FRAMEWORK 1.0 / Evidence protocol

Evidence protocol

Evidence must remain traceable from an observation to a source, transformation, claim, model, and review record.

Evidence levels

A stated idea without inspectable support.

A source with author, date, location, method, license, and stable identifier.

Source converted through a declared coding or measurement rule with uncertainty.

Multiple independent sources or methods support the same bounded claim.

Independent teams reproduce the result or explain bounded disagreement.

A preregistered model performs on unseen future data with harms evaluated.

Stable source ID, claim ID, evidence ID, model ID, review ID, and version.

Author or collector, institution, date, geography, population, acquisition method, and custody.

License, consent, access condition, privacy class, retention, deletion, and derivative-use limits.

Extraction, coding, cleaning, normalization, imputation, aggregation, dimensional reduction, and software version.

Coverage, missingness, uncertainty, bias, source dependence, translation, and known anomalies.

Supports, contradicts, qualifies, uses, tests, reviews, supersedes, corrects, or disputes.

Preserve supporting and opposing evidence; do not collapse disagreement into one score.

A reliable source may still be coded incorrectly; a disputed source may contain inspectable observations.

Corrections create a new version linked to the superseded record; the old version remains auditable.

Traceability does not require public exposure of personal, restricted, or high-risk data.