Logos52
journal / 2026 06 13 tsumugu tocfl a to c complete

Tsumugu: TOCFL A–C complete — every character and word, A1 through C1

journal updated 2026-06-13

Tsumugu: TOCFL A–C complete (2026-06-13, capstone)

The coverage campaign closed today. The Tsumugu Encoding Dictionary now holds an entry for every TOCFL character and every TOCFL word, A1 through C1.

The numbers

  • Words: 6,006 / 6,006 (100%).
  • Characters: 100%, A1–C1.
  • Corpus: 9,663 entries, validation green — ids unique, filenames canonical.
  • C2 is not in the 華語八千詞 list (it tops out at C1); true 精通級 coverage would need an external word list — a separate, future campaign.

How the last mile closed

The day’s arc: the seed-gate unblock let authoring rip → the character set closed (A1–B2 100%, then the 55 C1 tail) → the bottleneck flipped to words, where the lane had been draining a frequency corpus rather than the TOCFL list → a TOCFL-seeded, band-ordered word campaign ran in one parallel wave (4 author-only agents, ~4,047 entries, zero collisions) to 99.2% → the 55 C1 characters authored, unblocking the final 46 C1 words.

Those final 46 were authored directly (the expand.py compact-text path is host-pathed and won’t run in the Cowork sandbox, so the JSON was built straight against the schema, zhuyin via gen_helpers, filenames via validate.py’s own slug function). Tree validated green at 9,663; register scan clean. Highlights: 帳篷 (cloth over a bamboo frame), 主宰 (宀 + 辛, the knife-wielder in the house, sourced), 嘴脣 with its 脣/唇 scholarNote.

Close-out bookkeeping

  • Reconciler ran (reconcile_queues.py --apply): 3,991 drifted word statuses synced queuedauthored. The 579 still queued are the non-TOCFL frequency-corpus tail — not gaps. Backup written.
  • Commit is host-side only (.git mount-blocked in the sandbox) — the 46 entries + reconciled queues are authored and valid on disk, committed via a host-side git add -A / author-loop run.

What remains (not blocking the milestone)

  • The 289-entry single-source audit (Composer cross-check of words/chars authored from a single Grok crib) — still parked; quality insurance, not coverage.
  • Scholar-note side-placement: 68 orthographic-twin pairs carry the note on one side only — cosmetic, undecided (mirror, standardize onto the orthodox form, or leave).
  • Beyond TOCFL: the 579-word frequency tail if the corpus should grow past the exam list; and the recurring researched→queued gate worth fixing at crib_diff so future crib runs auto-promote.

The dictionary’s exam-coverage spine is done. Remaining work is refinement and reach, not gaps.