Watch
1
0
Fork
You've already forked Seyyed_arc
0
forked from hesabix/arc
Seyyed_arc/docs/ai-agent/evals/tool-scale/latest.json
2026-08-21 00:15:57 +00:00

240 lines
No EOL
6.9 KiB
JSON

{
"sizes": [
{
"size": 180,
"real_tools": 180,
"synthetic_tools": 0,
"keyword_linear_ms": 38.067,
"inverted_index_build_ms": 101.57,
"inverted_index_query_ms": 5.396,
"inverted_index_query_p50_ms": 5.358,
"inverted_index_query_p95_ms": 5.526,
"inverted_postings": 522,
"index_memory_bytes_est": 120048,
"semantic_linear_ms": 39.234,
"ranking_latency_ms": 38.067,
"semantic_search_latency_ms": 39.234,
"discovery_latency_ms": 77.301,
"schema_loading_latency_ms": 3.643,
"database_queries": 0,
"top15": [
"search_invoices",
"get_sales_report",
"get_invoice_details",
"delete_invoice",
"create_invoice"
],
"inverted_top15": [
"search_invoices",
"get_sales_report",
"delete_invoice",
"create_invoice",
"delete_memory_entry"
],
"production_ready": false
},
{
"size": 500,
"real_tools": 180,
"synthetic_tools": 320,
"keyword_linear_ms": 69.998,
"inverted_index_build_ms": 181.453,
"inverted_index_query_ms": 5.805,
"inverted_index_query_p50_ms": 5.798,
"inverted_index_query_p95_ms": 5.892,
"inverted_postings": 845,
"index_memory_bytes_est": 339200,
"semantic_linear_ms": 105.663,
"ranking_latency_ms": 69.998,
"semantic_search_latency_ms": 105.663,
"discovery_latency_ms": 175.661,
"schema_loading_latency_ms": 3.515,
"database_queries": 0,
"top15": [
"search_invoices",
"get_sales_report",
"get_invoice_details",
"delete_invoice",
"create_invoice"
],
"inverted_top15": [
"search_invoices",
"get_sales_report",
"delete_invoice",
"create_invoice",
"delete_memory_entry"
],
"production_ready": false
},
{
"size": 1200,
"real_tools": 180,
"synthetic_tools": 1020,
"keyword_linear_ms": 100.839,
"inverted_index_build_ms": 397.414,
"inverted_index_query_ms": 6.454,
"inverted_index_query_p50_ms": 6.44,
"inverted_index_query_p95_ms": 6.597,
"inverted_postings": 1545,
"index_memory_bytes_est": 818640,
"semantic_linear_ms": 246.336,
"ranking_latency_ms": 100.839,
"semantic_search_latency_ms": 246.336,
"discovery_latency_ms": 347.175,
"schema_loading_latency_ms": 3.619,
"database_queries": 0,
"top15": [
"search_invoices",
"get_sales_report",
"get_invoice_details",
"delete_invoice",
"create_invoice"
],
"inverted_top15": [
"search_invoices",
"get_sales_report",
"delete_invoice",
"create_invoice",
"delete_memory_entry"
],
"production_ready": false
},
{
"size": 5000,
"real_tools": 180,
"synthetic_tools": 4820,
"keyword_linear_ms": 334.283,
"inverted_index_build_ms": 1456.405,
"inverted_index_query_ms": 10.55,
"inverted_index_query_p50_ms": 10.626,
"inverted_index_query_p95_ms": 10.937,
"inverted_postings": 5345,
"index_memory_bytes_est": 3421376,
"semantic_linear_ms": 1053.128,
"ranking_latency_ms": 334.283,
"semantic_search_latency_ms": 1053.128,
"discovery_latency_ms": 1387.411,
"schema_loading_latency_ms": 3.603,
"database_queries": 0,
"top15": [
"search_invoices",
"get_sales_report",
"get_invoice_details",
"delete_invoice",
"create_invoice"
],
"inverted_top15": [
"search_invoices",
"get_sales_report",
"delete_invoice",
"create_invoice",
"delete_memory_entry"
],
"production_ready": false
},
{
"size": 10000,
"real_tools": 180,
"synthetic_tools": 9820,
"keyword_linear_ms": 635.31,
"inverted_index_build_ms": 3033.773,
"inverted_index_query_ms": 13.662,
"inverted_index_query_p50_ms": 13.602,
"inverted_index_query_p95_ms": 13.914,
"inverted_postings": 10345,
"index_memory_bytes_est": 6846112,
"semantic_linear_ms": 2131.595,
"ranking_latency_ms": 635.31,
"semantic_search_latency_ms": 2131.595,
"discovery_latency_ms": 2766.905,
"schema_loading_latency_ms": 5.317,
"database_queries": 0,
"top15": [
"search_invoices",
"get_sales_report",
"get_invoice_details",
"delete_invoice",
"create_invoice"
],
"inverted_top15": [
"search_invoices",
"get_sales_report",
"delete_invoice",
"create_invoice",
"delete_memory_entry"
],
"production_ready": false
}
],
"size_100000": {
"size": 100000,
"real_tools": 180,
"synthetic_tools": 99820,
"keyword_linear_ms": null,
"inverted_index_build_ms": 30796.249,
"inverted_index_query_ms": 122.269,
"inverted_index_query_p50_ms": 120.801,
"inverted_index_query_p95_ms": 126.76,
"inverted_postings": 100345,
"index_memory_bytes_est": 68490352,
"semantic_linear_ms": null,
"ranking_latency_ms": null,
"semantic_search_latency_ms": null,
"discovery_latency_ms": 0,
"schema_loading_latency_ms": null,
"database_queries": 0,
"top15": [],
"inverted_top15": [
"search_invoices",
"get_sales_report",
"delete_invoice",
"create_invoice",
"delete_memory_entry"
],
"production_ready": false
},
"stress_10k": [
{
"concurrency": 1,
"queries": 1,
"wall_ms": 18.536,
"p50_ms": 16.324,
"p95_ms": 16.324
},
{
"concurrency": 10,
"queries": 10,
"wall_ms": 227.407,
"p50_ms": 87.168,
"p95_ms": 138.001
},
{
"concurrency": 50,
"queries": 50,
"wall_ms": 1591.009,
"p50_ms": 1395.796,
"p95_ms": 1507.56
},
{
"concurrency": 100,
"queries": 100,
"wall_ms": 3166.629,
"p50_ms": 2766.092,
"p95_ms": 3006.562
}
],
"memory_current_kb": 15373.2,
"memory_peak_kb": 217806.9,
"note": "Synthetic names never enter production registry. Inverted index is derived posting lists. Linear semantic at 10k is not a production design. 100k skips semantic scan.",
"index_study": {
"A_inverted_keyword": "In-process inverted index is the Phase 10 candidate generator.",
"B_postgres_gin_trigram": "Optional later for fuzzy tokens; not required at 10k.",
"C_pgvector_ann": "Existing pgvector is for knowledge chunks, not tools. No new vector DB.",
"D_hnsw": "Preferred future ANN when a versioned neural embedding exists.",
"E_hybrid_inverted_plus_vector": "Merge is implemented behind VECTOR_CANDIDATE_RETRIEVAL=False.",
"phase_10_gate": "Do not ship 10k linear semantic. Keyword inverted index is the online path.",
"semantic_10k_ms": 2131.595,
"inverted_10k_query_ms": 13.662,
"keyword_10k_ms": 635.31
}
}