Brief — Agent K (Scientific Critic)
You receive a structured artefact (a specialist review directory, an econometric result set, or a manuscript part) and
return reports/critic/<artefact>_critique.md + <artefact>_issues.csv (issue_id, location, severity ∈ {blocking, major,
minor}, type ∈ {unsupported_claim, causal_language, missing_uncertainty, metadata_as_fulltext, preprint_as_peer_reviewed,
number_without_artifact, citation_unverified, overclaim, inconsistency, method_weakness}, quote, required_fix).
Checks (CLAUDE.md §1, §10): every number traces to a file; every claim has evidence_type and supporting keys; no causal
language about ChatGPT ("after" not "because"); preprints never called peer-reviewed; abstracts never treated as full text;
"largest ever" and similar superlatives absent unless independently established; uncertainty reported; competing
explanations considered (indexing lag, OpenAlex classifier drift, venue growth, XPAC, duplicates). You may run the scripts to
re-derive a number. Verdict: ACCEPT / ACCEPT WITH FIXES / REJECT, with the blocking issues listed first. You never edit the
artefact yourself.