69 lines
2.4 KiB
JSON
69 lines
2.4 KiB
JSON
{
|
|
"schemaVersion": 1,
|
|
"tuningVersion": "1.0.1",
|
|
"status": "frozen",
|
|
"purpose": "Weight, alias, and ontology tuning only; never used as held-out release evidence.",
|
|
"index": "index.json",
|
|
"caseRoot": "cases",
|
|
"goldRoot": "gold",
|
|
"caseCount": 60,
|
|
"casesPerIntent": 10,
|
|
"intents": [
|
|
"web-application-delivery",
|
|
"api-backend-delivery",
|
|
"test-qa-automation",
|
|
"security-review-hardening",
|
|
"deployment-devops",
|
|
"agent-mcp-development"
|
|
],
|
|
"separation": {
|
|
"heldOutInputsIncluded": false,
|
|
"heldOutGoldIncluded": false,
|
|
"taskFamilyFingerprintsDisjoint": true,
|
|
"verifiedAgainstHeldOutCaseCount": 180
|
|
},
|
|
"provenance": {
|
|
"source": "independent-tuning-gold-equivalence-audit",
|
|
"version": "1.0.1",
|
|
"reviewedAt": "2026-07-17"
|
|
},
|
|
"equivalenceAudit": {
|
|
"auditId": "aas-v1-tuning-gold-equivalence-1.0.1",
|
|
"path": "../../reviews/tuning-gold-equivalence-audit.json",
|
|
"sha256": "d280cf0021d3d3be527ece98e355bf93724d96a1bd6d718bb8cf48703478121e",
|
|
"diagnosticSha256": "ae33eb201792dbcd5b28e5a71de36594f6c0e38709b6c86848de4420328f7001",
|
|
"omittedInclusionsAudited": 19,
|
|
"changedPairs": 7
|
|
},
|
|
"reviews": [
|
|
{
|
|
"reviewer": "gold-equivalence-independent-alpha",
|
|
"decision": "approved",
|
|
"scope": "primary-disjoint-a",
|
|
"report": "../../reviews/reviewer-tuning-equivalence-alpha.json",
|
|
"reportSha256": "ec48b476a76a0332114e65db25badd92737ac6547c5a397d9456b2be4fb11689"
|
|
},
|
|
{
|
|
"reviewer": "gold-equivalence-independent-beta",
|
|
"decision": "approved",
|
|
"scope": "primary-disjoint-b",
|
|
"report": "../../reviews/reviewer-tuning-equivalence-beta.json",
|
|
"reportSha256": "01cbe88f9d55690b32e61cbd567c3560987654dbd4bee70daec926d740e060c3"
|
|
},
|
|
{
|
|
"reviewer": "gold-equivalence-independent-adjudicator",
|
|
"decision": "approved",
|
|
"scope": "independent-adjudication",
|
|
"report": "../../reviews/reviewer-tuning-equivalence-adjudicator.json",
|
|
"reportSha256": "ad8c87f03287989b7dbbacaebc42aa11e9f4564a0f08af2b6361e02d389bd160"
|
|
},
|
|
{
|
|
"reviewer": "gold-equivalence-independent-ci-tiebreak",
|
|
"decision": "approved",
|
|
"scope": "single-claim-tiebreak",
|
|
"report": "../../reviews/reviewer-tuning-equivalence-ci-tiebreak.json",
|
|
"reportSha256": "dbc01ab488cd904370af0a27f9f14a92891edd0187549ace5c48d8cc88d9bebf"
|
|
}
|
|
]
|
|
}
|