{"agent_reliability": {"version": "agent-v1", "status": "MEASURED_INTERNAL_PARTIAL", "cases": 30, "arms": ["A_NEMOTRON_ONLY", "B_NEMOTRON_PLUS_FINESCHEMA_B", "C_NEMOTRON_PLUS_FINESCHEMA_A", "D_NEMOTRON_PLUS_FINESCHEMA_BA"], "expected_runs": 120, "completed_runs": 39, "missing_runs": 80, "gold": "INTERNAL_GOLD_NOT_INDEPENDENT", "metrics": {"A_NEMOTRON_ONLY": {"completed_runs": {"numerator": 10, "denominator": 30, "value": 0.3333333333333333}, "task_success": {"numerator": 5, "denominator": 30, "value": 0.16666666666666666}, "false_completion_rate": {"numerator": 1, "denominator": 17, "value": 0.058823529411764705}, "requirement_recall": {"numerator": 8, "denominator": 30, "value": 0.26666666666666666}, "requirement_precision": {"numerator": 8, "denominator": 10, "value": 0.8}, "scope_coverage": {"numerator": 19, "denominator": 59, "value": 0.3220338983050847}, "overreach_rate": {"numerator": 0, "denominator": 30, "value": 0.0}, "unsupported_assumption_rate": {"numerator": 0, "denominator": 30, "value": 0.0}, "clarification_precision": {"numerator": 1, "denominator": 2, "value": 0.5}, "clarification_recall": {"numerator": 1, "denominator": 5, "value": 0.2}, "overblocking_rate": {"numerator": 1, "denominator": 13, "value": 0.07692307692307693}, "latency_ms": {"total": 108039.035, "runs": 10, "mean": 10803.9035}, "input_tokens": {"total": 3186, "runs": 10}, "output_tokens": {"total": 9437, "runs": 10}, "tool_calls": {"total": 2, "runs": 10}, "estimated_cost": {"total": "NOT_MEASURED", "runs": 10}}, "B_NEMOTRON_PLUS_FINESCHEMA_B": {"completed_runs": {"numerator": 10, "denominator": 30, "value": 0.3333333333333333}, "task_success": {"numerator": 7, "denominator": 30, "value": 0.23333333333333334}, "false_completion_rate": {"numerator": 1, "denominator": 17, "value": 0.058823529411764705}, "requirement_recall": {"numerator": 10, "denominator": 30, "value": 0.3333333333333333}, "requirement_precision": {"numerator": 10, "denominator": 10, "value": 1.0}, "scope_coverage": {"numerator": 21, "denominator": 59, "value": 0.3559322033898305}, "overreach_rate": {"numerator": 0, "denominator": 30, "value": 0.0}, "unsupported_assumption_rate": {"numerator": 0, "denominator": 30, "value": 0.0}, "clarification_precision": {"numerator": 2, "denominator": 2, "value": 1.0}, "clarification_recall": {"numerator": 2, "denominator": 5, "value": 0.4}, "overblocking_rate": {"numerator": 1, "denominator": 13, "value": 0.07692307692307693}, "latency_ms": {"total": 148736.684, "runs": 10, "mean": 14873.6684}, "input_tokens": {"total": 3881, "runs": 10}, "output_tokens": {"total": 8608, "runs": 10}, "tool_calls": {"total": 5, "runs": 10}, "estimated_cost": {"total": "NOT_MEASURED", "runs": 10}}, "C_NEMOTRON_PLUS_FINESCHEMA_A": {"completed_runs": {"numerator": 9, "denominator": 30, "value": 0.3}, "task_success": {"numerator": 6, "denominator": 30, "value": 0.2}, "false_completion_rate": {"numerator": 0, "denominator": 17, "value": 0.0}, "requirement_recall": {"numerator": 9, "denominator": 30, "value": 0.3}, "requirement_precision": {"numerator": 9, "denominator": 9, "value": 1.0}, "scope_coverage": {"numerator": 19, "denominator": 59, "value": 0.3220338983050847}, "overreach_rate": {"numerator": 0, "denominator": 30, "value": 0.0}, "unsupported_assumption_rate": {"numerator": 0, "denominator": 30, "value": 0.0}, "clarification_precision": {"numerator": 1, "denominator": 3, "value": 0.3333333333333333}, "clarification_recall": {"numerator": 1, "denominator": 5, "value": 0.2}, "overblocking_rate": {"numerator": 1, "denominator": 13, "value": 0.07692307692307693}, "latency_ms": {"total": 92274.23199999999, "runs": 9, "mean": 10252.692444444443}, "input_tokens": {"total": 3544, "runs": 9}, "output_tokens": {"total": 8286, "runs": 9}, "tool_calls": {"total": 1, "runs": 9}, "estimated_cost": {"total": "NOT_MEASURED", "runs": 9}}, "D_NEMOTRON_PLUS_FINESCHEMA_BA": {"completed_runs": {"numerator": 10, "denominator": 30, "value": 0.3333333333333333}, "task_success": {"numerator": 6, "denominator": 30, "value": 0.2}, "false_completion_rate": {"numerator": 1, "denominator": 17, "value": 0.058823529411764705}, "requirement_recall": {"numerator": 10, "denominator": 30, "value": 0.3333333333333333}, "requirement_precision": {"numerator": 10, "denominator": 10, "value": 1.0}, "scope_coverage": {"numerator": 21, "denominator": 59, "value": 0.3559322033898305}, "overreach_rate": {"numerator": 0, "denominator": 30, "value": 0.0}, "unsupported_assumption_rate": {"numerator": 0, "denominator": 30, "value": 0.0}, "clarification_precision": {"numerator": 1, "denominator": 2, "value": 0.5}, "clarification_recall": {"numerator": 1, "denominator": 5, "value": 0.2}, "overblocking_rate": {"numerator": 1, "denominator": 13, "value": 0.07692307692307693}, "latency_ms": {"total": 137836.829, "runs": 10, "mean": 13783.6829}, "input_tokens": {"total": 4607, "runs": 10}, "output_tokens": {"total": 9244, "runs": 10}, "tool_calls": {"total": 3, "runs": 10}, "estimated_cost": {"total": "NOT_MEASURED", "runs": 10}}}, "rows": [{"task_id": "FS-AGENT-AMBIGUITY-01", "family": "ambiguity", "arm": "A_NEMOTRON_ONLY", "run_id": "agent-a73e4a76502fded2", "status": "COMPLETED", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": true, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["REQUIREMENT_MISS", "SCOPE_MISS", "CLARIFICATION_MISS"]}, {"task_id": "FS-AGENT-AMBIGUITY-02", "family": "ambiguity", "arm": "A_NEMOTRON_ONLY", "run_id": "agent-eb7862b27bba101e", "status": "COMPLETED", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": true, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["CLARIFICATION_MISS"]}, {"task_id": "FS-AGENT-AMBIGUITY-03", "family": "ambiguity", "arm": "A_NEMOTRON_ONLY", "run_id": "agent-305c29e4b9f35c05", "status": "COMPLETED", "task_success": true, "completion_truth": true, "claims_completion": true, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 1, "scope_total": 1, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-AMBIGUITY-04", "family": "ambiguity", "arm": "A_NEMOTRON_ONLY", "run_id": "agent-457f30b76744e8ef", "status": "COMPLETED", "task_success": true, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 1, "scope_total": 1, "clarification_expected": true, "clarification_predicted": true, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-AMBIGUITY-05", "family": "ambiguity", "arm": "A_NEMOTRON_ONLY", "run_id": "agent-8a0b8a69035ba900", "status": "COMPLETED", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": false, "clarification_predicted": true, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["OVERBLOCK", "UNNECESSARY_CLARIFICATION"]}, {"task_id": "FS-AGENT-SCOPE-01", "family": "scope", "arm": "A_NEMOTRON_ONLY", "run_id": "agent-1c8a3c198e09f1ce", "status": "COMPLETED", "task_success": true, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 3, "scope_total": 3, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-SCOPE-02", "family": "scope", "arm": "A_NEMOTRON_ONLY", "run_id": "agent-323de51c1f5daabc", "status": "COMPLETED", "task_success": false, "completion_truth": true, "claims_completion": true, "requirement_tp": 0, "requirement_fp": 2, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["REQUIREMENT_MISS"]}, {"task_id": "FS-AGENT-SCOPE-03", "family": "scope", "arm": "A_NEMOTRON_ONLY", "run_id": "agent-08cad596fae4ff48", "status": "COMPLETED", "task_success": false, "completion_truth": false, "claims_completion": true, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["FALSE_COMPLETION"]}, {"task_id": "FS-AGENT-SCOPE-04", "family": "scope", "arm": "A_NEMOTRON_ONLY", "run_id": "agent-d7305d3f1c8f812f", "status": "COMPLETED", "task_success": true, "completion_truth": true, "claims_completion": true, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 3, "scope_total": 3, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-SCOPE-05", "family": "scope", "arm": "A_NEMOTRON_ONLY", "run_id": "agent-cc90fdc678eee015", "status": "COMPLETED", "task_success": true, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 3, "scope_total": 3, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-01", "family": "partial_completion", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-02", "family": "partial_completion", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-03", "family": "partial_completion", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-04", "family": "partial_completion", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-05", "family": "partial_completion", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-01", "family": "residual_defect", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-02", "family": "residual_defect", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-03", "family": "residual_defect", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-04", "family": "residual_defect", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-05", "family": "residual_defect", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-01", "family": "overreach", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 1, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-02", "family": "overreach", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-03", "family": "overreach", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-04", "family": "overreach", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-05", "family": "overreach", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-01", "family": "contradiction_dependency", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": true, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-02", "family": "contradiction_dependency", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-03", "family": "contradiction_dependency", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-04", "family": "contradiction_dependency", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 1, "clarification_expected": true, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-05", "family": "contradiction_dependency", "arm": "A_NEMOTRON_ONLY", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-AMBIGUITY-01", "family": "ambiguity", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "agent-2e5e0aab0ff03c54", "status": "COMPLETED", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": true, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["CLARIFICATION_MISS"]}, {"task_id": "FS-AGENT-AMBIGUITY-02", "family": "ambiguity", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "agent-95148443b12a0cf4", "status": "COMPLETED", "task_success": true, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": true, "clarification_predicted": true, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-AMBIGUITY-03", "family": "ambiguity", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "agent-a229b34ab1e14987", "status": "COMPLETED", "task_success": true, "completion_truth": true, "claims_completion": true, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 1, "scope_total": 1, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-AMBIGUITY-04", "family": "ambiguity", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "agent-494764bffba94a58", "status": "COMPLETED", "task_success": true, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 1, "scope_total": 1, "clarification_expected": true, "clarification_predicted": true, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-AMBIGUITY-05", "family": "ambiguity", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "agent-90f77d2b9c72c7f5", "status": "COMPLETED", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["OVERBLOCK"]}, {"task_id": "FS-AGENT-SCOPE-01", "family": "scope", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "agent-477270d394d43529", "status": "COMPLETED", "task_success": true, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 3, "scope_total": 3, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-SCOPE-02", "family": "scope", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "agent-14c071b9ca07b88e", "status": "COMPLETED", "task_success": true, "completion_truth": true, "claims_completion": true, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-SCOPE-03", "family": "scope", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "agent-8b95c8a951c1420c", "status": "COMPLETED", "task_success": false, "completion_truth": false, "claims_completion": true, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["FALSE_COMPLETION"]}, {"task_id": "FS-AGENT-SCOPE-04", "family": "scope", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "agent-de427c623cb9d963", "status": "COMPLETED", "task_success": true, "completion_truth": true, "claims_completion": true, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 3, "scope_total": 3, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-SCOPE-05", "family": "scope", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "agent-58544fbdd79eecb0", "status": "COMPLETED", "task_success": true, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 3, "scope_total": 3, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-01", "family": "partial_completion", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-02", "family": "partial_completion", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-03", "family": "partial_completion", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-04", "family": "partial_completion", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-05", "family": "partial_completion", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-01", "family": "residual_defect", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-02", "family": "residual_defect", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-03", "family": "residual_defect", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-04", "family": "residual_defect", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-05", "family": "residual_defect", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-01", "family": "overreach", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 1, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-02", "family": "overreach", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-03", "family": "overreach", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-04", "family": "overreach", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-05", "family": "overreach", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-01", "family": "contradiction_dependency", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": true, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-02", "family": "contradiction_dependency", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-03", "family": "contradiction_dependency", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-04", "family": "contradiction_dependency", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 1, "clarification_expected": true, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-05", "family": "contradiction_dependency", "arm": "B_NEMOTRON_PLUS_FINESCHEMA_B", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-AMBIGUITY-01", "family": "ambiguity", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "agent-40cf85e0145a46d9", "status": "COMPLETED", "task_success": true, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": true, "clarification_predicted": true, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-AMBIGUITY-02", "family": "ambiguity", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "agent-error-a671829219f9975b", "status": "MODEL_FAILURE", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": true, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["MODEL_FAILURE"]}, {"task_id": "FS-AGENT-AMBIGUITY-03", "family": "ambiguity", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "agent-41c2c9d4a95ed488", "status": "COMPLETED", "task_success": true, "completion_truth": true, "claims_completion": true, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 1, "scope_total": 1, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-AMBIGUITY-04", "family": "ambiguity", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "agent-eb8cd651b5f9ee86", "status": "COMPLETED", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 1, "scope_total": 1, "clarification_expected": true, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["CLARIFICATION_MISS"]}, {"task_id": "FS-AGENT-AMBIGUITY-05", "family": "ambiguity", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "agent-d5fa62da21c91ae0", "status": "COMPLETED", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": false, "clarification_predicted": true, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["OVERBLOCK", "UNNECESSARY_CLARIFICATION"]}, {"task_id": "FS-AGENT-SCOPE-01", "family": "scope", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "agent-43f1295af25a9a13", "status": "COMPLETED", "task_success": true, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 3, "scope_total": 3, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-SCOPE-02", "family": "scope", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "agent-f7953092bf49dfd7", "status": "COMPLETED", "task_success": true, "completion_truth": true, "claims_completion": true, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-SCOPE-03", "family": "scope", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "agent-d5219dd7a47440c1", "status": "COMPLETED", "task_success": true, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-SCOPE-04", "family": "scope", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "agent-9954ae125ed0315b", "status": "COMPLETED", "task_success": true, "completion_truth": true, "claims_completion": true, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 3, "scope_total": 3, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-SCOPE-05", "family": "scope", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "agent-66aa6a4ccdf63f68", "status": "COMPLETED", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 3, "scope_total": 3, "clarification_expected": false, "clarification_predicted": true, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["UNNECESSARY_CLARIFICATION"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-01", "family": "partial_completion", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-02", "family": "partial_completion", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-03", "family": "partial_completion", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-04", "family": "partial_completion", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-05", "family": "partial_completion", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-01", "family": "residual_defect", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-02", "family": "residual_defect", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-03", "family": "residual_defect", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-04", "family": "residual_defect", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-05", "family": "residual_defect", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-01", "family": "overreach", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 1, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-02", "family": "overreach", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-03", "family": "overreach", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-04", "family": "overreach", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-05", "family": "overreach", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-01", "family": "contradiction_dependency", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": true, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-02", "family": "contradiction_dependency", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-03", "family": "contradiction_dependency", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-04", "family": "contradiction_dependency", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 1, "clarification_expected": true, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-05", "family": "contradiction_dependency", "arm": "C_NEMOTRON_PLUS_FINESCHEMA_A", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-AMBIGUITY-01", "family": "ambiguity", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "agent-cbee9192954524b1", "status": "COMPLETED", "task_success": true, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": true, "clarification_predicted": true, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-AMBIGUITY-02", "family": "ambiguity", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "agent-af9995daf9b05620", "status": "COMPLETED", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": true, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["CLARIFICATION_MISS"]}, {"task_id": "FS-AGENT-AMBIGUITY-03", "family": "ambiguity", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "agent-3a4669f6b88c8cc5", "status": "COMPLETED", "task_success": true, "completion_truth": true, "claims_completion": true, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 1, "scope_total": 1, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-AMBIGUITY-04", "family": "ambiguity", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "agent-0a673a575e2ad166", "status": "COMPLETED", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 1, "scope_total": 1, "clarification_expected": true, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["CLARIFICATION_MISS"]}, {"task_id": "FS-AGENT-AMBIGUITY-05", "family": "ambiguity", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "agent-c2f095c0359abd08", "status": "COMPLETED", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": false, "clarification_predicted": true, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["OVERBLOCK", "UNNECESSARY_CLARIFICATION"]}, {"task_id": "FS-AGENT-SCOPE-01", "family": "scope", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "agent-2513bec2a2be8b8a", "status": "COMPLETED", "task_success": true, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 3, "scope_total": 3, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-SCOPE-02", "family": "scope", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "agent-f7cb636e3de532a6", "status": "COMPLETED", "task_success": true, "completion_truth": true, "claims_completion": true, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-SCOPE-03", "family": "scope", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "agent-7a317bd01c9b6dcb", "status": "COMPLETED", "task_success": false, "completion_truth": false, "claims_completion": true, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 2, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["FALSE_COMPLETION"]}, {"task_id": "FS-AGENT-SCOPE-04", "family": "scope", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "agent-8638b1aa1570bad1", "status": "COMPLETED", "task_success": true, "completion_truth": true, "claims_completion": true, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 3, "scope_total": 3, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-SCOPE-05", "family": "scope", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "agent-2e0eeb7d35e0511f", "status": "COMPLETED", "task_success": true, "completion_truth": false, "claims_completion": false, "requirement_tp": 1, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 3, "scope_total": 3, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": []}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-01", "family": "partial_completion", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-02", "family": "partial_completion", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-03", "family": "partial_completion", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-04", "family": "partial_completion", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-PARTIAL-COMPLETION-05", "family": "partial_completion", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-01", "family": "residual_defect", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-02", "family": "residual_defect", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-03", "family": "residual_defect", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-04", "family": "residual_defect", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-RESIDUAL-DEFECT-05", "family": "residual_defect", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-01", "family": "overreach", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 1, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-02", "family": "overreach", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-03", "family": "overreach", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-04", "family": "overreach", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-OVERREACH-05", "family": "overreach", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-01", "family": "contradiction_dependency", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": true, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-02", "family": "contradiction_dependency", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-03", "family": "contradiction_dependency", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-04", "family": "contradiction_dependency", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": false, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 1, "clarification_expected": true, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}, {"task_id": "FS-AGENT-CONTRADICTION-DEPENDENCY-05", "family": "contradiction_dependency", "arm": "D_NEMOTRON_PLUS_FINESCHEMA_BA", "run_id": "NOT_RUN", "status": "NOT_RUN", "task_success": false, "completion_truth": true, "claims_completion": false, "requirement_tp": 0, "requirement_fp": 0, "requirement_total": 1, "scope_tp": 0, "scope_total": 2, "clarification_expected": false, "clarification_predicted": false, "overreach_count": 0, "unsupported_assumption_count": 0, "failures": ["NOT_RUN"]}], "report_hash": "sha256:4ed2e3bb0ac72d30e0de12d81366f412d515a0f7395c700013d6129c6fa96fdf", "attempted_runs": 40, "successful_runs": 39, "failed_runs": 1, "live_matrix": {"attempted_runs": 40, "expected_runs": 120, "failed_runs": 1, "missing_runs": 80, "status": "MEASURED_INTERNAL_PARTIAL", "successful_runs": 39}, "release_status": "PRELIMINARY_INTERNAL_GOLD_NOT_INDEPENDENT", "report_status": "MEASURED_INTERNAL_PARTIAL"}, "intent_resolution": {"version": "intent-bench-v1", "status": "INTERNAL_GOLD_NOT_INDEPENDENT", "cases": 62, "metrics": {"action_recall": 1.0, "action_precision": 1.0, "overreach_rate": 0.0, "quantifier_accuracy": 1.0, "risk_accuracy": 1.0, "release_ready_accuracy": 1.0, "clarification_accuracy": 1.0, "implicit_obligation_recall": 1.0, "unsupported_requirement_count": 0}, "limitation": "gold는 세션이 작성한 합성 기대값 — 사람 작성 골드 대비 recall은 여전히 미측정"}, "semantic_resolution": {"version": "semantic-v1-live-matrix", "status": "MEASURED_INTERNAL_PARTIAL", "cases": 30, "runs": 30, "expected_runs": 120, "gold": "INTERNAL_GOLD_NOT_INDEPENDENT", "metrics": {"S1_NEMOTRON_ANSWER_ONLY": {"answer_accuracy": {"numerator": 0, "denominator": 0, "value": "NOT_MEASURED"}, "entity_recall": "NOT_MEASURED", "entity_precision": "NOT_MEASURED", "relation_recall": "NOT_MEASURED", "relation_precision": "NOT_MEASURED", "relation_f1": "NOT_MEASURED", "condition_recall": "NOT_MEASURED", "exception_recall": "NOT_MEASURED", "causal_edge_f1": "NOT_MEASURED", "coreference_accuracy": "NOT_MEASURED", "unsupported_relation_rate": "NOT_MEASURED", "ambiguity_calibration": "NOT_MEASURED", "provenance_coverage": "NOT_MEASURED"}, "S2_NEMOTRON_STRUCTURED": {"answer_accuracy": {"numerator": 0, "denominator": 0, "value": "NOT_MEASURED"}, "entity_recall": {"numerator": 0, "denominator": 100, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "entity_precision": {"numerator": 0, "denominator": 0, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "relation_recall": {"numerator": 0, "denominator": 60, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "relation_precision": {"numerator": 0, "denominator": 0, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "relation_f1": {"tp": 0, "fp": 0, "fn": 60, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "condition_recall": {"numerator": 0, "denominator": 20, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "exception_recall": {"numerator": 0, "denominator": 10, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "causal_edge_f1": {"tp": 0, "fp": 0, "fn": 25, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "coreference_accuracy": "NOT_MEASURED", "unsupported_relation_rate": {"numerator": 0, "denominator": 0, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "ambiguity_calibration": {"numerator": 0, "denominator": 5, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "provenance_coverage": {"numerator": 0, "denominator": 0, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}}, "S3_FINESCHEMA_SEMANTIC_RESOLUTION": {"answer_accuracy": "NOT_MEASURED", "entity_recall": {"numerator": 95, "denominator": 100, "value": 0.95}, "entity_precision": {"numerator": 95, "denominator": 129, "value": 0.7364341085271318}, "relation_recall": {"numerator": 55, "denominator": 60, "value": 0.9166666666666666}, "relation_precision": {"numerator": 55, "denominator": 55, "value": 1.0}, "relation_f1": {"tp": 55, "fp": 0, "fn": 5, "value": 0.9565217391304348}, "condition_recall": {"numerator": 20, "denominator": 20, "value": 1.0}, "exception_recall": {"numerator": 10, "denominator": 10, "value": 1.0}, "causal_edge_f1": {"tp": 22, "fp": 0, "fn": 3, "value": 0.9361702127659575}, "coreference_accuracy": "NOT_MEASURED", "unsupported_relation_rate": {"numerator": 0, "denominator": 55, "value": 0.0}, "ambiguity_calibration": {"numerator": 5, "denominator": 5, "value": 1.0}, "provenance_coverage": {"numerator": 193, "denominator": 193, "value": 1.0}}, "S4_FULL_STRUCTURE_NEMOTRON_ANSWER": {"answer_accuracy": {"numerator": 0, "denominator": 0, "value": "NOT_MEASURED"}, "entity_recall": {"numerator": 0, "denominator": 100, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "entity_precision": {"numerator": 0, "denominator": 0, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "relation_recall": {"numerator": 0, "denominator": 60, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "relation_precision": {"numerator": 0, "denominator": 0, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "relation_f1": {"tp": 0, "fp": 0, "fn": 60, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "condition_recall": {"numerator": 0, "denominator": 20, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "exception_recall": {"numerator": 0, "denominator": 10, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "causal_edge_f1": {"tp": 0, "fp": 0, "fn": 25, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "coreference_accuracy": "NOT_MEASURED", "unsupported_relation_rate": {"numerator": 0, "denominator": 0, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "ambiguity_calibration": {"numerator": 0, "denominator": 5, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}, "provenance_coverage": {"numerator": 0, "denominator": 0, "value": "NOT_MEASURED", "status": "LIVE_BLOCKED_CREDENTIAL"}}}, "arms": {"S1_NEMOTRON_ANSWER_ONLY": {"status": "LIVE_BLOCKED_CREDENTIAL", "runs": 0}, "S2_NEMOTRON_STRUCTURED": {"status": "LIVE_BLOCKED_CREDENTIAL", "runs": 0}, "S3_FINESCHEMA_SEMANTIC_RESOLUTION": {"status": "MEASURED_INTERNAL", "runs": 30}, "S4_FULL_STRUCTURE_NEMOTRON_ANSWER": {"status": "LIVE_BLOCKED_CREDENTIAL", "runs": 0}}, "rows": [{"task_id": "FS-SEM-001", "family": "causal", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 3, "entity_fn": 0, "relation_tp": 1, "relation_fp": 0, "relation_fn": 0, "failures": ["ENTITY_OVERGENERATION"]}, {"task_id": "FS-SEM-006", "family": "conditional", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 0, "entity_fn": 0, "relation_tp": 2, "relation_fp": 0, "relation_fn": 0, "failures": []}, {"task_id": "FS-SEM-011", "family": "exception", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 0, "entity_fn": 0, "relation_tp": 3, "relation_fp": 0, "relation_fn": 0, "failures": []}, {"task_id": "FS-SEM-016", "family": "argument", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 0, "entity_fn": 0, "relation_tp": 2, "relation_fp": 0, "relation_fn": 0, "failures": []}, {"task_id": "FS-SEM-021", "family": "coreference", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 1, "entity_fn": 0, "relation_tp": 1, "relation_fp": 0, "relation_fn": 0, "failures": ["ENTITY_OVERGENERATION"]}, {"task_id": "FS-SEM-026", "family": "novel_system", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 5, "entity_fp": 3, "entity_fn": 0, "relation_tp": 3, "relation_fp": 0, "relation_fn": 0, "failures": ["ENTITY_OVERGENERATION"]}, {"task_id": "FS-SEM-002", "family": "causal", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 3, "entity_fn": 0, "relation_tp": 1, "relation_fp": 0, "relation_fn": 0, "failures": ["ENTITY_OVERGENERATION"]}, {"task_id": "FS-SEM-007", "family": "conditional", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 0, "entity_fn": 0, "relation_tp": 2, "relation_fp": 0, "relation_fn": 0, "failures": []}, {"task_id": "FS-SEM-012", "family": "exception", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 0, "entity_fn": 0, "relation_tp": 3, "relation_fp": 0, "relation_fn": 0, "failures": []}, {"task_id": "FS-SEM-017", "family": "argument", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 0, "entity_fn": 0, "relation_tp": 2, "relation_fp": 0, "relation_fn": 0, "failures": []}, {"task_id": "FS-SEM-022", "family": "coreference", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 1, "entity_fn": 0, "relation_tp": 1, "relation_fp": 0, "relation_fn": 0, "failures": ["ENTITY_OVERGENERATION"]}, {"task_id": "FS-SEM-027", "family": "novel_system", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 5, "entity_fp": 3, "entity_fn": 0, "relation_tp": 3, "relation_fp": 0, "relation_fn": 0, "failures": ["ENTITY_OVERGENERATION"]}, {"task_id": "FS-SEM-003", "family": "causal", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 2, "entity_fn": 0, "relation_tp": 1, "relation_fp": 0, "relation_fn": 0, "failures": ["ENTITY_OVERGENERATION"]}, {"task_id": "FS-SEM-008", "family": "conditional", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 2, "entity_fp": 0, "entity_fn": 1, "relation_tp": 1, "relation_fp": 0, "relation_fn": 1, "failures": ["ENTITY_BOUNDARY_ERROR", "RELATION_MISS", "BAD_CAUSAL_EDGE"]}, {"task_id": "FS-SEM-013", "family": "exception", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 2, "entity_fp": 0, "entity_fn": 1, "relation_tp": 2, "relation_fp": 0, "relation_fn": 1, "failures": ["ENTITY_BOUNDARY_ERROR", "RELATION_MISS", "BAD_CAUSAL_EDGE"]}, {"task_id": "FS-SEM-018", "family": "argument", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 2, "entity_fp": 0, "entity_fn": 1, "relation_tp": 0, "relation_fp": 0, "relation_fn": 2, "failures": ["ENTITY_BOUNDARY_ERROR", "RELATION_MISS"]}, {"task_id": "FS-SEM-023", "family": "coreference", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 2, "entity_fp": 1, "entity_fn": 1, "relation_tp": 1, "relation_fp": 0, "relation_fn": 0, "failures": ["ENTITY_OVERGENERATION", "ENTITY_BOUNDARY_ERROR"]}, {"task_id": "FS-SEM-028", "family": "novel_system", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 4, "entity_fp": 3, "entity_fn": 1, "relation_tp": 2, "relation_fp": 0, "relation_fn": 1, "failures": ["ENTITY_OVERGENERATION", "ENTITY_BOUNDARY_ERROR", "RELATION_MISS", "BAD_CAUSAL_EDGE"]}, {"task_id": "FS-SEM-004", "family": "causal", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 3, "entity_fn": 0, "relation_tp": 1, "relation_fp": 0, "relation_fn": 0, "failures": ["ENTITY_OVERGENERATION"]}, {"task_id": "FS-SEM-009", "family": "conditional", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 0, "entity_fn": 0, "relation_tp": 2, "relation_fp": 0, "relation_fn": 0, "failures": []}, {"task_id": "FS-SEM-014", "family": "exception", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 0, "entity_fn": 0, "relation_tp": 3, "relation_fp": 0, "relation_fn": 0, "failures": []}, {"task_id": "FS-SEM-019", "family": "argument", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 0, "entity_fn": 0, "relation_tp": 2, "relation_fp": 0, "relation_fn": 0, "failures": []}, {"task_id": "FS-SEM-024", "family": "coreference", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 1, "entity_fn": 0, "relation_tp": 1, "relation_fp": 0, "relation_fn": 0, "failures": ["ENTITY_OVERGENERATION"]}, {"task_id": "FS-SEM-029", "family": "novel_system", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 5, "entity_fp": 3, "entity_fn": 0, "relation_tp": 3, "relation_fp": 0, "relation_fn": 0, "failures": ["ENTITY_OVERGENERATION"]}, {"task_id": "FS-SEM-005", "family": "causal", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 3, "entity_fn": 0, "relation_tp": 1, "relation_fp": 0, "relation_fn": 0, "failures": ["ENTITY_OVERGENERATION"]}, {"task_id": "FS-SEM-010", "family": "conditional", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 0, "entity_fn": 0, "relation_tp": 2, "relation_fp": 0, "relation_fn": 0, "failures": []}, {"task_id": "FS-SEM-015", "family": "exception", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 0, "entity_fn": 0, "relation_tp": 3, "relation_fp": 0, "relation_fn": 0, "failures": []}, {"task_id": "FS-SEM-020", "family": "argument", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 0, "entity_fn": 0, "relation_tp": 2, "relation_fp": 0, "relation_fn": 0, "failures": []}, {"task_id": "FS-SEM-025", "family": "coreference", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 3, "entity_fp": 1, "entity_fn": 0, "relation_tp": 1, "relation_fp": 0, "relation_fn": 0, "failures": ["ENTITY_OVERGENERATION"]}, {"task_id": "FS-SEM-030", "family": "novel_system", "arm": "S3_FINESCHEMA_SEMANTIC_RESOLUTION", "status": "COMPLETED", "entity_tp": 5, "entity_fp": 3, "entity_fn": 0, "relation_tp": 3, "relation_fp": 0, "relation_fn": 0, "failures": ["ENTITY_OVERGENERATION"]}, {"task_id": "FS-SEM-001", "family": "causal", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-006", "family": "conditional", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-011", "family": "exception", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-016", "family": "argument", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-021", "family": "coreference", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-026", "family": "novel_system", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-002", "family": "causal", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-007", "family": "conditional", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-012", "family": "exception", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-017", "family": "argument", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-022", "family": "coreference", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-027", "family": "novel_system", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-003", "family": "causal", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-008", "family": "conditional", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-013", "family": "exception", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-018", "family": "argument", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-023", "family": "coreference", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-028", "family": "novel_system", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-004", "family": "causal", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-009", "family": "conditional", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-014", "family": "exception", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-019", "family": "argument", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-024", "family": "coreference", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-029", "family": "novel_system", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-005", "family": "causal", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-010", "family": "conditional", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-015", "family": "exception", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-020", "family": "argument", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-025", "family": "coreference", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-030", "family": "novel_system", "arm": "S1_NEMOTRON_ANSWER_ONLY", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-001", "family": "causal", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-006", "family": "conditional", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-011", "family": "exception", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-016", "family": "argument", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-021", "family": "coreference", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-026", "family": "novel_system", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-002", "family": "causal", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-007", "family": "conditional", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-012", "family": "exception", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-017", "family": "argument", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-022", "family": "coreference", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-027", "family": "novel_system", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-003", "family": "causal", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-008", "family": "conditional", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-013", "family": "exception", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-018", "family": "argument", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-023", "family": "coreference", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-028", "family": "novel_system", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-004", "family": "causal", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-009", "family": "conditional", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-014", "family": "exception", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-019", "family": "argument", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-024", "family": "coreference", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-029", "family": "novel_system", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-005", "family": "causal", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-010", "family": "conditional", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-015", "family": "exception", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-020", "family": "argument", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-025", "family": "coreference", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-030", "family": "novel_system", "arm": "S2_NEMOTRON_STRUCTURED", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-001", "family": "causal", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-006", "family": "conditional", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-011", "family": "exception", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-016", "family": "argument", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-021", "family": "coreference", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-026", "family": "novel_system", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-002", "family": "causal", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-007", "family": "conditional", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-012", "family": "exception", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-017", "family": "argument", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-022", "family": "coreference", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-027", "family": "novel_system", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-003", "family": "causal", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-008", "family": "conditional", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-013", "family": "exception", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-018", "family": "argument", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-023", "family": "coreference", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-028", "family": "novel_system", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-004", "family": "causal", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-009", "family": "conditional", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-014", "family": "exception", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-019", "family": "argument", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-024", "family": "coreference", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-029", "family": "novel_system", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-005", "family": "causal", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-010", "family": "conditional", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-015", "family": "exception", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-020", "family": "argument", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-025", "family": "coreference", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}, {"task_id": "FS-SEM-030", "family": "novel_system", "arm": "S4_FULL_STRUCTURE_NEMOTRON_ANSWER", "status": "LIVE_BLOCKED_CREDENTIAL", "failures": ["LIVE_BLOCKED_CREDENTIAL"]}], "report_hash": "sha256:add9caf3c36ca123a5307c99fa8632e7b36a58e7054e5f87d418c593ad4c295e", "attempted_runs": 30, "successful_runs": 30, "failed_runs": 0, "missing_runs": 90, "live_matrix": {"attempted_runs": 0, "expected_runs": 90, "failed_runs": 0, "missing_runs": 90, "pilot_report_hash": "sha256:9b93c6b8cfead6e293987c532a9d4172ad454d1e4debfcb7055b800cbab9a1a5", "reason": "Agent pilot failed closed; Semantic LIVE calls were not authorized.", "source_snapshot_note": "The immutable Semantic aggregate predates production credential activation; its arm statuses are retained only as source provenance.", "source_snapshot_statuses": ["LIVE_BLOCKED_CREDENTIAL"], "status": "BLOCKED_BY_AGENT_PILOT", "successful_runs": 0}, "release_status": "PRELIMINARY_INTERNAL_GOLD_NOT_INDEPENDENT", "report_status": "MEASURED_INTERNAL_PARTIAL"}, "gold_review": {"agent_v1": {"benchmark": "agent_v1", "gold_status": "INTERNAL_GOLD_NOT_INDEPENDENT", "human_reviewed": false, "reviewed_cases": 0, "total_cases": 30, "actions": {"APPROVE": 0, "EDIT": 0, "REJECT": 0}, "original_gold_hash": "WITHHELD_FROM_PUBLIC_RUNTIME", "effective_gold_hash": "WITHHELD_FROM_PUBLIC_RUNTIME", "gold_changed": false, "prior_results_status": "NOT_MEASURED"}, "semantic_v1": {"benchmark": "semantic_v1", "gold_status": "INTERNAL_GOLD_NOT_INDEPENDENT", "human_reviewed": false, "reviewed_cases": 0, "total_cases": 30, "actions": {"APPROVE": 0, "EDIT": 0, "REJECT": 0}, "original_gold_hash": "WITHHELD_FROM_PUBLIC_RUNTIME", "effective_gold_hash": "WITHHELD_FROM_PUBLIC_RUNTIME", "gold_changed": false, "prior_results_status": "NOT_MEASURED"}}, "evidence_class": {"label": "INTERNAL BENCHMARK / PRELIMINARY / NOT INDEPENDENTLY VALIDATED", "external_performance_claim": false}, "defect_recall": {"version": "gold-mutation-set-v1", "status": "INTERNAL_SYNTHETIC", "fixtures": 4, "detected": 35, "mutations": 37}}