Prose has no scope (promoted)

Status (2026-07-17): promoted. The theory now lives in vocabulary collisions are prevented at write time, not resolved at read time — the corpus-scale no-scope argument, the silent-composition failure, the prevention-over-scoping remedy, the position-in-a-schema analysis, the multi-word-coinage naming rule, and the load-bearing-polysemy caveat all live there. This file keeps only what is workshop-specific: evidence pointers the note cannot link, decisions the thread produced, and the open policy work this workshop still owns.

  • distillation-usage-audit.md — the 464-instance sweep: scope leakage observed at corpus scale, and the cost benchmark for checking a captured common word (semantic audit) vs a coined compound (lexical search)
  • the distillation control trap (thread closed 2026-07-17; its control-regime diagnostic is now the classifying question in kb/reference/link-vocabulary.md) — the operational cost of one misresolved sense: an artifact borrowing a maintenance regime its control story doesn't support
  • migration-plan.md — the thread's conclusions enacted: minimality constraint, no-successor-term end state, and the naming decisions below

Decisions this thread produced

  • One term, one sense, corpus-wide — the uniqueness invariant, enforced at write time; adopted as the reading of "collision handling" for this workshop's candidate model (imports, qualified names, and shadowing have no prose implementation).
  • Multi-word coinages as the default naming rule for technical terms (operator, 2026-07-17).
  • discovery lifecycle adopted as the migration's incoming technical term — the compound, not bare "discovery" — applying the naming rule to the migration's own Wave 0.

Flagged tangent (parked for the policy pass)

"prose" is itself a captured common word in KB-technical use — a representational-form category, already on the outward-docs exclusion list. Same exposure class as distillation, but its technical sense sits close to the ordinary one, so misresolution may be low-consequence. Suggests the policy pass wants a graded exposure test — how far apart are the senses, and what rides on resolving them? — rather than a binary common-word rule.

Open policy work (this workshop)

  • Where to encode the naming rule: the vocabulary-policy output of this workshop, cp-skill-write's naming guidance, or both.
  • Write-time enforcement of the uniqueness invariant: the candidate mechanisms (reserved-term registry, slot-escape lint, coinage collision screen, naming-review gate, clausal-binding link check) are now collected in write-time vocabulary collision controls; this workshop keeps the policy question of where the controls read their term list from.
  • Whether collection-local vocabulary is compatible with the invariant at all (safe only for terms that cannot co-load with a colliding sense, or are coined so collision is impossible).