Cascade inconsistency: bullets counted but experienceCount: 0 for non-standard experience headings (laverne fixture)

[Vaishnavi1709]

Problem

The openresume-laverne-word-quartz corpus fixture exposes a scoring inconsistency for student and volunteer-heavy resumes:

cascade.experienceCount   → 0       (parsed.experience[].length)
score.bulletCount         → 6       (extractBulletsFromText over rawText)
score.completeness.missing → ["work experience", "phone", "LinkedIn"]

The resume contains sections headed "VOLUNTEER EXPERIENCE" and "ON CAMPUS INVOLVEMENT" with 6 experience-shaped bullet points. The parser scores those 6 bullets for Specificity and Structure (document-scoped), but experienceCount stays 0 and Completeness deducts a full check for missing "work experience" — because experience extraction is section-scoped on SECTION_KEYWORDS.experience (regex.ts:96–105) while bullet counting runs document-scoped in extractBulletsFromText (score.ts:508).

Current snapshot: tests/fixtures/pdfs/word/openresume-laverne-word-quartz.expected.json

"experienceCount": 0,
"bulletCount": 6,
"completeness": {
  "missing": ["LinkedIn", "phone", "work experience"]
}

Root cause

Two layers in the pipeline treat "experience" differently:

1. Tier 1 parser (openresume.ts:107): findSection(sections, "experience") returns null because matchSectionHeader (regex.ts:172) tests each heading against SECTION_KEYWORDS.experience (regex.ts:96–105). That list contains only canonical professional-experience synonyms ("experience", "work experience", "professional experience", "employment", "employment history", "work history", "career", "career history"). Headings like "VOLUNTEER EXPERIENCE" and "ON CAMPUS INVOLVEMENT" match nothing, so extractExperience(null) returns [] and parsed.experience stays empty.

2. Anonymous scorer (score.ts:530): extractBulletsFromText(input.rawText) scans the raw text string without any section boundary awareness. It finds all 6 bullet lines regardless of their parent heading. The Completeness check at score.ts:567–571 tests expEntries.length > 0 (where expEntries = input.parsed.experience ?? []), labels the failure "work experience", and deducts a full check slot.

The three keywords "volunteer", "volunteering", and "activities" currently live in SECTION_KEYWORDS.other (regex.ts:123–139), whose sole purpose is to act as a section boundary terminator — nothing renders an other section. Headings containing those words terminate the preceding section but do not open an experience section.

Decided fix — Compound A + C

Decision: student/volunteer headings count toward experienceCount AND drop the "work experience" Completeness deduction when bulletCount > 0.

Part A — Expand SECTION_KEYWORDS.experience (regex.ts:96–105)

Add student and non-standard synonyms so matchSectionHeader routes those headings into the experience section:

"volunteer experience"
"community service"
"leadership"
"involvement"
"on campus involvement"
"campus involvement"
"extracurricular"
"extracurricular activities"
"activities"          ← moved from other
"volunteer"           ← moved from other
"volunteering"        ← moved from other

Keyword move: "volunteer", "volunteering", and "activities" are removed from SECTION_KEYWORDS.other and added to SECTION_KEYWORDS.experience. They previously served as boundary terminators; after the move they open an experience section instead. Accept this trade-off (see Risks).

Part C — Adjust the Completeness scorer (score.ts:567–571)

De-emphasize the "work experience" check when the document has bullets but no parsed experience entries — indicating that extraction missed content rather than that the resume has none:

// score.ts ~line 567
completenessChecks.push({
  key: "experience",
  passed: expEntries.length > 0 || bullets.length > 0,
  label: "work experience",
});

This lets the bullet signal (already computed at line 530) serve as a fallback pass condition for the experience completeness check. The scorer still rewards resumes that produce parsed experience entries (enabling the dates sub-check at lines 580–587), but stops penalizing resumes where extraction failed to section-scope the content.

Acceptance criteria

1. After Part A: experienceCount ≥ 1 for the laverne fixture (at least one of "VOLUNTEER EXPERIENCE" or "ON CAMPUS INVOLVEMENT" is parsed into parsed.experience).
2. After Part C: "work experience" is absent from score.completeness.missing for the laverne fixture.
3. The laverne expected.json snapshot is updated to reflect both changes:
   • "experienceCount": ≥ 1
   • "work experience" removed from completeness.missing
   • Overall score will increase (exact value to be captured in the updated snapshot run).
4. Existing fixtures whose experienceCount > 0 are unaffected — the keyword expansion only adds new match paths, it does not change existing ones.
5. The google-docs-skia-proxy-nonstandard-headers and weasyprint-cairo-nonstandard-headers fixtures (which also have experienceCount: 0 and "work experience" in missing) should be evaluated for whether their headings now match; if they do, those snapshots update too. If they don't, that confirms the fix scope is bounded.
6. npm run test passes (including corpus snapshot verification) with UPDATE=1 re-run to bake the new snapshots.

Risks

Two-column boundary bleed. "volunteer", "volunteering", and "activities" previously terminated the preceding section in SECTION_KEYWORDS.other. Moving them to experience means a two-column resume with a sidebar "VOLUNTEER" label will now open an experience section instead of closing the preceding one, potentially bleeding sidebar content into the experience entries. Mitigations:

• The isTwoColumn layout probe (from LayoutProbes) is already in scope; a follow-up could gate the expanded keywords behind a !isTwoColumn condition if bleed is observed in practice.
• The existing two-column corpus fixtures (google-docs-skia-proxy-two-column, weasyprint-cairo-two-column, chromium-asymmetric-sidebar) must be re-run with UPDATE=1 and their snapshots eyeballed for regression before merging.

False-positive "leadership" / "involvement". A resume that uses "Leadership" as a sidebar widget label rather than a section heading could bleed into experience. The split-letter recovery already in matchSectionHeader (regex.ts:172) does not change this risk; it applies equally to any added keyword.

[s-annam]

Clarification (rewrite) — Body rewritten to current code citations (field is parsed.experience[] not work_experiences; SECTION_KEYWORDS.experience at regex.ts:96–105; SECTION_KEYWORDS.other volunteer/volunteering/activities at regex.ts:131–134; extractBulletsFromText at score.ts:508; "work experience" label at score.ts:570; experienceCount snapshot key at corpus.test.ts:113); fix narrowed to decided compound A+C: expand SECTION_KEYWORDS.experience with student/volunteer synonyms (moving volunteer, volunteering, activities out of other) and make the Completeness "work experience" check pass when bullets.length > 0, with two-column bleed risk documented.

[s-annam]

Implementation Complete — 5c96ac3 (PR #121)

Note: the original symptom (laverne experienceCount: 0 + "work experience" in completeness.missing) was already resolved by #115's anchor-fallback — the on-disk snapshot already showed experienceCount: 1. AC1/AC2 were met before this PR. Per decision, implemented the full Part A + C as a generalization beyond the single fixture.

Acceptance Criteria

• AC1 — laverne experienceCount ≥ 1 — now 2 ("ON CAMPUS INVOLVEMENT" also routes after Part A; "VOLUNTEER EXPERIENCE" already routed via feat: section-recognition L2/L3 — anchor fallback + visual bounds #115)
• AC2 — "work experience" absent from laverne missing — confirmed (already true via feat: section-recognition L2/L3 — anchor fallback + visual bounds #115; preserved)
• AC3 — laverne snapshot updated — experienceCount 1→2, confidence 0→0.9, escalation ocr→none
• AC4 — existing experienceCount > 0 fixtures unaffected — verified; all two-column fixtures unchanged (no boundary bleed)
• AC5 — nonstandard-headers fixtures evaluated — both google-docs-skia-proxy-nonstandard-headers and weasyprint-cairo-nonstandard-headers now match: experienceCount 2→3, educationCount 2→1
• AC6 — npm run test passes with snapshots re-baked — 454/454 green; npm run typecheck clean

Additional Changes

• Part A — expanded sections.config.json experience.aliases with student/volunteer synonyms; removed volunteer/volunteering from other.
• Part C — score.ts experience completeness check: passed: expEntries.length > 0 || bullets.length > 0.
• Deviation from body — "activities" alias was moved from achievements → experience (not other as the body stated; it had migrated to achievements in the refactor: externalize section keyword config to sections.config.json #113 config externalization). Effect: student-projects-activities fixture achievementsCount 4→0, experienceCount 2→3 — score-neutral (whole score block unchanged), confidence 0→0.86.
• Tests updated — sections.test.ts and score.test.ts hardcoded expectations adjusted for the alias expansion.
• awesome-cv fixture: score 63→67, "work experience" drops from missing.

[s-annam]

Implementation Complete — 4cb5745

Landed via PR #121 (squash). Both decided parts shipped.

Acceptance Criteria

• Part A — Expand SECTION_KEYWORDS.experience — Added student/non-standard synonyms (volunteer experience, community service, leadership, involvement, on campus involvement, campus involvement, extracurricular(s), activities); moved volunteer/volunteering/activities out of SECTION_KEYWORDS.other into experience so matchSectionHeader routes those headings into the experience section. Config now lives in sections.config.json.
• Part C — Completeness bullets fallback — score.ts no longer deducts the "work experience" check when bulletCount > 0 but parsed experience is empty, treating it as missed extraction rather than absent content.

Additional Changes

• Fixture snapshots regenerated — openresume-laverne-word-quartz now reflects experienceCount > 0 and drops "work experience" from completeness.missing; ~28 *.expected.json snapshots updated for the keyword/scoring shift.
• Two-column fixtures unchanged — chromium-asymmetric-sidebar, chromium-two-column-sidebar, google-docs-two-column, weasyprint-two-column, two-column-achievements-sidebar verified no regression.
• Tests — regex.test.ts, sections.test.ts, score.test.ts extended; verify + fallow CI green.

Reviewed and approved by @Vaishnavi1709.