Skip to content

Latest commit

 

History

History
32 lines (22 loc) · 2.15 KB

File metadata and controls

32 lines (22 loc) · 2.15 KB

Methods

pcr combines deterministic routing with multiple detector families. Each detector should declare applicable input types, dependency status, method limits, and false-positive risk.

Detector Families

Area Examples Notes
Raw data rules Digit distributions, repeated rows/columns, column relationships, category shape, missingness concentration Useful for surfacing review leads; many signals are weak and context-sensitive.
Summary-stat cross-checks N/mean/SD/SE/CI/count/percentage/t/df/p consistency More deterministic when input tables are parsed correctly and columns are identifiable.
Statistical text APA/NHST consistency through statcheck Depends on text extraction quality and R package availability.
Likert/integer feasibility GRIM/GRIMMER/DEBIT and SPRITE-style reconstruction Depends on scale assumptions and R package availability.
P-value collections Domain validity and just-significant clustering signals Domain errors are strong; distributional signals need careful context.
Project audit References, cited claims, paper-mill light signals, cross-material reconciliation, code reruns Designed for mixed manuscript/data/code packages.
Image triage Internal duplicates, copy-move candidates, metadata checks, blot/gel checklist Weak-signal triage only; use original images for serious review.
Provenance and corpus SHA-256 ledger, change verification, local cross-manuscript similarity Useful for package history and local corpus screening.

Routing

Use routing to decide tool applicability:

pcr-audit route path/to/input --json build/route.json

The route output records ready tools, missing dependencies, skipped checks, and unsupported inputs as structured data. Missing tools and dependencies should be recorded as level: info, not converted into risk findings.

Outputs

Every finding JSON should include tool_id, tool_name, detector_runtime, dependency_status, source, input_type, and findings[].

Markdown reports are readable summaries. JSON reports are the preferred machine-readable artifact for merging, automation, and audit trails.