Skip to content

Cover-Letter Issue Schema

Simplified, field-compatible variant of paper-audit/references/ISSUE_SCHEMA.md. A cover-letter finding can be ingested by paper-audit's deep-review consolidation later by adding the dropped fields with default values.

Canonical record

json
{
  "title": "short issue title",
  "quote": "exact quote from the cover letter",
  "explanation": "why this matters and what remains problematic",
  "comment_type": "claim_accuracy|journal_fit|declaration_missing|presentation|tone|disclosure_consistency",
  "severity": "major|moderate|minor",
  "source_kind": "script|llm",
  "confidence": "high|medium|low|unverified",
  "source_section": "header|opening|body|contributions|fit|declarations|closing",
  "char_offset": 128,
  "manuscript_section_anchor": "abstract|introduction|results|conclusion|none",
  "evidence_anchor": [
    {
      "type": "citation|figure_or_table|metric|section|analysis_artifact|missing",
      "text": "visible anchor in manuscript"
    }
  ],
  "claim_strength": "unsupported|observed|supported|strong",
  "missing_evidence": ["specific support that is absent or unverified"],
  "allowed_wording": "bounded wording that stays within the evidence",
  "forbidden_wording": [
    "unbounded wording that would require stronger evidence"
  ],
  "quote_verified": true
}

Required fields

  • title
  • quote
  • explanation
  • comment_type
  • severity
  • source_kind

Optional fields

  • confidence — high / medium / low / unverified; demote to unverified when the cover-letter quote cannot be located.
  • source_section — which logical section of the letter the issue lives in; deterministic claim mapping emits header, opening, body, or closing, while semantic lanes may use contributions, fit, or declarations.
  • char_offset — zero-based character offset of the claim in the analyzed letter text; -1 when the sentence cannot be located.
  • manuscript_section_anchor — which section of the manuscript the claim should be supported by; none when the issue is about the letter itself (e.g. tone) and not a claim-vs-manuscript alignment.
  • evidence_anchor / claim_strength / missing_evidence / allowed_wording / forbidden_wording — added when the issue is about claim accuracy. Follows the CLAIM_EVIDENCE_CONTRACT.md contract.
  • quote_verified — populated by verify_letter_against_manuscript.py.

Dropped vs. paper-audit canonical

These fields are intentionally absent in v1; add them with default values when ingesting into paper-audit:

Dropped fieldDefault for ingestion
review_lane"presubmission_readiness"
root_cause_keyderive from comment_type + source_section + quote hash
gate_blockerfalse (cover-letter does not have a gate mode in v1)
related_sectionsuse manuscript_section_anchor instead

Comment-type semantics

  • claim_accuracy — letter claim is unsupported or overclaims relative to manuscript evidence.
  • journal_fit — pitch is mismatched with the target venue's scope or tier.
  • declaration_missing — required declaration absent (see template's required_declarations).
  • presentation — length, paragraph shape, citation tilde, or other surface form.
  • tone — AI-tone words, forbidden phrases, marketing language, or structural AI-trace signals (parallel openings, uniform sentence length, diverse promotional vocabulary).
  • disclosure_consistency — the cover letter and the manuscript disagree on generative-AI disclosure (one discloses and the other is silent, or the two contradict on polarity). Emitted by align_check.py, which reads both documents.

Severity guidance

  • major — declaration_missing for a template's required items; overt overclaim where claim_strength is unsupported; length exceeding the template's hard ceiling by ≥20%; journal_fit verdict of LOW.
  • moderate — overclaim that is observed but pushed beyond observed scope; missing optional but recommended declaration when manuscript needs it; journal_fit verdict of MEDIUM on the most-loaded sub-axis; an AI-disclosure inconsistency between the letter and the manuscript (disclosure_consistency).
  • minor — paragraph length warnings; AI-tone term frequency; weak topic starters; non-required journal-specific phrasings.

Released under the MIT License.