{"schema":1,"historical":true,"generatedAt":"2026-09-09T22:20:34.367Z","sourceRevision":"fe25dd3d877da4b6257b17946a8289bb20a1b0d7","sourceSha256":"3e4785a02f26b3d9ecda2e2f1d76d07fcaba589b4b84c7523081eb0d4fff79dd","corpusHash":"a600e51263fec471f946df0c99d9b11d2cd4531b31053e0bfc2b2527fcee058b","labelCorpusHash":"1420d480e1522b0a1a07215766217c6aa8e9d4ca0d2eb0ab189d3b903523e256","modelArtifact":"Xenova/bge-small-en-v1.5@q8 rev ea104dacec62","configVersion":"bge-small-en-v1.5:q8:cls:norm:tok512","rubricVersion":"0.2.0-draft","suites":[{"id":"ask-retrieval@calibration","n":25,"status":"ran","metrics":{"hitAt5Rate":0.8888888888888888,"ndcgAt5":0.6626809312390524,"mrr":0.8381834215167548,"inCorpusN":18,"oodN":4,"adjacencyN":3},"ci":{"low":0.672002348698212,"high":0.9689804773543876},"confusion":null,"gate":{"passed":true,"enforcedAtMeasurement":false}},{"id":"ask-retrieval@holdout","n":15,"status":"ran","metrics":{"hitAt5Rate":1,"ndcgAt5":0.5648438200082895,"mrr":1,"inCorpusN":10,"oodN":3,"adjacencyN":2},"ci":{"low":0.7224672001371107,"high":1},"confusion":null,"gate":{"passed":true,"enforcedAtMeasurement":true}},{"id":"policy-goldens@calibration","n":21,"status":"ran","metrics":{},"ci":null,"confusion":null,"gate":{"passed":true,"enforcedAtMeasurement":false}},{"id":"negative-intent@calibration","n":389,"status":"ran","metrics":{"precision":1,"controlN":92,"controlFired":0,"targetN":297,"targetFired":297,"targetRecall":1},"ci":{"low":0.9872309526893244,"high":1},"confusion":null,"gate":{"passed":true,"enforcedAtMeasurement":false}},{"id":"negative-intent@holdout","n":389,"status":"ran","metrics":{"precision":1,"controlN":91,"controlFired":0,"targetN":298,"targetFired":298,"targetRecall":1},"ci":{"low":0.9872732565112068,"high":1},"confusion":null,"gate":{"passed":true,"enforcedAtMeasurement":true}},{"id":"jd-requirements@calibration","n":14,"status":"ran","metrics":{"documents":14,"goldAtoms":453,"extractionRecallMacro":0.9865351268712612,"extractionRecallLow":0.9756202480992398,"extractionRecallHigh":0.9957983193277312,"protectedTypingPrecision":1,"protectedTypingRecall":1,"coordinationCollapses":1,"omitted":0},"ci":{"low":0.9756202480992398,"high":0.9957983193277312},"confusion":{"labels":["hard_gate","quantified_experience","tool_vendor","capability_outcome","domain","preference","unclassified"],"rows":[{"actual":"hard_gate","counts":[43,0,0,0,0,0,0]},{"actual":"quantified_experience","counts":[0,37,0,0,0,0,0]},{"actual":"tool_vendor","counts":[0,0,36,9,0,1,3]},{"actual":"capability_outcome","counts":[0,0,5,91,0,0,57]},{"actual":"domain","counts":[0,0,0,0,2,1,3]},{"actual":"preference","counts":[0,0,0,0,2,57,43]},{"actual":"unclassified","counts":[0,0,0,6,0,0,51]}]},"gate":{"passed":true,"enforcedAtMeasurement":false}},{"id":"jd-requirements@holdout","n":12,"status":"ran","metrics":{"documents":12,"goldAtoms":414,"extractionRecallMacro":0.9280540161466888,"extractionRecallLow":0.8914138045064769,"extractionRecallHigh":0.959439712187557,"protectedTypingPrecision":0.9041095890410958,"protectedTypingRecall":0.9295774647887324,"coordinationCollapses":3,"omitted":0},"ci":{"low":0.8914138045064769,"high":0.959439712187557},"confusion":{"labels":["hard_gate","quantified_experience","tool_vendor","capability_outcome","domain","preference","unclassified"],"rows":[{"actual":"hard_gate","counts":[35,1,0,1,0,0,2]},{"actual":"quantified_experience","counts":[0,30,0,0,0,0,2]},{"actual":"tool_vendor","counts":[0,1,29,4,0,0,8]},{"actual":"capability_outcome","counts":[0,1,1,63,0,0,63]},{"actual":"domain","counts":[0,0,0,0,3,0,2]},{"actual":"preference","counts":[1,0,0,0,0,41,52]},{"actual":"unclassified","counts":[4,0,0,7,0,0,32]}]},"gate":{"passed":false,"enforcedAtMeasurement":true}},{"id":"jd-requirements@calibration-2","n":14,"status":"ran","metrics":{"documents":14,"goldAtoms":549,"extractionRecallMacro":0.9791893344433265,"extractionRecallLow":0.9571771704045844,"extractionRecallHigh":0.9940689537463732,"protectedTypingPrecision":1,"protectedTypingRecall":0.9891304347826086,"coordinationCollapses":0,"omitted":0},"ci":{"low":0.9571771704045844,"high":0.9940689537463732},"confusion":{"labels":["hard_gate","quantified_experience","tool_vendor","capability_outcome","domain","preference","unclassified"],"rows":[{"actual":"hard_gate","counts":[41,0,0,0,0,0,1]},{"actual":"quantified_experience","counts":[0,50,0,0,0,0,0]},{"actual":"tool_vendor","counts":[0,0,33,5,0,1,7]},{"actual":"capability_outcome","counts":[0,0,10,85,0,0,138]},{"actual":"domain","counts":[0,0,0,0,0,0,0]},{"actual":"preference","counts":[0,0,0,0,0,41,47]},{"actual":"unclassified","counts":[0,0,0,5,0,2,72]}]},"gate":{"passed":true,"enforcedAtMeasurement":false}}]}