Methodology

QRI reports are scored against the methodology below, which acts as the source of truth for evaluation criteria, stage caps, evidence handling, readiness tags, confidence levels, assurance notes, critical quantum blockers, user urgency status, and review status.

For generated reports, an evidence-gathering prompt and OpenRouter web search assemble a timestamped source dossier before scoring begins.

Maintainers can add project-specific curated evidence and human notes; those notes are treated as leads that models must verify, apply, qualify, or reject with an explicit human_note_treatment trace.

Evaluator prompts then analyze the project against the QRI specification using the evidence dossier and additional primary-source search where support is missing.

A final synthesis prompt reconciles disagreements, normalizes evidence, fact-checks claims against supplied materials, separates score-reducing quantum issues from assurance-only caveats, and produces one structured report. Public pages show the exact model IDs used when a report has been generated.

Generated reports are treated as evidence-assisted drafts unless a maintainer marks them reviewed or published.