mateclaw/mateclaw-server/src/test/resources/agent-evaluation/goal-service-baseline-v1.json

178 lines
5.1 KiB
JSON

{
"schemaVersion" : 1,
"suiteId" : "goal-service-boundaries-v1",
"suiteSha256" : "ca38905298a2b8cf1379b14f04b4a4e52b759504d8aa2314c52e6a37eb3cdaa6",
"revisionLabel" : "b166793d1+cycle015-working",
"productionClassSha256" : {
"vip.mate.goal.service.GoalServiceImpl" : "7151a5e4a8918f16f8e17d9c9fa478086d15b24a1d705cb87811505be40c5387",
"vip.mate.goal.model.GoalCriteriaCodec" : "6963dbb3a89736cc0e75ebd40d68d3df7b6537d2e35a946e39f1dad7a9345d7b",
"vip.mate.goal.model.GoalEntity" : "0ba945500694f12bce2ac302927e8d468b303a36f96c78be2a07143539a07163",
"vip.mate.goal.model.GoalEvaluationResult" : "bc150a255e8503a8760b5ff323562da4c2f93ba11c9ec17df65334a18f725d87"
},
"databaseProduct" : "H2",
"latestMigration" : "193",
"mockedBoundaries" : [ "MemoryManager" ],
"executionMode" : "offline_h2_goal_service_scenarios",
"onlineModelCalls" : 0,
"agentTaskSuccessRate" : "not_measured",
"onlineCost" : "not_measured",
"matchedCases" : 6,
"mismatchedCases" : 0,
"cases" : [ {
"id" : "current-revision-completion",
"task" : "current revision completion",
"source" : "cycles 004/006/010/013 real service boundaries",
"expected" : {
"status" : "completed",
"score" : 1.0,
"criteriaCount" : 1,
"passedCount" : 1,
"firstCriterionText" : "report",
"evaluationRevision" : 1,
"recordedEvalCallsUsed" : 1,
"completionCode" : 200
},
"actual" : {
"status" : "completed",
"score" : 1.0,
"criteriaCount" : 1,
"passedCount" : 1,
"firstCriterionText" : "report",
"evaluationRevision" : 1,
"recordedEvalCallsUsed" : 1,
"completionCode" : 200
},
"matched" : true,
"error" : null
}, {
"id" : "append-before-verdict",
"task" : "append before verdict",
"source" : "cycles 004/006/010/013 real service boundaries",
"expected" : {
"status" : "active",
"score" : 0.5,
"criteriaCount" : 2,
"passedCount" : 1,
"firstCriterionText" : "report",
"evaluationRevision" : 0,
"recordedEvalCallsUsed" : 1,
"completionCode" : 409
},
"actual" : {
"status" : "active",
"score" : 0.5,
"criteriaCount" : 2,
"passedCount" : 1,
"firstCriterionText" : "report",
"evaluationRevision" : 0,
"recordedEvalCallsUsed" : 1,
"completionCode" : 409
},
"matched" : true,
"error" : null
}, {
"id" : "pause-before-verdict",
"task" : "pause before verdict",
"source" : "cycles 004/006/010/013 real service boundaries",
"expected" : {
"status" : "paused",
"score" : 0.0,
"criteriaCount" : 1,
"passedCount" : 0,
"firstCriterionText" : "report",
"evaluationRevision" : 0,
"recordedEvalCallsUsed" : 2,
"completionCode" : 409
},
"actual" : {
"status" : "paused",
"score" : 0.0,
"criteriaCount" : 1,
"passedCount" : 0,
"firstCriterionText" : "report",
"evaluationRevision" : 0,
"recordedEvalCallsUsed" : 2,
"completionCode" : 409
},
"matched" : true,
"error" : null
}, {
"id" : "append-before-bootstrap",
"task" : "append before bootstrap",
"source" : "cycles 004/006/010/013 real service boundaries",
"expected" : {
"status" : "active",
"score" : 0.0,
"criteriaCount" : 1,
"passedCount" : 0,
"firstCriterionText" : "user appendix",
"evaluationRevision" : 0,
"recordedEvalCallsUsed" : 1,
"completionCode" : 0
},
"actual" : {
"status" : "active",
"score" : 0.0,
"criteriaCount" : 1,
"passedCount" : 0,
"firstCriterionText" : "user appendix",
"evaluationRevision" : 0,
"recordedEvalCallsUsed" : 1,
"completionCode" : 0
},
"matched" : true,
"error" : null
}, {
"id" : "replace-definition",
"task" : "replace definition",
"source" : "cycles 004/006/010/013 real service boundaries",
"expected" : {
"status" : "active",
"score" : 0.0,
"criteriaCount" : 0,
"passedCount" : 0,
"firstCriterionText" : null,
"evaluationRevision" : 1,
"recordedEvalCallsUsed" : 2,
"completionCode" : 409
},
"actual" : {
"status" : "active",
"score" : 0.0,
"criteriaCount" : 0,
"passedCount" : 0,
"firstCriterionText" : null,
"evaluationRevision" : 1,
"recordedEvalCallsUsed" : 2,
"completionCode" : 409
},
"matched" : true,
"error" : null
}, {
"id" : "aba-definition",
"task" : "aba definition",
"source" : "cycles 004/006/010/013 real service boundaries",
"expected" : {
"status" : "active",
"score" : 0.0,
"criteriaCount" : 0,
"passedCount" : 0,
"firstCriterionText" : null,
"evaluationRevision" : 2,
"recordedEvalCallsUsed" : 1,
"completionCode" : 0
},
"actual" : {
"status" : "active",
"score" : 0.0,
"criteriaCount" : 0,
"passedCount" : 0,
"firstCriterionText" : null,
"evaluationRevision" : 2,
"recordedEvalCallsUsed" : 1,
"completionCode" : 0
},
"matched" : true,
"error" : null
} ]
}