arc/docs/ai-agent/evals/phase-12/latest.json
2026-08-21 00:15:57 +00:00

323 lines
No EOL
7 KiB
JSON

{
"canary": {
"INDEXED_CANDIDATE_RETRIEVAL": false,
"INDEXED_CANDIDATE_ROLLOUT_PERCENT": 5,
"INDEXED_CANDIDATE_FALLBACK": true,
"INDEXED_CANDIDATE_KILL_SWITCH": false,
"min_live_samples": 500
},
"other_planes": {
"ADAPTIVE_DISCOVERY_K": false,
"ADAPTIVE_K_ROLLOUT_PERCENT": 0,
"HYBRID_TOOL_DISCOVERY": false,
"PROGRESSIVE_SCHEMA_LOADING": false
},
"gold": {
"candidate_recall": [
{
"n": 20,
"queries": 443,
"candidate_recall": 0.9549,
"latency_p50_ms": 0.87,
"latency_p95_ms": 1.283,
"avg_latency_ms": 0.887
},
{
"n": 50,
"queries": 443,
"candidate_recall": 0.9955,
"latency_p50_ms": 0.862,
"latency_p95_ms": 1.84,
"avg_latency_ms": 0.918
},
{
"n": 100,
"queries": 443,
"candidate_recall": 1.0,
"latency_p50_ms": 0.849,
"latency_p95_ms": 1.302,
"avg_latency_ms": 0.867
},
{
"n": 200,
"queries": 443,
"candidate_recall": 1.0,
"latency_p50_ms": 0.837,
"latency_p95_ms": 1.328,
"avg_latency_ms": 0.866
}
],
"indexed_final": [
{
"k": 10,
"candidate_n": 100,
"recall": 0.9616,
"mrr": 0.7931,
"ndcg": 0.8091
},
{
"k": 15,
"candidate_n": 100,
"recall": 0.9819,
"mrr": 0.7945,
"ndcg": 0.8166
},
{
"k": 20,
"candidate_n": 100,
"recall": 0.9887,
"mrr": 0.7948,
"ndcg": 0.8191
},
{
"k": 48,
"candidate_n": 100,
"recall": 0.9887,
"mrr": 0.7947,
"ndcg": 0.8233
}
],
"baseline_final": [
{
"k": 10,
"recall": 0.9616
},
{
"k": 15,
"recall": 0.9819
},
{
"k": 20,
"recall": 0.9887
},
{
"k": 48,
"recall": 0.9887
}
],
"latency": [
{
"size": 180,
"control_p50_ms": 8.714,
"control_p95_ms": 9.093,
"indexed_p50_ms": 3.637,
"indexed_p95_ms": 4.337
},
{
"size": 1200,
"control_p50_ms": 20.429,
"control_p95_ms": 23.074,
"indexed_p50_ms": 4.306,
"indexed_p95_ms": 4.644
},
{
"size": 5000,
"control_p50_ms": 60.278,
"control_p95_ms": 61.425,
"indexed_p50_ms": 4.787,
"indexed_p95_ms": 5.393
},
{
"size": 10000,
"control_p50_ms": 117.374,
"control_p95_ms": 136.86,
"indexed_p50_ms": 6.805,
"indexed_p95_ms": 7.306
}
]
},
"live": {
"counters": {},
"control": {
"requests": 0,
"average_k": 0.0,
"p50_k": 0,
"p95_k": 0,
"avg_schema_tokens": 0,
"avg_latency_ms": 0.0
},
"experiment": {
"requests": 0,
"average_k": 0.0,
"p50_k": 0,
"p95_k": 0,
"avg_schema_tokens": 0,
"avg_latency_ms": 0.0
},
"indexed_control": {
"requests": 0,
"indexed_requests": 0,
"fallback_requests": 0,
"fallback_rate": null,
"avg_candidate_count": 0.0,
"discovery_latency": {
"p50": 0,
"p95": 0,
"p99": 0,
"avg_ms": 0.0
},
"request_latency": {
"p50": 0,
"p95": 0,
"p99": 0
},
"task_completion_rate": null,
"fallback_reasons": {},
"by_channel": {},
"by_mutation": {},
"high_risk_requests": 0
},
"indexed_experiment": {
"requests": 0,
"indexed_requests": 0,
"fallback_requests": 0,
"fallback_rate": null,
"avg_candidate_count": 0.0,
"discovery_latency": {
"p50": 0,
"p95": 0,
"p99": 0,
"avg_ms": 0.0
},
"request_latency": {
"p50": 0,
"p95": 0,
"p99": 0
},
"task_completion_rate": null,
"fallback_reasons": {},
"by_channel": {},
"by_mutation": {},
"high_risk_requests": 0
},
"k_histogram": {
"10": 0,
"15": 0,
"20": 0,
"fallback": 0
},
"by_channel": {},
"high_risk_requests": {},
"rediscovery_queued": 0,
"rediscovery_recovered": 0,
"rediscovery_recovery_rate": null,
"discovery_success_without_rediscovery": null,
"task_completion": null,
"live_requests": 0,
"min_live_samples": 500,
"decision": "HOLD",
"indexed_min_live_samples": 500,
"indexed_decision": "HOLD",
"candidate_index_hit": 0,
"candidate_index_empty": 0,
"indexed_candidate_fallback": 0,
"candidate_index_fallback": 0,
"indexed_retrieval_miss": 0,
"index_consistency_failure": 0,
"unauthorized_candidate": 0,
"unauthorized_rediscovery": 0,
"comparison": {
"task_completion": {
"control": {
"n": 0,
"rate": null,
"ci95": [
null,
null
]
},
"experiment": {
"n": 0,
"rate": null,
"ci95": [
null,
null
]
},
"delta": null,
"ci_overlap": null,
"enough_for_significance": false
},
"fallback_rate": {
"control": {
"n": 0,
"rate": null,
"ci95": [
null,
null
]
},
"experiment": {
"n": 0,
"rate": null,
"ci95": [
null,
null
]
},
"delta": null,
"ci_overlap": null,
"enough_for_significance": false
},
"indexed_retrieval_miss": {
"control": {
"n": 0,
"rate": null,
"ci95": [
null,
null
]
},
"experiment": {
"n": 0,
"rate": null,
"ci95": [
null,
null
]
},
"delta": null,
"ci_overlap": null,
"enough_for_significance": false
},
"tool_success": {
"control": {
"n": 0,
"rate": null,
"ci95": [
null,
null
]
},
"experiment": {
"n": 0,
"rate": null,
"ci95": [
null,
null
]
},
"delta": null,
"ci_overlap": null,
"enough_for_significance": false
}
},
"indexed_promotion": {
"decision": "HOLD",
"case": "insufficient_samples",
"reason": "need_500_per_arm",
"control_requests": 0,
"experiment_requests": 0,
"min_samples": 500
}
},
"promotion": {
"decision": "HOLD",
"case": "insufficient_samples",
"reason": "need_500_per_arm",
"control_requests": 0,
"experiment_requests": 0,
"min_samples": 500
},
"note": "Rollout is 5%. Promotion requires 500 live discovery requests per arm. This process snapshot is not production traffic."
}