{"cases":[{"case_gates":[{"gate":"accepted","run_id":"authored-judgment-1"},{"gate":"accepted","run_id":"authored-judgment-2"},{"gate":"accepted","run_id":"authored-judgment-3"}],"case_id":"same-pass","goal":"Inspect the same-pass condition.","session_id":"session-0","targets":[{"assessed":3,"evaluator":"authored-goal","expected":3,"observations":[{"accepted":true,"error":null,"explanation":"Authored judgment; not a performance measurement.","label":null,"run_id":"authored-judgment-1","status":"assessed","value":1},{"accepted":true,"error":null,"explanation":"Authored judgment; not a performance measurement.","label":null,"run_id":"authored-judgment-2","status":"assessed","value":1},{"accepted":true,"error":null,"explanation":"Authored judgment; not a performance measurement.","label":null,"run_id":"authored-judgment-3","status":"assessed","value":1}],"observed_gate_disagreement":false,"observed_score_disagreement":false,"observed_values":[1],"passed":3,"rating":{"kind":"numeric","max":1,"min":0,"pass_at_least":0.8},"rejected":0,"revision":"control-1; no model or AWS call","span_id":null,"state":"same_gate","trace_id":null,"unassessed":0}]},{"case_gates":[{"gate":"rejected","run_id":"authored-judgment-1"},{"gate":"rejected","run_id":"authored-judgment-2"},{"gate":"rejected","run_id":"authored-judgment-3"}],"case_id":"same-reject","goal":"Inspect the same-reject condition.","session_id":"session-1","targets":[{"assessed":3,"evaluator":"authored-goal","expected":3,"observations":[{"accepted":false,"error":null,"explanation":"Authored judgment; not a performance measurement.","label":null,"run_id":"authored-judgment-1","status":"assessed","value":0},{"accepted":false,"error":null,"explanation":"Authored judgment; not a performance measurement.","label":null,"run_id":"authored-judgment-2","status":"assessed","value":0},{"accepted":false,"error":null,"explanation":"Authored judgment; not a performance measurement.","label":null,"run_id":"authored-judgment-3","status":"assessed","value":0}],"observed_gate_disagreement":false,"observed_score_disagreement":false,"observed_values":[0],"passed":0,"rating":{"kind":"numeric","max":1,"min":0,"pass_at_least":0.8},"rejected":3,"revision":"control-1; no model or AWS call","span_id":null,"state":"same_gate","trace_id":null,"unassessed":0}]},{"case_gates":[{"gate":"accepted","run_id":"authored-judgment-1"},{"gate":"rejected","run_id":"authored-judgment-2"},{"gate":"accepted","run_id":"authored-judgment-3"}],"case_id":"gate-flip","goal":"Inspect the gate-flip condition.","session_id":"session-2","targets":[{"assessed":3,"evaluator":"authored-goal","expected":3,"observations":[{"accepted":true,"error":null,"explanation":"Authored judgment; not a performance measurement.","label":null,"run_id":"authored-judgment-1","status":"assessed","value":1},{"accepted":false,"error":null,"explanation":"Authored judgment; not a performance measurement.","label":null,"run_id":"authored-judgment-2","status":"assessed","value":0},{"accepted":true,"error":null,"explanation":"Authored judgment; not a performance measurement.","label":null,"run_id":"authored-judgment-3","status":"assessed","value":1}],"observed_gate_disagreement":true,"observed_score_disagreement":true,"observed_values":[0,1],"passed":2,"rating":{"kind":"numeric","max":1,"min":0,"pass_at_least":0.8},"rejected":1,"revision":"control-1; no model or AWS call","span_id":null,"state":"disagreement","trace_id":null,"unassessed":0}]},{"case_gates":[{"gate":"accepted","run_id":"authored-judgment-1"},{"gate":"accepted","run_id":"authored-judgment-2"},{"gate":"accepted","run_id":"authored-judgment-3"}],"case_id":"score-only-change","goal":"Inspect the score-only-change condition.","session_id":"session-3","targets":[{"assessed":3,"evaluator":"authored-goal","expected":3,"observations":[{"accepted":true,"error":null,"explanation":"Authored judgment; not a performance measurement.","label":null,"run_id":"authored-judgment-1","status":"assessed","value":0.8},{"accepted":true,"error":null,"explanation":"Authored judgment; not a performance measurement.","label":null,"run_id":"authored-judgment-2","status":"assessed","value":0.9},{"accepted":true,"error":null,"explanation":"Authored judgment; not a performance measurement.","label":null,"run_id":"authored-judgment-3","status":"assessed","value":1}],"observed_gate_disagreement":false,"observed_score_disagreement":true,"observed_values":[0.8,0.9,1],"passed":3,"rating":{"kind":"numeric","max":1,"min":0,"pass_at_least":0.8},"rejected":0,"revision":"control-1; no model or AWS call","span_id":null,"state":"same_gate","trace_id":null,"unassessed":0}]},{"case_gates":[{"gate":"accepted","run_id":"authored-judgment-1"},{"gate":"incomplete","run_id":"authored-judgment-2"},{"gate":"incomplete","run_id":"authored-judgment-3"}],"case_id":"partial-judgments","goal":"Inspect the partial-judgments condition.","session_id":"session-4","targets":[{"assessed":1,"evaluator":"authored-goal","expected":3,"observations":[{"accepted":true,"error":null,"explanation":"Authored judgment; not a performance measurement.","label":null,"run_id":"authored-judgment-1","status":"assessed","value":1},{"accepted":null,"error":"Authored unavailable-context control.","explanation":"Authored judgment; not a performance measurement.","label":null,"run_id":"authored-judgment-2","status":"skipped","value":null},{"accepted":null,"error":null,"explanation":"No result for this declared target.","label":null,"run_id":"authored-judgment-3","status":"missing","value":null}],"observed_gate_disagreement":false,"observed_score_disagreement":false,"observed_values":[1],"passed":1,"rating":{"kind":"numeric","max":1,"min":0,"pass_at_least":0.8},"rejected":0,"revision":"control-1; no model or AWS call","span_id":null,"state":"incomplete","trace_id":null,"unassessed":2}]}],"configuration":{"model":"synthetic-control","model_parameters":{},"prompt_sha256":"aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","skills":{},"tools_sha256":"bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb"},"dataset":{"cases":[{"expected_skills":[],"goal":"Inspect the same-pass condition.","id":"same-pass"},{"expected_skills":[],"goal":"Inspect the same-reject condition.","id":"same-reject"},{"expected_skills":[],"goal":"Inspect the gate-flip condition.","id":"gate-flip"},{"expected_skills":[],"goal":"Inspect the score-only-change condition.","id":"score-only-change"},{"expected_skills":[],"goal":"Inspect the partial-judgments condition.","id":"partial-judgments"}],"id":"judge-stability-controls","version":"1"},"evaluators":[{"id":"authored-goal","level":"session","rating":{"kind":"numeric","max":1,"min":0,"pass_at_least":0.8},"revision":"control-1; no model or AWS call"}],"fixed_record_sha256":"8bd84193b28c09bffe8d140d9dacaf46d729c5300f9babc4b709a06d7dcb021c","mode":"fixed_record_judge_repetitions","provenance_kind":"synthetic","schema_version":"evalarc.judge-stability.v1","scope":"Descriptive agreement of imported judgments on the same declared recording and rubric. No agent or judge is executed. Repetition IDs and evaluator revisions are caller-declared; agreement does not establish independence, accuracy, calibration or task success.","sources":[{"bytes":5912,"file":"inputs/0001.json","provenance":{"description":"Five authored controls with three judgment sets on identical simulated spans. Scores and placeholder identities are synthetic; no model or AWS evaluation.","kind":"synthetic"},"run_id":"authored-judgment-1","sha256":"d3807680a44436922ed827cd7a6a120984f8e0d7478d3d93fee6da1e7ff13a70"},{"bytes":5993,"file":"inputs/0002.json","provenance":{"description":"Five authored controls with three judgment sets on identical simulated spans. Scores and placeholder identities are synthetic; no model or AWS evaluation.","kind":"synthetic"},"run_id":"authored-judgment-2","sha256":"230e9ddd3d6deac2785272e476d661f65597eef99e5059d9146fdc0628c7db9a"},{"bytes":5602,"file":"inputs/0003.json","provenance":{"description":"Five authored controls with three judgment sets on identical simulated spans. Scores and placeholder identities are synthetic; no model or AWS evaluation.","kind":"synthetic"},"run_id":"authored-judgment-3","sha256":"5b8bcfa81c62644211f6b527dfa607af5a631f8e9437001ff874d512cc6e616a"}],"summary":{"all_rejected_targets":1,"complete_targets":4,"gate_disagreements":1,"incomplete_targets":1,"not_applicable_targets":0,"repetitions":3,"required_targets":5,"score_disagreements":2}}