280 lines
9.2 KiB
JSON
280 lines
9.2 KiB
JSON
{
|
||
"run_id": "eval-LyUcB42oK9",
|
||
"timestamp": "2026-03-23T15:21:58.788Z",
|
||
"mode": "single-pass-strict",
|
||
"use_mock": true,
|
||
"prompt_version": "normalizer_v1_1_1",
|
||
"dataset": {
|
||
"source": "file",
|
||
"file": "normalizer_eval_v1_1_30cases.json"
|
||
},
|
||
"cases_total": 5,
|
||
"metrics": {
|
||
"schema_validation_pass_rate": 100,
|
||
"intent_class_accuracy": 40,
|
||
"route_hint_accuracy": 100,
|
||
"causal_flag_accuracy": 100,
|
||
"high_confidence_error_rate": 0
|
||
},
|
||
"baseline_metrics": {
|
||
"schema_validation_pass_rate": 100,
|
||
"intent_class_accuracy": 72.73,
|
||
"route_hint_accuracy": 90.91,
|
||
"causal_flag_accuracy": 81.82,
|
||
"high_confidence_error_rate": 9.09
|
||
},
|
||
"baseline_delta": {
|
||
"schema_validation_pass_rate": 0,
|
||
"intent_class_accuracy": -32.73,
|
||
"route_hint_accuracy": 9.09,
|
||
"causal_flag_accuracy": 18.18,
|
||
"high_confidence_error_rate": -9.09
|
||
},
|
||
"class_accuracy": {
|
||
"drilldown_explain": {
|
||
"total": 2,
|
||
"passed": 2,
|
||
"accuracy_percent": 100
|
||
},
|
||
"period_close_risk": {
|
||
"total": 2,
|
||
"passed": 0,
|
||
"accuracy_percent": 0
|
||
},
|
||
"anomaly_probe": {
|
||
"total": 1,
|
||
"passed": 0,
|
||
"accuracy_percent": 0
|
||
}
|
||
},
|
||
"budget": {
|
||
"requests_total": 0,
|
||
"retries_used": 0,
|
||
"guidance": {
|
||
"forensic_calls_max": 10,
|
||
"final_eval_calls_max": 30,
|
||
"target_total_calls_max": 40,
|
||
"hard_cap_calls_max": 45
|
||
}
|
||
},
|
||
"mismatches": [
|
||
{
|
||
"case_id": "V11-OT-003",
|
||
"expected_intent_class": "period_close_risk",
|
||
"actual_intent_class": "heavy_analytical",
|
||
"expected_route_hint": "batch_refresh_then_store",
|
||
"actual_route_hint": "batch_refresh_then_store",
|
||
"expected_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false
|
||
},
|
||
"actual_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false,
|
||
"needs_exact_object_trace": false,
|
||
"needs_ranking": true,
|
||
"needs_anomaly_summary": false,
|
||
"needs_runtime_truth": false,
|
||
"needs_period_cut": true,
|
||
"needs_evidence": false
|
||
},
|
||
"comment": "Route chosen correctly, but intent_class drifted into a neighboring taxonomy bucket.",
|
||
"trace_id": "LyziNYqDRnmIav"
|
||
},
|
||
{
|
||
"case_id": "V11-OT-004",
|
||
"expected_intent_class": "anomaly_probe",
|
||
"actual_intent_class": "rule_based_account_control",
|
||
"expected_route_hint": "store_feature_risk",
|
||
"actual_route_hint": "store_feature_risk",
|
||
"expected_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false
|
||
},
|
||
"actual_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false,
|
||
"needs_exact_object_trace": false,
|
||
"needs_ranking": false,
|
||
"needs_anomaly_summary": true,
|
||
"needs_runtime_truth": false,
|
||
"needs_period_cut": true,
|
||
"needs_evidence": false
|
||
},
|
||
"comment": "Route chosen correctly, but intent_class drifted into a neighboring taxonomy bucket.",
|
||
"trace_id": "mo09Q4j0hOVWgw"
|
||
},
|
||
{
|
||
"case_id": "V11-OT-005",
|
||
"expected_intent_class": "period_close_risk",
|
||
"actual_intent_class": "heavy_analytical",
|
||
"expected_route_hint": "batch_refresh_then_store",
|
||
"actual_route_hint": "batch_refresh_then_store",
|
||
"expected_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false
|
||
},
|
||
"actual_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false,
|
||
"needs_exact_object_trace": false,
|
||
"needs_ranking": true,
|
||
"needs_anomaly_summary": false,
|
||
"needs_runtime_truth": false,
|
||
"needs_period_cut": false,
|
||
"needs_evidence": false
|
||
},
|
||
"comment": "Route chosen correctly, but intent_class drifted into a neighboring taxonomy bucket.",
|
||
"trace_id": "zLXsDrsg2vR5Qs"
|
||
}
|
||
],
|
||
"bad_confidence_cases": [],
|
||
"results": [
|
||
{
|
||
"case_id": "NQ-008",
|
||
"raw_question": "Покажи по банку документ №TRX-88 и связанную проводку по 51.",
|
||
"validation_passed": true,
|
||
"intent_match": true,
|
||
"route_match": true,
|
||
"causal_flags_match": true,
|
||
"expected_intent_class": "drilldown_explain",
|
||
"actual_intent_class": "drilldown_explain",
|
||
"expected_route_hint": "live_mcp_drilldown",
|
||
"actual_route_hint": "live_mcp_drilldown",
|
||
"expected_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false
|
||
},
|
||
"actual_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false,
|
||
"needs_exact_object_trace": true,
|
||
"needs_ranking": false,
|
||
"needs_anomaly_summary": false,
|
||
"needs_runtime_truth": true,
|
||
"needs_period_cut": false,
|
||
"needs_evidence": true
|
||
},
|
||
"confidence_overall": "low",
|
||
"trace_id": "rnUogTOSYvvR0g",
|
||
"request_count_for_case": 0
|
||
},
|
||
{
|
||
"case_id": "V11-DD-005",
|
||
"raw_question": "Покажи карточку конкретной операции DOC-7781 и связанную проводку.",
|
||
"validation_passed": true,
|
||
"intent_match": true,
|
||
"route_match": true,
|
||
"causal_flags_match": true,
|
||
"expected_intent_class": "drilldown_explain",
|
||
"actual_intent_class": "drilldown_explain",
|
||
"expected_route_hint": "live_mcp_drilldown",
|
||
"actual_route_hint": "live_mcp_drilldown",
|
||
"expected_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false
|
||
},
|
||
"actual_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false,
|
||
"needs_exact_object_trace": true,
|
||
"needs_ranking": false,
|
||
"needs_anomaly_summary": false,
|
||
"needs_runtime_truth": true,
|
||
"needs_period_cut": false,
|
||
"needs_evidence": true
|
||
},
|
||
"confidence_overall": "low",
|
||
"trace_id": "LSKGq9-z1wUR10",
|
||
"request_count_for_case": 0
|
||
},
|
||
{
|
||
"case_id": "V11-OT-003",
|
||
"raw_question": "Перед закрытием периода что у нас может взорваться в последний день?",
|
||
"validation_passed": true,
|
||
"intent_match": false,
|
||
"route_match": true,
|
||
"causal_flags_match": true,
|
||
"expected_intent_class": "period_close_risk",
|
||
"actual_intent_class": "heavy_analytical",
|
||
"expected_route_hint": "batch_refresh_then_store",
|
||
"actual_route_hint": "batch_refresh_then_store",
|
||
"expected_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false
|
||
},
|
||
"actual_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false,
|
||
"needs_exact_object_trace": false,
|
||
"needs_ranking": true,
|
||
"needs_anomaly_summary": false,
|
||
"needs_runtime_truth": false,
|
||
"needs_period_cut": true,
|
||
"needs_evidence": false
|
||
},
|
||
"confidence_overall": "medium",
|
||
"trace_id": "LyziNYqDRnmIav",
|
||
"request_count_for_case": 0
|
||
},
|
||
{
|
||
"case_id": "V11-OT-004",
|
||
"raw_question": "Где по июню выглядит подозрительно, но без точечного документа, просто дай зоны риска.",
|
||
"validation_passed": true,
|
||
"intent_match": false,
|
||
"route_match": true,
|
||
"causal_flags_match": true,
|
||
"expected_intent_class": "anomaly_probe",
|
||
"actual_intent_class": "rule_based_account_control",
|
||
"expected_route_hint": "store_feature_risk",
|
||
"actual_route_hint": "store_feature_risk",
|
||
"expected_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false
|
||
},
|
||
"actual_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false,
|
||
"needs_exact_object_trace": false,
|
||
"needs_ranking": false,
|
||
"needs_anomaly_summary": true,
|
||
"needs_runtime_truth": false,
|
||
"needs_period_cut": true,
|
||
"needs_evidence": false
|
||
},
|
||
"confidence_overall": "medium",
|
||
"trace_id": "mo09Q4j0hOVWgw",
|
||
"request_count_for_case": 0
|
||
},
|
||
{
|
||
"case_id": "V11-OT-005",
|
||
"raw_question": "Че-то все криво на предзакрытии, где самые опасные места?",
|
||
"validation_passed": true,
|
||
"intent_match": false,
|
||
"route_match": true,
|
||
"causal_flags_match": true,
|
||
"expected_intent_class": "period_close_risk",
|
||
"actual_intent_class": "heavy_analytical",
|
||
"expected_route_hint": "batch_refresh_then_store",
|
||
"actual_route_hint": "batch_refresh_then_store",
|
||
"expected_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false
|
||
},
|
||
"actual_requires": {
|
||
"needs_cross_entity_join": false,
|
||
"needs_causal_chain": false,
|
||
"needs_exact_object_trace": false,
|
||
"needs_ranking": true,
|
||
"needs_anomaly_summary": false,
|
||
"needs_runtime_truth": false,
|
||
"needs_period_cut": false,
|
||
"needs_evidence": false
|
||
},
|
||
"confidence_overall": "low",
|
||
"trace_id": "zLXsDrsg2vR5Qs",
|
||
"request_count_for_case": 0
|
||
}
|
||
]
|
||
}
|