Skip to content

docs: WO33 word channel — lexical floor measured, disambiguator arm launched - #292

Merged
ronaldtse merged 72 commits into
mainfrom
docs/wo33-word-channel
Nov 3, 2026
Merged

ronaldtse merged 72 commits into
mainfrom
docs/wo33-word-channel

Conversation

@ronaldtse

Copy link
Copy Markdown
Contributor

Summary

  • WO33 spec (TODO.sota/33-lexicon-disambiguator.md): lexical-channel decomposition — pure most-frequent-reading lookup = 25.15 WER / 13.18 DER, OOV 18.10% (482,609 types); run-035 variant-disambiguation tagger (word-emb + char-CNN + BiLSTM over variant indices) launched with gates 14.30 (beats char BiLSTM) / 10.13 (news-successor candidate)
  • run-030-heb-noisy HF rerun verdict: DER 8.72 vs crown 8.18 — WO26 negative re-confirmed; Hebrew text levers stay closed
  • RESULTS.md: WO32 lineage + WO33 launch recorded; 00-README index row 33 added

Test plan

  • docs-only change; numbers match bucket verdicts (run-030-heb-noisy/verdict.json, run-034-bilstm/wikinews_multiref.json)

Ronald Tse added 30 commits October 10, 2026 06:48
Ronald Tse added 29 commits October 11, 2026 18:01
…erified); 1.0 card corrected-protocol metrics + decode-contract note (WO44 D5); draft cards
…ilter/gold-nikud/static-int8 arms; matrix post-WO38/40
@ronaldtse
ronaldtse merged commit 816c728 into main Nov 3, 2026
7 of 11 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant