Shared lexicon (a): one bullet-glyph vocabulary — ⁃ and — are parser bullets the scorer silently drops

[s-annam]

Part (a) of #653. Not a good first issue, and do not batch it — this one changes scoring behaviour and needs a reviewed, field-by-field rebaseline.

Decisions (2026-09-29)

• One shared glyph vocabulary, one import-nothing leaf. Parser and scorer both derive their bullet classes from it. No second hand-written copy anywhere.
• The scorer adopts the parser's glyphs. ⁃ (U+2043) and — (em dash) become scorer bullets. The scorer keeps the glyphs it already recognises that the parser lacks (▶ ► · �) — dropping them would move scores the other way for no reason; the implementer verifies the actual union from the current sources (see the table) and records each glyph's adjudication in the leaf docblock.
• Leading hyphen - and en-dash – are the ambiguous pair (date-range fragment / negated number, see cleanRewriteLine eats the minus sign off a line-initial negative number (-5% churn → 5% churn) #821). They are expressed as an explicit, documented named subset in the leaf (e.g. AMBIGUOUS_DASH_BULLETS), not a second copy of the list, with the cleanRewriteLine eats the minus sign off a line-initial negative number (-5% churn → 5% churn) #821 reason in the docblock. Maintainer intent: these two stay parser-only — i.e. the scorer must not gain new hyphen/en-dash handling from this issue. NOTE: the scorer's classes today (BULLET_MARKER_RE, apply-overrides.ts LEADING_MARKER_RE) already include - and –. The implementer must not silently drop them from the scorer (that would deflate Specificity on every hyphen-bulleted résumé); if the maintainer's "parser-only" is read literally, stop and ask before removing them. Default if unanswered: the scorer's current -/– handling is unchanged, and the named subset documents that they are the only glyphs the parser and scorer treat by separate rules.
• Score change and a corpus/score rebaseline are allowed. The implementer must justify every moved baseline field by field (fixture, which glyph, before → after Specificity/bullet count) in the PR description. Bump ATS_SCORE_ALGO_VERSION (currently "1.8", src/lib/score/score.ts) and add the version's entry to its docblock changelog, following the 1.x entries (which glyph set changed, that WHICH bullets are pooled moves, so Specificity/Structure move). The bump also invalidates the Local-first storage foundation: IndexedDB module for resumes and jobs #321 resume-library cache, which is intended.

Problem

The bullet-glyph vocabulary exists as divergent copies, and the divergence is live: glyphs the parser treats as bullets are invisible to the scorer, so those bullets drop out of the Specificity pool.

Verified divergence (line numbers re-checked 2026-09-29; they drift, so anchor on the symbol):

site	class
src/lib/heuristics/line-primitives.ts:41 (BULLET_CLASS)	[•‣▪●◦⁃*\-–—]
src/lib/heuristics/sections.ts:152 (VISUAL_BULLET_RE)	[•‣▪●◦⁃*\-–—]
src/lib/heuristics/regex.ts:383 (LEADING_BULLET_RE)	[•‣▪●◦⁃*\-–—]
src/lib/score/score.ts:137 (BULLET_MARKER_RE)	[-*•●–▪◦‣▶►·�]
src/lib/score/score.ts:831 (LONE_BULLET_RE)	[•●▪◦‣▶►·�]
src/lib/edit/apply-overrides.ts:176 (LEADING_MARKER_RE)	[-*•●–▪◦‣▶►·�]
src/lib/score/group-bullets.ts:35 (LEADING_MARKER_RE)	same as the scorer's, plus numbered markers
src/lib/heuristics/markdown-emit.ts:66 (LEADING_BULLET_RE)	[•▪◦‣·⁃∙●⬤▸▶*\-]

Also near-copies to reconcile or consciously leave (say which in the PR): src/lib/heuristics/extract/summary.ts:18 (SUMMARY_BULLET_RE), src/lib/heuristics/localize/achievements.ts (mirrors group-bullets' marker set), and the comment-only mirror in extract/skills.ts:24. (The original issue listed extract/group-bullets.ts and webllm/localize/achievements.ts; those paths are now score/group-bullets.ts and heuristics/localize/achievements.ts.)

⁃ (U+2043 hyphen bullet) and — (em dash) are parser bullets absent from the scorer's class. A résumé using either has its Experience bullets segmented correctly by the parser and then silently skipped by extractBulletsFromLines (score.ts ~:851), deflating Specificity. Conversely ▶, ►, · and � are scorer-only.

Why this is separate from #653(b)/(c)/(d)

Merging the sets changes scores on existing résumés. That is a deliberate behaviour change requiring a version bump and a reviewed rebaseline, which is why it cannot ride along with the behaviour-preserving consolidations.

Implementation plan

1. Mint a new import-nothing leaf for the glyph vocabulary. House precedent is extract/title-shape.ts. Not line-primitives.ts — it imports regex.ts, and deriving there re-creates the cycle regex.ts (LEADING_BULLET_RE's comment) exists to avoid, dragging the large regex.ts and its module-eval Intl gazetteer onto the eager entry graph via the score/edit consumers (the feat: primary role targeting, editable headline, and job-search handoff (#598) #605 failure class).
2. Adjudicate each glyph in the union: parser-bullet / scorer-bullet / both, with the reason written in the leaf docblock. Per the Decisions above: the union is shared, -/– are the documented ambiguous subset.
3. Derive all sites above from the leaf as named exported sets, preserving any distinction that survives adjudication (the ambiguous-dash subset; any scorer-only glyph) as its own derived constant rather than collapsing it.
4. Bump ATS_SCORE_ALGO_VERSION and add its changelog entry.
5. Rebaseline the corpus snapshots (npm run bake-fixtures); review the score deltas fixture by fixture and explain each in the PR description.

Acceptance criteria

• One import-nothing leaf owns the glyph vocabulary; every site in the table derives from it (no remaining hand-written glyph class), and any near-copy left alone is named with a reason in the PR.
• The leaf imports nothing, asserted by a test (there is no existing test for title-shape.ts's contract — write one; scripts/check-eager-graph.mjs is the eager-graph counterpart).
• Per-glyph adjudication, including the -/– named subset and the cleanRewriteLine eats the minus sign off a line-initial negative number (-5% churn → 5% churn) #821 reason, is written in the leaf's docblock, not just in the PR.
• The scorer treats ⁃ and — as bullets; the scorer does not lose any glyph it recognises today.
• ATS_SCORE_ALGO_VERSION bumped with a docblock changelog entry.
• Corpus rebaseline reviewed fixture-by-fixture; every moved baseline justified field by field in the PR description.
• A regression test covers a ⁃-bulleted and an —-bulleted résumé reaching the Specificity pool.
• The / entry chunk does not gain regex.ts (build-verified; npm run check:eager passes).

Parent: #653. Roadmap item 5(a) of #646.

[s-annam]

Clarification (rewrite) — Body rewritten with a ## Decisions (2026-09-29) section: one shared import-nothing glyph leaf, the scorer adopts the parser's glyphs (⁃, —), -/– are an explicit documented named subset (#821), the rebaseline is allowed with per-field justification and an ATS_SCORE_ALGO_VERSION bump; stale file/line refs refreshed.