{
  "entries": [
    {
      "overall_cited_correctness": 0.5379811722335331,
      "overall_cited_correctness_ci95": [
        0.504244,
        0.573097
      ],
      "overall_cited_correctness_micro": 0.5324945805132907,
      "per_type": {
        "aggregation": 0.28585022137653715,
        "single_hop": 0.534629404617254,
        "timeline": 0.4074074074074074,
        "dossier": 0.0,
        "unanswerable": 1.0,
        "false_premise": 1.0
      },
      "overall_uncited_correctness": 0.607002794868102,
      "overall_uncited_correctness_ci95": [
        0.569941,
        0.643288
      ],
      "overall_uncited_correctness_micro": 0.6019453583143412,
      "per_type_uncited": {
        "aggregation": 0.4506896115995591,
        "single_hop": 0.5965978128797084,
        "timeline": 0.48148148148148145,
        "dossier": 0.11324786324786325,
        "unanswerable": 1.0,
        "false_premise": 1.0
      },
      "citation_precision": 0.4849284095151864,
      "citation_recall": 0.2401242524771324,
      "retrieval": {
        "recall@5": 0.2504164204250172,
        "recall@20": 0.3211723154139726,
        "ndcg@10": 0.31493977508001825
      },
      "n_tasks": 1038,
      "judge_model": "gpt-5.5-2026-04-23",
      "judge_prompt_version": "v2",
      "judge_errors": 0,
      "premise_id_rate": 1.0,
      "premise_refused_n": 38,
      "tokens_total": 15348570,
      "tokens_per_task": 14786.7,
      "cost_usd_total": 54.9128,
      "cost_usd_per_task": 0.052903,
      "system_name": "agentic-sonnet-5 (reference)",
      "split": "full",
      "dataset_version": "v1.0"
    },
    {
      "overall_cited_correctness": 0.489914937072187,
      "overall_cited_correctness_ci95": [
        0.453786,
        0.532675
      ],
      "overall_cited_correctness_micro": 0.48839730703892553,
      "per_type": {
        "aggregation": 0.3129205629205629,
        "single_hop": 0.48602673147023084,
        "timeline": 0.2962962962962963,
        "dossier": 0.03174603174603175,
        "unanswerable": 0.8125,
        "false_premise": 1.0
      },
      "overall_uncited_correctness": 0.5412926037278725,
      "overall_uncited_correctness_ci95": [
        0.498013,
        0.58164
      ],
      "overall_uncited_correctness_micro": 0.5239564492454666,
      "per_type_uncited": {
        "aggregation": 0.3635933135933136,
        "single_hop": 0.5176184690157959,
        "timeline": 0.48148148148148145,
        "dossier": 0.072562358276644,
        "unanswerable": 0.8125,
        "false_premise": 1.0
      },
      "citation_precision": 0.4826618457300275,
      "citation_recall": 0.20792912132057378,
      "retrieval": {
        "recall@5": 0.27161833585958484,
        "recall@20": 0.5943520456421887,
        "ndcg@10": 0.40104580262011313
      },
      "n_tasks": 1038,
      "judge_model": "gpt-5.5-2026-04-23",
      "judge_prompt_version": "v2",
      "judge_errors": 0,
      "premise_id_rate": 0.0,
      "premise_refused_n": 38,
      "system_name": "hybrid (reference)",
      "split": "full",
      "dataset_version": "v1.0"
    },
    {
      "overall_cited_correctness": 0.48812783009323873,
      "overall_cited_correctness_ci95": [
        0.458655,
        0.519757
      ],
      "overall_cited_correctness_micro": 0.4603254862251342,
      "per_type": {
        "aggregation": 0.16953022253774133,
        "single_hop": 0.4629404617253949,
        "timeline": 0.2962962962962963,
        "dossier": 0.0,
        "unanswerable": 1.0,
        "false_premise": 1.0
      },
      "overall_uncited_correctness": 0.5977148368031501,
      "overall_uncited_correctness_ci95": [
        0.562424,
        0.634442
      ],
      "overall_uncited_correctness_micro": 0.6136945437469721,
      "per_type_uncited": {
        "aggregation": 0.48349292060481747,
        "single_hop": 0.6087484811664642,
        "timeline": 0.4444444444444444,
        "dossier": 0.0496031746031746,
        "unanswerable": 1.0,
        "false_premise": 1.0
      },
      "citation_precision": 0.3677817560864538,
      "citation_recall": 0.20557436442922716,
      "retrieval": {
        "recall@5": 0.2719555810668802,
        "recall@20": 0.35699685050841434,
        "ndcg@10": 0.3462384319696253
      },
      "n_tasks": 1038,
      "judge_model": "gpt-5.5-2026-04-23",
      "judge_prompt_version": "v2",
      "judge_errors": 0,
      "premise_id_rate": 1.0,
      "premise_refused_n": 38,
      "tokens_total": 20329715,
      "tokens_per_task": 19585.5,
      "cost_usd_total": 118.537,
      "cost_usd_per_task": 0.114197,
      "system_name": "agentic-opus-4-8 (reference)",
      "split": "full",
      "dataset_version": "v1.0"
    },
    {
      "overall_cited_correctness": 0.4834211295519011,
      "overall_cited_correctness_ci95": [
        0.44542,
        0.522442
      ],
      "overall_cited_correctness_micro": 0.475368241264195,
      "per_type": {
        "aggregation": 0.24713724713724716,
        "single_hop": 0.47630619684082626,
        "timeline": 0.3333333333333333,
        "dossier": 0.0,
        "unanswerable": 0.84375,
        "false_premise": 1.0
      },
      "overall_uncited_correctness": 0.5182429565061704,
      "overall_uncited_correctness_ci95": [
        0.479916,
        0.556624
      ],
      "overall_uncited_correctness_micro": 0.5123080569612362,
      "per_type_uncited": {
        "aggregation": 0.31104291104291104,
        "single_hop": 0.511543134872418,
        "timeline": 0.4074074074074074,
        "dossier": 0.03571428571428571,
        "unanswerable": 0.84375,
        "false_premise": 1.0
      },
      "citation_precision": 0.462448347107438,
      "citation_recall": 0.20913474121465642,
      "retrieval": {
        "recall@5": 0.27110883483610854,
        "recall@20": 0.5894933149531674,
        "ndcg@10": 0.4053735887641372
      },
      "n_tasks": 1038,
      "judge_model": "gpt-5.5-2026-04-23",
      "judge_prompt_version": "v2",
      "judge_errors": 0,
      "premise_id_rate": 0.0,
      "premise_refused_n": 38,
      "system_name": "bm25 (reference)",
      "split": "full",
      "dataset_version": "v1.0"
    },
    {
      "overall_cited_correctness": 0.43378096243182523,
      "overall_cited_correctness_ci95": [
        0.398354,
        0.467841
      ],
      "overall_cited_correctness_micro": 0.39318165355585494,
      "per_type": {
        "aggregation": 0.21732032784664362,
        "single_hop": 0.3827460510328068,
        "timeline": 0.18518518518518517,
        "dossier": 0.0,
        "unanswerable": 0.84375,
        "false_premise": 0.9736842105263158
      },
      "overall_uncited_correctness": 0.48146383967843565,
      "overall_uncited_correctness_ci95": [
        0.435265,
        0.525402
      ],
      "overall_uncited_correctness_micro": 0.42827508720420193,
      "per_type_uncited": {
        "aggregation": 0.27908208960840536,
        "single_hop": 0.41433778857837184,
        "timeline": 0.2962962962962963,
        "dossier": 0.0816326530612245,
        "unanswerable": 0.84375,
        "false_premise": 0.9736842105263158
      },
      "citation_precision": 0.40643939393939393,
      "citation_recall": 0.1603905187044457,
      "retrieval": {
        "recall@5": 0.21868990486596868,
        "recall@20": 0.5169732306951362,
        "ndcg@10": 0.33172381577777293
      },
      "n_tasks": 1038,
      "judge_model": "gpt-5.5-2026-04-23",
      "judge_prompt_version": "v2",
      "judge_errors": 0,
      "premise_id_rate": 0.0,
      "premise_refused_n": 37,
      "system_name": "dense (reference)",
      "split": "full",
      "dataset_version": "v1.0"
    },
    {
      "overall_cited_correctness": 0.328125,
      "overall_cited_correctness_ci95": [
        0.317708,
        0.333333
      ],
      "overall_cited_correctness_micro": 0.06647398843930635,
      "per_type": {
        "aggregation": 0.0,
        "single_hop": 0.0,
        "timeline": 0.0,
        "dossier": 0.0,
        "unanswerable": 0.96875,
        "false_premise": 1.0
      },
      "overall_uncited_correctness": 0.34576067073409117,
      "overall_uncited_correctness_ci95": [
        0.327549,
        0.366601
      ],
      "overall_uncited_correctness_micro": 0.08346331467718751,
      "per_type_uncited": {
        "aggregation": 0.014729014729014731,
        "single_hop": 0.01701093560145808,
        "timeline": 0.07407407407407407,
        "dossier": 0.0,
        "unanswerable": 0.96875,
        "false_premise": 1.0
      },
      "citation_precision": 0.0,
      "citation_recall": 0.0,
      "retrieval": {
        "recall@5": 0.0,
        "recall@20": 0.0,
        "ndcg@10": 0.0
      },
      "n_tasks": 1038,
      "judge_model": "gpt-5.5-2026-04-23",
      "judge_prompt_version": "v2",
      "judge_errors": 0,
      "premise_id_rate": 0.0,
      "premise_refused_n": 38,
      "system_name": "closed_book (reference)",
      "split": "full",
      "dataset_version": "v1.0"
    },
    {
      "overall_cited_correctness": 0.30372807017543857,
      "overall_cited_correctness_ci95": [
        0.280154,
        0.323739
      ],
      "overall_cited_correctness_micro": 0.06165703275529865,
      "per_type": {
        "aggregation": 0.0,
        "single_hop": 0.0,
        "timeline": 0.0,
        "dossier": 0.0,
        "unanswerable": 0.875,
        "false_premise": 0.9473684210526315
      },
      "overall_uncited_correctness": 0.34629198402327305,
      "overall_uncited_correctness_ci95": [
        0.31106,
        0.380829
      ],
      "overall_uncited_correctness_micro": 0.1048186238937684,
      "per_type_uncited": {
        "aggregation": 0.02524082524082524,
        "single_hop": 0.04495747266099635,
        "timeline": 0.18518518518518517,
        "dossier": 0.0,
        "unanswerable": 0.875,
        "false_premise": 0.9473684210526315
      },
      "citation_precision": 0.0,
      "citation_recall": 0.0,
      "retrieval": {
        "recall@5": 0.0,
        "recall@20": 0.0,
        "ndcg@10": 0.0
      },
      "n_tasks": 1038,
      "judge_model": "gpt-5.5-2026-04-23",
      "judge_prompt_version": "v2",
      "judge_errors": 0,
      "premise_id_rate": 0.0,
      "premise_refused_n": 36,
      "system_name": "parametric (reference)",
      "split": "full",
      "dataset_version": "v1.0"
    }
  ]
}