmctrl/hyprctl/swaymsg. Pure:
parse_wm_command+ backend adapters. Distinct from Mouse Grid/Pilot. → adr-v2-070. - Spoken Regex Builder — "four digits dash two digits" →
\d{4}-\d{2}. Pure:nl_to_regexcompositional grammar. Builds search patterns, not literal code. → adr-v2-066. - Citation-by-Voice from local BibTeX/CSL — "cite Vaswani 2017" → formatted citation, offline. Pure:
resolve_citationfuzzy match +format_citation. Distinct from RAG (retrieve+generate). → adr-v2-071.
Ship-now pure (do first)¶
7 Structured-Markup, #6 Terminal Command Safety, #5 Find-and-Replace, #9 Spoken Regex —¶
cleanest fully dependency-free cores (pure text functions), each an accessibility/safety win a cloud tool can't offer.
Citations here have not been re-verified against research/verify_refs.py.