An inspectable engineering example

EvalDeck

Change a saved answer and inspect exactly which output contracts pass, fail, or regress from a synthetic baseline.

Example contract

Baseline answerCandidate answer

These are hand-authored replay examples, not live model outputs. The browser ports the Python check rules; no Python interpreter or provider runs here. Numeric JSON values are restricted to finite JavaScript numbers with safe integer values; the native CLI supports larger Python integers.

Contract results

Check	Baseline	Candidate	Candidate explanation

Inspect the contract

The gate compares this case’s pass/fail status. The native CLI also validates JSONL datasets, fingerprints changed expectations, detects removed cases, and writes escaped HTML reports. Passing a contract does not establish factual or model quality.