Evaluations, Claims, and Evidence

Evaluation

An Evaluation assesses a fixed Prediction Set with selected Performance metrics and compatible Robustness, Fairness, or Explainability modules. Its setup and execution can change state while work is planned or running.

Knotest compiles the requested configuration and revalidates current Project inputs when the Evaluation starts. A prior Ready state is not a reserved ticket.

Claim

At a terminal outcome, the Evaluation seals a Claim. The outcome is exactly completed, failed, or cancelled. The Claim binds the compiled plan, ordered stage records, provenance, and hashed Evidence.

A completed Claim never changes. Renaming the Evaluation affects separate presentation metadata rather than rewriting the Claim.

Evidence

Evidence is the content-identified output generated during an Evaluation: metrics, tables, images, logs, and failure diagnostics. Evidence completeness is separate from Claim outcome, so a failed or cancelled Claim may retain useful Evidence without claiming a complete assessment.

Evaluation (planned or running)
              |
              v
Claim (completed, failed, or cancelled) --> Evidence

Report drafts and exported reports reference sealed Evidence. They are deliverables, not replacements for the Claim or new Evaluation Evidence.