fix(heuristics): drop empty-title (date-only) entry blocks before scoring (#145)

[rohithgollapalli]

Summary

A header/anchor line whose entire text is a date leaves nothing after stripDateRange, so parseEntryBlocks emits a block with empty headerLines. All three extractors mapped that straight through to a { title: "" } / { name: "" } entry with score 0 — a title-less item in output that also diluted section confidence via avgScore. (Surfaced during #144 review; pre-existing and shared across the extractors.)

Adds a shared finalizeEntries(built, hasLabel) helper in extract/shared.ts that drops empty-label entries and packages the survivors as the standard { value, confidence } pair — keeping the score-0 phantom out of the avgScore denominator. Routes extractAchievements / extractProjects / extractExperience through it (predicates: title, name, and title || company for experience). Filtering on the built entry's label rather than on empty headerLines also catches a URL-only header, which liftHeaderLabel collapses to an empty label — so every anchor path (first_line flat-list, first_line buildEntryBlock, date_range) is covered at once. Per-block mappers and scoring are unchanged.

Closes #145

Test plan

• npm run typecheck clean
• npm run test green (723 passed; +5 new extract-fields cases: date-only bullet, date-only header, date-only project, date-only experience, and a no-dilution regression guard)
• Corpus re-baked — no fixture content changed vs main (confirmed via --ignore-cr-at-eol); no real résumé carried a phantom empty entry, matching the issue's expectation
• eslint clean on changed files

[s-annam]

PR Review: fix(heuristics): drop empty-title (date-only) entry blocks before scoring (#145)

Summary

Correct root-cause fix. A date-only anchor reduces to an empty label after stripDateRange, producing a phantom { title: "" } / { name: "" } score-0 entry that also diluted section confidence via avgScore. The new shared finalizeEntries helper drops the phantom and packages survivors as the standard { value, confidence } pair.

Highlights

• DRY: one helper replaces the duplicated blocks.length === 0 → avgScore(...) tail across all three entry extractors.
• Filtering on the built entry's label (not on empty headerLines) also catches URL-only headers that liftHeaderLabel collapses to an empty label — covers every anchor path at once.
• Behavior-preserving for the empty case (empty built → avgScore([]) → 0), so no regression vs the old early return.
• Predicates match the returned label fields (title, name, title || company for experience).

Verification

• npm run typecheck clean
• npm run test — 723 passed (+5 new cases incl. no-dilution regression guard)
• Corpus unchanged vs main

Verdict

Action: APPROVE
Rationale: No findings. Targeted, well-tested fix.

[s-annam]

Re-approve after rebase onto main + conflict resolution in experience.ts (import-line only: kept #159's US_LOCATION_RE/INTL_LOCATION_RE, swapped avgScore→finalizeEntries; avgScore now unused in file). Gates green post-rebase: typecheck clean, 732 tests passing.

[s-annam]

Re-approve after updating branch to base (160 merged; disjoint files, clean update — no conflict). Gates green on prior rebase: typecheck clean, 732 tests.