Agent-created results plus fresh deterministic checks of evidence records; no independent model baseline.

PASS different-normal-a: Actual saved note reconciled with its declared criteria; checker executed.
PASS different-normal-b: Actual saved note reconciled with its declared criteria; checker executed.
PASS stale-evidence-rejected: A PASS quote not present in the artifact was rejected.
PASS missing-input: Actual saved clarification requests source material instead of inventing it.

node.exe studio/prompt-spec/scripts/check-contract.mjs studio/prompt-spec/demo/output/contract-a.json studio/prompt-spec/demo/output
exit=0
{"checked":2,"scope":"evidence record and quoted text; human judgement not inferred","results":["PASS","PASS"]}

node.exe studio/prompt-spec/scripts/check-contract.mjs studio/prompt-spec/demo/output/contract-b.json studio/prompt-spec/demo/output
exit=0
{"checked":2,"scope":"evidence record and quoted text; human judgement not inferred","results":["PASS","PASS"]}

node.exe studio/prompt-spec/scripts/check-contract.mjs studio/prompt-spec/demo/output/reject-stale-contract.json studio/prompt-spec/demo/output
exit=1
Evidence no longer contains: Checkout succeeded with 100 payments

