Portfolio pattern signals — Evidence

The five signal families

Failed/mixed replication dependency: papers citing foundations with failed or mixed repeat-attempt records in FLoRA.
Known problematic-paper concentration: portfolio works present in source datasets such as PPS or recorded as retracted.
Flagged-reference rate: papers whose extracted references include a source-attributed problematic-paper or retraction observation.
Screening provenance gap: completed screening records whose public signal remains limited or withheld.
Citation-overlap cluster: works in the same scope with identical fingerprints across at least five extracted DOI references.

Windows and denominators

Every snapshot stores current and baseline numerators, denominators, rates, and bounded examples. A rate is not interpreted until at least 10 eligible records exist. Lower-coverage portfolios return insufficient_data, not a clean result. Stale graph inputs produce limited.

Public snapshots cover catalogue source and subject. The public API exposes these aggregates. Partner snapshots are returned only through an authenticated publisher key and cannot be selected by public query parameters.

Interpretation boundary: severity means triage urgency. It is not a probability of wrongdoing, scientific error, or shared authorship. No person-level scores are created.

Known limits

Reference coverage depends on cached full text and resolvable DOI strings. Exact citation overlap is deliberately conservative and can reflect standard methods, shared datasets, or field conventions. External sources have uneven disciplinary and historical coverage. All examples require inspection in their original scholarly context.