Examples & contracts
Machine-readable examples and contracts.
Copy-ready EvalGlass artifacts — example scorecard and RunRecord JSON, an authority-safe config, a CI workflow, and the typed contract catalog — for developers and AI agents.
These are illustrative pre-alpha examples: the shapes match the current docs, but contract stability is still settling before a 0.1 tag. Nothing here implies generated data or an LLM judge can self-approve a gate — a fresh setup gates nothing.
Examples & templates
- scorecard.json the bounded run summary — verdict, metrics, authority, baseline state. Docs →
- runrecord.json the complete reproducible record — embedded scorecard, scores, provenance. Docs →
- evalglass.yaml authority-safe config — metrics, datasets, traces, thresholds, data policy. Docs →
- github-actions-ci.yml CI workflow preserving informational/pass/fail/blocked exit semantics. Docs →
- judge-calibration.json an uncalibrated judge cannot gate. Docs →
- baseline-comparison.json a delta is a regression claim only when comparable. Docs →
- trace-import.jsonl exported-trace import shape (OpenInference/OTel/local). Docs →
Contract catalog
The typed contracts — Scorecard, RunRecord, and config — are cataloged as machine-readable JSON at /contracts.json, each with field summaries, a pre-alpha status caveat, and links to the authoritative docs. It is a caveated catalog, not a stability promise.