Files
poc-grok-websearch/benchmarks/screening/results/screening-2026-10-15-v1/summary.json

636 lines
20 KiB
JSON

{
"retrospective": true,
"run_id": "screening-2026-10-15-v1",
"reference_date": "2026-10-15",
"method": "12 prompts idênticos x sete IDs; três por tema; critérios congelados antes das novas inferências; recuperações de rejeições compartilhadas de infraestrutura documentadas à parte",
"http_attempts": 91,
"infrastructure_rejections": 7,
"unknown_cost_infrastructure_attempts": 7,
"manifest": {
"dataset_sha256": "5487cc3c1a0458d8064ceced8b5bf6610905fad30f5deaa581c063566cb7cfad",
"prices_sha256": "8876c4caaf82a4541fe0f79626b778dd7f0980209655c4d679bb6197a7012257",
"rules_sha256": "26d3b045b29940d457137d7edd117c3fad022c01f85ee669d7b564cbfa5d9001",
"study_sha256": "161ff2f534695771e38ffa3d7cb971845c4b9cc143006d8b8caaaec78477e0c3",
"system_sha256": "8a6fa58dbc2539af95cf138848caf4ac704abb6ee738ff197df2859f19788b42",
"protocol_sha256": "91f14676799ed4c0bb802bb8fa61369613a905aa0680d29fee028009cb14d61f",
"models": [
"xai.grok-4.20-0309-non-reasoning",
"xai.grok-4.20-0309-reasoning",
"xai.grok-4.20-non-reasoning",
"xai.grok-4.20-reasoning",
"xai.grok-4.3",
"xai.grok-4.6",
"xai.grok-4.7"
],
"reference_date": "2026-10-15",
"concurrency": 2,
"max_output_tokens": 2400,
"timeout_seconds": 150,
"retries": 0,
"region": "us-chicago-1",
"endpoint": "/v1/responses",
"retrospective": true
},
"rules": {
"version": "v1",
"retrospective": true,
"weights": {
"cost": 0.5,
"search_reliability": 0.3,
"latency": 0.2
},
"category_weight": 0.25,
"cutoff_score": 70,
"minimum_confirmed_searches": 11,
"required_known_costs": 12,
"primary_score": "Por intenção: 100*(0.50*menor_custo/custo_modelo + 0.30*busca_confiável + 0.20*menor_latência/latência_modelo). Eficiência recebe zero quando a busca não é confiável. Média de três intenções por tema; cada tema pesa 25%.",
"reliability": "HTTP 200, custo conhecido, busca completed com fontes e citações, texto final não vazio e resposta não truncada; não é nota de correção factual.",
"normalization": "Melhor custo e melhor latência entre respostas confiáveis da mesma intenção; falhas não entram como melhores.",
"aliases": "Notas dos IDs da mesma configuração são promediadas; ambos devem satisfazer os gates. Selecionar duas configurações distintas com nota >=70, por ordem de nota. Não forçar dois aprovados.",
"ties": [
"menor custo médio por interação",
"menor mediana de latência",
"nome da configuração em ordem alfabética"
],
"sampling": "12 consultas: três por tema; seleção usa apenas hashes dos prompts, sem acesso aos resultados financeiros na função de seleção.",
"uncertainty": "Bootstrap estratificado e pareado de 2000 reamostragens, seed 20261015; intervalo descritivo de variabilidade de casos, não confiança populacional nem avaliação factual."
},
"total": {
"interactions": 84,
"known_costs": 84,
"reliable_searches": 84,
"total_usd": "9.45816885",
"mean_usd": "0.1125972482142857142857142857",
"input_usd": "4.57596475",
"cached_input_usd": "1.2390016",
"output_usd": "0.9532025",
"web_search_usd": "2.690",
"input_tokens": 5097015,
"cached_input_tokens": 2469760,
"output_tokens": 195286,
"reasoning_tokens": 132001,
"web_search_calls": 538,
"mean_total_tokens": "63003.58333333333333333333333",
"median_seconds": 26.227
},
"by_model": {
"xai.grok-4.20-0309-non-reasoning": {
"model": "xai.grok-4.20-0309-non-reasoning",
"configuration": "grok-4.20-non-reasoning",
"score": 74.01065453292512,
"by_category": {
"culinaria": 69.21498503484565,
"esportes": 72.89993143074679,
"clima": 76.60157508419303,
"entretenimento": 77.326126581915
},
"operational_gate_passed": true,
"passed": true,
"interactions": 12,
"known_costs": 12,
"reliable_searches": 12,
"total_usd": "0.87450885",
"mean_usd": "0.0728757375",
"input_usd": "0.38518125",
"cached_input_usd": "0.0536576",
"output_usd": "0.0306700",
"web_search_usd": "0.405",
"input_tokens": 576433,
"cached_input_tokens": 268288,
"output_tokens": 12268,
"reasoning_tokens": 0,
"web_search_calls": 81,
"mean_total_tokens": "49058.41666666666666666666667",
"median_seconds": 12.6395,
"case_bootstrap_95_range": [
70.93604846765469,
77.02627691399306
]
},
"xai.grok-4.20-0309-reasoning": {
"model": "xai.grok-4.20-0309-reasoning",
"configuration": "grok-4.20-reasoning",
"score": 67.926059113904,
"by_category": {
"culinaria": 66.939300973136,
"esportes": 68.27832931596575,
"clima": 63.521998133167465,
"entretenimento": 72.96460803334682
},
"operational_gate_passed": true,
"passed": false,
"interactions": 12,
"known_costs": 12,
"reliable_searches": 12,
"total_usd": "0.79318440",
"mean_usd": "0.06609870",
"input_usd": "0.37172750",
"cached_input_usd": "0.0377344",
"output_usd": "0.0837225",
"web_search_usd": "0.300",
"input_tokens": 486054,
"cached_input_tokens": 188672,
"output_tokens": 33489,
"reasoning_tokens": 24547,
"web_search_calls": 60,
"mean_total_tokens": "43295.25",
"median_seconds": 33.048500000000004,
"case_bootstrap_95_range": [
63.8727041625088,
71.66759604432121
]
},
"xai.grok-4.20-non-reasoning": {
"model": "xai.grok-4.20-non-reasoning",
"configuration": "grok-4.20-non-reasoning",
"score": 72.76146105075577,
"by_category": {
"culinaria": 67.86373397720189,
"esportes": 73.92056055882125,
"clima": 67.80423848945831,
"entretenimento": 81.45731117754164
},
"operational_gate_passed": true,
"passed": true,
"interactions": 12,
"known_costs": 12,
"reliable_searches": 12,
"total_usd": "0.90652565",
"mean_usd": "0.07554380416666666666666666667",
"input_usd": "0.43264375",
"cached_input_usd": "0.0505344",
"output_usd": "0.0283475",
"web_search_usd": "0.395",
"input_tokens": 598787,
"cached_input_tokens": 252672,
"output_tokens": 11339,
"reasoning_tokens": 0,
"web_search_calls": 79,
"mean_total_tokens": "50843.83333333333333333333333",
"median_seconds": 12.383,
"case_bootstrap_95_range": [
69.96543401558118,
76.11276600401955
]
},
"xai.grok-4.20-reasoning": {
"model": "xai.grok-4.20-reasoning",
"configuration": "grok-4.20-reasoning",
"score": 72.23809313842376,
"by_category": {
"culinaria": 63.908826140174874,
"esportes": 68.14352956483388,
"clima": 69.00759929665293,
"entretenimento": 87.89241755203336
},
"operational_gate_passed": true,
"passed": true,
"interactions": 12,
"known_costs": 12,
"reliable_searches": 12,
"total_usd": "0.72970235",
"mean_usd": "0.06080852916666666666666666667",
"input_usd": "0.34650875",
"cached_input_usd": "0.0276736",
"output_usd": "0.0755200",
"web_search_usd": "0.280",
"input_tokens": 415575,
"cached_input_tokens": 138368,
"output_tokens": 30208,
"reasoning_tokens": 22140,
"web_search_calls": 56,
"mean_total_tokens": "37148.58333333333333333333333",
"median_seconds": 29.1545,
"case_bootstrap_95_range": [
69.52862471644511,
75.30139296340401
]
},
"xai.grok-4.3": {
"model": "xai.grok-4.3",
"configuration": "grok-4.3",
"score": 97.14903660620999,
"by_category": {
"culinaria": 100.0,
"esportes": 99.2752401820327,
"clima": 98.54118555711648,
"entretenimento": 90.7797206856908
},
"operational_gate_passed": true,
"passed": true,
"interactions": 12,
"known_costs": 12,
"reliable_searches": 12,
"total_usd": "0.50883360",
"mean_usd": "0.04240280",
"input_usd": "0.27343750",
"cached_input_usd": "0.0065536",
"output_usd": "0.0338425",
"web_search_usd": "0.195",
"input_tokens": 251518,
"cached_input_tokens": 32768,
"output_tokens": 13537,
"reasoning_tokens": 8045,
"web_search_calls": 39,
"mean_total_tokens": "22087.91666666666666666666667",
"median_seconds": 11.370000000000001,
"case_bootstrap_95_range": [
95.68117500841532,
99.30834306882593
]
},
"xai.grok-4.6": {
"model": "xai.grok-4.6",
"configuration": "grok-4.6",
"score": 45.50858471800458,
"by_category": {
"culinaria": 43.367917278455145,
"esportes": 47.432918850592024,
"clima": 44.64205288452623,
"entretenimento": 46.59144985844491
},
"operational_gate_passed": true,
"passed": false,
"interactions": 12,
"known_costs": 12,
"reliable_searches": 12,
"total_usd": "2.007070",
"mean_usd": "0.1672558333333333333333333333",
"input_usd": "1.049170",
"cached_input_usd": "0.223872",
"output_usd": "0.254028",
"web_search_usd": "0.480",
"input_tokens": 972329,
"cached_input_tokens": 447744,
"output_tokens": 42338,
"reasoning_tokens": 35045,
"web_search_calls": 96,
"mean_total_tokens": "84555.58333333333333333333333",
"median_seconds": 65.825,
"case_bootstrap_95_range": [
43.91545712499182,
47.38907940711982
]
},
"xai.grok-4.7": {
"model": "xai.grok-4.7",
"configuration": "grok-4.7",
"score": 43.344240404465566,
"by_category": {
"culinaria": 43.532006898649726,
"esportes": 45.84946646230471,
"clima": 41.33823470020622,
"entretenimento": 42.657253556701605
},
"operational_gate_passed": true,
"passed": false,
"interactions": 12,
"known_costs": 12,
"reliable_searches": 12,
"total_usd": "3.638344",
"mean_usd": "0.3031953333333333333333333333",
"input_usd": "1.717296",
"cached_input_usd": "0.838976",
"output_usd": "0.447072",
"web_search_usd": "0.635",
"input_tokens": 1796319,
"cached_input_tokens": 1141248,
"output_tokens": 52107,
"reasoning_tokens": 42224,
"web_search_calls": 127,
"mean_total_tokens": "154035.5",
"median_seconds": 50.155,
"case_bootstrap_95_range": [
40.6144424835464,
46.40546441452118
]
}
},
"configurations": [
{
"configuration": "grok-4.3",
"members": [
"xai.grok-4.3"
],
"score": 97.14903660620999,
"operational_gate_passed": true,
"passed": true,
"interactions": 12,
"known_costs": 12,
"reliable_searches": 12,
"total_usd": "0.50883360",
"mean_usd": "0.04240280",
"input_usd": "0.27343750",
"cached_input_usd": "0.0065536",
"output_usd": "0.0338425",
"web_search_usd": "0.195",
"input_tokens": 251518,
"cached_input_tokens": 32768,
"output_tokens": 13537,
"reasoning_tokens": 8045,
"web_search_calls": 39,
"mean_total_tokens": "22087.91666666666666666666667",
"median_seconds": 11.370000000000001,
"case_bootstrap_95_range": [
95.68117500841532,
99.30834306882593
],
"bootstrap_top2_frequency": 1.0
},
{
"configuration": "grok-4.20-non-reasoning",
"members": [
"xai.grok-4.20-0309-non-reasoning",
"xai.grok-4.20-non-reasoning"
],
"score": 73.38605779184044,
"operational_gate_passed": true,
"passed": true,
"interactions": 24,
"known_costs": 24,
"reliable_searches": 24,
"total_usd": "1.78103450",
"mean_usd": "0.07420977083333333333333333333",
"input_usd": "0.81782500",
"cached_input_usd": "0.1041920",
"output_usd": "0.0590175",
"web_search_usd": "0.800",
"input_tokens": 1175220,
"cached_input_tokens": 520960,
"output_tokens": 23607,
"reasoning_tokens": 0,
"web_search_calls": 160,
"mean_total_tokens": "49951.125",
"median_seconds": 12.6395,
"case_bootstrap_95_range": [
70.7982235073648,
76.06598395249495
],
"bootstrap_top2_frequency": 0.9975
},
{
"configuration": "grok-4.20-reasoning",
"members": [
"xai.grok-4.20-0309-reasoning",
"xai.grok-4.20-reasoning"
],
"score": 70.08207612616388,
"operational_gate_passed": true,
"passed": true,
"interactions": 24,
"known_costs": 24,
"reliable_searches": 24,
"total_usd": "1.52288675",
"mean_usd": "0.06345361458333333333333333333",
"input_usd": "0.71823625",
"cached_input_usd": "0.0654080",
"output_usd": "0.1592425",
"web_search_usd": "0.580",
"input_tokens": 901629,
"cached_input_tokens": 327040,
"output_tokens": 63697,
"reasoning_tokens": 46687,
"web_search_calls": 116,
"mean_total_tokens": "40221.91666666666666666666667",
"median_seconds": 29.606,
"case_bootstrap_95_range": [
67.02618846363141,
73.23189119873403
],
"bootstrap_top2_frequency": 0.0
},
{
"configuration": "grok-4.6",
"members": [
"xai.grok-4.6"
],
"score": 45.50858471800458,
"operational_gate_passed": true,
"passed": false,
"interactions": 12,
"known_costs": 12,
"reliable_searches": 12,
"total_usd": "2.007070",
"mean_usd": "0.1672558333333333333333333333",
"input_usd": "1.049170",
"cached_input_usd": "0.223872",
"output_usd": "0.254028",
"web_search_usd": "0.480",
"input_tokens": 972329,
"cached_input_tokens": 447744,
"output_tokens": 42338,
"reasoning_tokens": 35045,
"web_search_calls": 96,
"mean_total_tokens": "84555.58333333333333333333333",
"median_seconds": 65.825,
"case_bootstrap_95_range": [
43.91545712499182,
47.38907940711982
],
"bootstrap_top2_frequency": 0.0
},
{
"configuration": "grok-4.7",
"members": [
"xai.grok-4.7"
],
"score": 43.344240404465566,
"operational_gate_passed": true,
"passed": false,
"interactions": 12,
"known_costs": 12,
"reliable_searches": 12,
"total_usd": "3.638344",
"mean_usd": "0.3031953333333333333333333333",
"input_usd": "1.717296",
"cached_input_usd": "0.838976",
"output_usd": "0.447072",
"web_search_usd": "0.635",
"input_tokens": 1796319,
"cached_input_tokens": 1141248,
"output_tokens": 52107,
"reasoning_tokens": 42224,
"web_search_calls": 127,
"mean_total_tokens": "154035.5",
"median_seconds": 50.155,
"case_bootstrap_95_range": [
40.6144424835464,
46.40546441452118
],
"bootstrap_top2_frequency": 0.0
}
],
"selected_configurations": [
"grok-4.3",
"grok-4.20-non-reasoning"
],
"original_selection_confirmed": true,
"cache_discount_sensitivity": {
"secondary": true,
"method": "Mesmos tokens e latências observados; entrada em cache reprecificada como entrada normal, sem novas chamadas. Não simula execução fisicamente sem cache.",
"selected_configurations": [
"grok-4.3"
],
"configurations": [
{
"configuration": "grok-4.3",
"members": [
"xai.grok-4.3"
],
"score": 97.04049195066246,
"operational_gate_passed": true,
"passed": true,
"interactions": 12,
"known_costs": 12,
"reliable_searches": 12,
"total_usd": "0.54324000",
"mean_usd": "0.04527000",
"input_usd": "0.31439750",
"cached_input_usd": "0",
"output_usd": "0.0338425",
"web_search_usd": "0.195",
"input_tokens": 251518,
"cached_input_tokens": 32768,
"output_tokens": 13537,
"reasoning_tokens": 8045,
"web_search_calls": 39,
"mean_total_tokens": "22087.91666666666666666666667",
"median_seconds": 11.370000000000001,
"case_bootstrap_95_range": [
95.20345769097861,
99.30834306882593
],
"bootstrap_top2_frequency": 1.0
},
{
"configuration": "grok-4.20-non-reasoning",
"members": [
"xai.grok-4.20-0309-non-reasoning",
"xai.grok-4.20-non-reasoning"
],
"score": 67.93084405713441,
"operational_gate_passed": true,
"passed": false,
"interactions": 24,
"known_costs": 24,
"reliable_searches": 24,
"total_usd": "2.32804250",
"mean_usd": "0.09700177083333333333333333333",
"input_usd": "1.46902500",
"cached_input_usd": "0",
"output_usd": "0.0590175",
"web_search_usd": "0.800",
"input_tokens": 1175220,
"cached_input_tokens": 520960,
"output_tokens": 23607,
"reasoning_tokens": 0,
"web_search_calls": 160,
"mean_total_tokens": "49951.125",
"median_seconds": 12.6395,
"case_bootstrap_95_range": [
65.13208173487911,
70.85715751322269
],
"bootstrap_top2_frequency": 0.0905
},
{
"configuration": "grok-4.20-reasoning",
"members": [
"xai.grok-4.20-0309-reasoning",
"xai.grok-4.20-reasoning"
],
"score": 65.5062227964728,
"operational_gate_passed": true,
"passed": false,
"interactions": 24,
"known_costs": 24,
"reliable_searches": 24,
"total_usd": "1.86627875",
"mean_usd": "0.07776161458333333333333333333",
"input_usd": "1.12703625",
"cached_input_usd": "0",
"output_usd": "0.1592425",
"web_search_usd": "0.580",
"input_tokens": 901629,
"cached_input_tokens": 327040,
"output_tokens": 63697,
"reasoning_tokens": 46687,
"web_search_calls": 116,
"mean_total_tokens": "40221.91666666666666666666667",
"median_seconds": 29.606,
"case_bootstrap_95_range": [
63.234031445180484,
67.79390479015198
],
"bootstrap_top2_frequency": 0.0
},
{
"configuration": "grok-4.6",
"members": [
"xai.grok-4.6"
],
"score": 43.10385037403032,
"operational_gate_passed": true,
"passed": false,
"interactions": 12,
"known_costs": 12,
"reliable_searches": 12,
"total_usd": "2.678686",
"mean_usd": "0.2232238333333333333333333333",
"input_usd": "1.944658",
"cached_input_usd": "0",
"output_usd": "0.254028",
"web_search_usd": "0.480",
"input_tokens": 972329,
"cached_input_tokens": 447744,
"output_tokens": 42338,
"reasoning_tokens": 35045,
"web_search_calls": 96,
"mean_total_tokens": "84555.58333333333333333333333",
"median_seconds": 65.825,
"case_bootstrap_95_range": [
41.369607075039134,
45.069965045507836
],
"bootstrap_top2_frequency": 0.0
},
{
"configuration": "grok-4.7",
"members": [
"xai.grok-4.7"
],
"score": 40.876305988100796,
"operational_gate_passed": true,
"passed": false,
"interactions": 12,
"known_costs": 12,
"reliable_searches": 12,
"total_usd": "6.155272",
"mean_usd": "0.5129393333333333333333333333",
"input_usd": "5.073200",
"cached_input_usd": "0",
"output_usd": "0.447072",
"web_search_usd": "0.635",
"input_tokens": 1796319,
"cached_input_tokens": 1141248,
"output_tokens": 52107,
"reasoning_tokens": 42224,
"web_search_calls": 127,
"mean_total_tokens": "154035.5",
"median_seconds": 50.155,
"case_bootstrap_95_range": [
38.39072454922838,
43.65534628382452
],
"bootstrap_top2_frequency": 0.0
}
]
},
"original_selection": [
"grok-4.3",
"grok-4.20-non-reasoning"
],
"conclusion": "A amostra confirma os dois candidatos originais segundo a regra operacional congelada.",
"limits": "Nota operacional relativa, sem pontuação factual. Auditoria retrospectiva; 12 casos não permitem inferir o resultado de 100 intenções nos demais modelos."
}