forensic-review

name	forensic-review
description	Stress-test a manuscript by mapping supplied evidence to scientific concerns, consequences, and revision locations. Use for structured scientific critique, aspect-coverage audits, unresolved-concern tracking, and fatal-major-minor prioritization without impersonating an editor or reproducing peer-review corpora.

Forensic Review

Produce an attributable, challengeable scientific critique from supplied material. Do not imitate a named journal, editor, reviewer, or historical review.

Evidence discipline

Use only supplied material and outputs from tools actually executed in the current run. Distinguish:

• missing evidence: material is absent or unreported;
• negative evidence: supplied material directly demonstrates a flaw, failed criterion, inconsistency, or contradiction;
• uncertain: multiple readings remain plausible.

Never infer misconduct or fabrication from a statistical anomaly, writing style, missing item, or model impression. Never claim access to a private or real peer-review corpus. Anchor concerns to stable supplied locations.

Coverage

Review these six dimensions independently, when assessable:

1. Data chain and integrity — provenance, transformations, exclusions, measurements, and raw or analysis output trace when supplied.
2. Methods — design, setting, population, comparators, bias, analysis, uncertainty, sensitivity, and robustness.
3. Result accuracy — abstract, body, table, figure, supplement, and supplied analysis-output consistency.
4. Conclusion boundary — claim-to-evidence alignment, causal or mechanistic overreach, external validity, and uncertainty.
5. Novelty — bounded contribution relative only to the supplied or actually retrieved evidence base.
6. Scientific story — whether problem, gap, design choices, results, and discussion form a coherent, falsifiable, evidence-bounded chain.

Mark each area reviewed, partially reviewed, or not assessable. If a table, figure, supplement, or binary artifact cannot be parsed, mark its checks not assessable; do not infer its contents.

Build a claim-evidence matrix for consequential claims. When numerical material is readable, also build a numeric consistency ledger that maps each reported value across abstract, body, table, figure, supplement, and supplied analysis output, and records every conflict or unavailable comparison.

Concern ledger

Give every concern a stable ID and keep these fields distinct:

evidence state → anchor → concern → consequence → revision location → uncertainty

Set priority:

• fatal: supplied evidence defeats a central claim unless new design, evidence, or analysis is added; this is not an editorial verdict;
• major: materially weakens validity, interpretation, reproducibility, or reporting;
• minor: localized clarity, consistency, presentation, or reporting work.

Track status as unresolved, partially resolved, resolved in supplied material, or not assessable. A suggestion does not resolve a concern.

If required material cannot be parsed or the review boundary cannot be established, fail closed as BLOCKED. Do not auto-edit the manuscript.

Call record_checkpoint with the strongest evidence boundary, uncertainty, or stop condition before returning. Prefix the statement with [forensic-review].

Output

Return:

1. review boundary;
2. aspect coverage;
3. priority summary;
4. evidence-to-concern map;
5. claim-evidence matrix;
6. numeric consistency ledger or not assessable;
7. unresolved concern ledger;
8. revision order;
9. uncertainties and non-assessed areas.

Do not recommend acceptance or rejection or speak for an editor.