Validate Bibliography

Cross-reference citations in lecture files against bibliography entries. Two modes:

• Default (structural): missing entries, unused entries, malformed fields, typo candidates.
• --semantic: adds citation-drift detection (duplicate entries for the same paper), DOI verification via crossref, and citation-style consistency within each file.

Report saved to quality_reports/bib_audit_[structural|semantic].md.

Mode 1: Structural (default)

Steps

1. Read the bibliography file and extract all citation keys.

2. Scan lecture files for citation keys:

   • .tex: \cite{, \citet{, \citep{, \citeauthor{, \citeyear{, \textcite{, \parencite{
   • .qmd / .md: @key, [@key], [@key1; @key2]
   • Extract all unique citation keys used.

3. Cross-reference:

   • Missing entries (CRITICAL): cited in lectures, absent from .bib.
   • Unused entries (informational): in .bib but never cited.
   • Typo candidates: keys within edit-distance 2 of a .bib key (e.g., Smith2020 vs Smth2020).

4. Check entry quality:

   • Required fields present (author, title, year, journal/booktitle).
   • Author field properly formatted.
   • Year in 1900–current.
   • No malformed characters / encoding issues.
   • doi field normalized (no leading https://doi.org/).

5. Write report to quality_reports/bib_audit_structural.md.

Files scanned

Slides/*.tex
Quarto/*.qmd
guide/*.qmd
master_supporting_docs/**/*.tex

Bibliography location

Bibliography_base.bib at repo root by default; override via CLAUDE.md.

Mode 2: Semantic (--semantic)

Everything in Mode 1, plus:

2a. Citation drift detection

Multiple .bib entries describing the same paper under different keys. Symptoms:

• Smith2020 + Smith2020a with identical DOI or title.
• CallawaySantAnna2021 + CS2021 both pointing to the same paper.
• Collaborator-merged .bib files.

Detection heuristics (any → FLAG):

Check	Signal
Same DOI across keys	Hard-duplicate (CRITICAL)
Same title (case-insensitive, punct-stripped)	Likely duplicate (CRITICAL)
Same author+year+journal	Probable duplicate (MEDIUM)
Title Jaccard > 0.85 on tokens ≥ 4 chars	Soft-duplicate (LOW)

For each flagged pair: list both keys, where each is cited, and recommend a canonical key (prefer most-cited, then alphabetically first).

2b. DOI verification (optional; network)

For each entry with a doi, fetch https://api.crossref.org/works/{doi} and compare:

• First-author last name
• Year
• Title (Jaccard > 0.7 on normalized tokens)
• Container-title / journal (exact or abbreviation)

Severity:

• Author or title mismatch → CRITICAL (wrong paper)
• Year mismatch → MEDIUM (preprint vs published, or typo)
• Journal mismatch → LOW (legitimate preprint variants)

Rate limit: cap 50 lookups per run, 0.5s delay between calls. Cache in quality_reports/.doi_cache.json.

Opt-out: --skip-doi for offline or no-WebFetch environments.

2c. Style consistency within each file

For each file, count citation commands (\citet vs \citep vs \cite; @key vs [@key]). FLAG files with mixed styles without an obvious pattern (e.g., 20× \citep and 3× \cite in the same deck). Low-severity.

2d. Cite-claim sanity (flag-only)

Gated behind --cite-claim. For the top-10 most-cited works per file, WebFetch the crossref abstract and surface it beside the in-text context. No auto-judgment — humans decide if the claim matches.

This is existence/structure, not appropriateness. Deciding whether the cited paper actually says what the in-text claim attributes to it is `/verify-claims`'s job — it reads the source and grounds a supports / partial / contradicts verdict in quotes + pages (with the EXPLAINED escape for a defensible named alternative). --cite-claim only surfaces the abstract; for the verdict, run /verify-claims.

Report structure (quality_reports/bib_audit_semantic.md)

# Bibliography Semantic Audit

**Date:** YYYY-MM-DD
**Bibliography:** Bibliography_base.bib (N entries)
**Files scanned:** [list]

## Summary

| Check | Critical | Medium | Low |
|---|---|---|---|
| Structural | | | |
| Citation drift | | | |
| DOI verification | | | |
| Style consistency | 0 | 0 | |

## Critical Issues

### Duplicate entries
| Keys | Signal | Citations | Recommended canonical |
|---|---|---|---|

### DOI mismatches
| Key | Field | .bib value | crossref value |
|---|---|---|---|

## Medium / Low issues
…

## Next steps
1. Resolve duplicates — pick canonical key, update citations, remove orphans.
2. Fix DOI mismatches — verify paper in crossref or strip the wrong DOI.
3. Review style-consistency notes.

Exit behavior

• Structural: exit 0; report enumerates issues.
• Semantic: exit 0 if only LOW findings; exit 1 on any CRITICAL. Usable as a pre-submission gate.

Cross-references

• .claude/skills/review-paper/SKILL.md — pair for full pre-submission.
• .claude/skills/audit-reproducibility/SKILL.md — numeric-claims counterpart.
• .claude/skills/verify-claims/SKILL.md — citation appropriateness counterpart (does the cited paper support the claim?). This skill checks that a citation exists and is well-formed; /verify-claims checks that it holds.

What this skill does NOT do

• Judge whether a citation is used in the right context — whether the cited paper actually says what the claim attributes to it. That is /verify-claims's job (see 2d); this skill stays existence-and-structure only.
• Auto-fix your .bib file — all edits are recommendations.
• Check non-DOI identifiers (ISBN, arXiv, SSRN) — roadmap.