| name | claim-evidence-ledger |
| description | Use when auditing mathematical paper drafts for supported claims, missing citations, overclaims, proof status, experiment support, or source-to-text traceability. |
Claim Evidence Ledger
Build a claim ledger before polishing prose. The goal is to make every
substantive statement traceable to a proof, experiment, citation, or explicit
uncertainty.
Inputs
- Paper draft or target section.
- Source packet: theorem statements, proof notes, experiment logs, figures,
bibliography, related-work notes, reviewer comments, or prior drafts.
- Optional venue/audience constraints and citation policy.
If sources are missing, request the source packet or produce only an evidence
request.
Output Contract
Return a table with:
| Field | Meaning |
|---|
claim | The exact or paraphrased paper claim |
location | Section, paragraph, theorem, figure, or equation |
type | theorem, proof, experiment, citation, positioning, limitation, future-work |
support | Source file, theorem, log, citation key, or reviewer comment |
status | supported, citation-needed, needs-proof, needs-experiment, overclaimed, unsupported, uncertain |
action | Keep, cite, soften, remove, verify, or ask human |
Workflow
- Extract claims from the target text.
- Classify each claim by mathematical, empirical, literature, or contribution
role.
- Match claims to source evidence.
- For mathematical claims, distinguish "has a proof sketch" from "assumptions,
dependencies, and proof obligations have been audited"; route the latter to
proof-obligation-and-assumption-audit when needed.
- Mark unsupported claims instead of rewriting them as if they were true.
- Suggest edits that preserve the strongest source-backed version.
Review Rules
- Do not invent citation keys, theorem numbers, experiments, or author claims.
- Keep theorem/proof status distinct from conjecture, heuristic, or empirical
evidence.
- Do not mark a theorem claim fully supported when assumptions, dependency fit,
or edge cases are still unaudited.
- Treat abstract, introduction, and conclusion claims as high risk because they
compress and often overstate results.