{
  "schema": "benchmark-website-readout.v1",
  "evidence_cutoff": "2026-11-09",
  "pricing_as_of": "2026-11-08",
  "metrics": {
    "groups": {
      "all": {
        "tasks": 100,
        "pairs": 300,
        "resolved": 268,
        "unresolved": 32,
        "correct_before": 268,
        "correct_after": 264,
        "regressions": 4,
        "regression_cases": [
          "longbench/multi_news/5021490f86fe35b2f55ffaa25a755dabfe084d68d55c55e5",
          "longbench/multi_news/cef9105b35619a56715bd8971c08f04183c9884ea5315029",
          "longbench/musique/d1ff18285ee3c6148bb4f8466a3d65d965432b694aba3ac7"
        ],
        "before_usd": 3.163555,
        "after_usd": 3.0072511,
        "compression_usd": 0.0844821,
        "net_saved_usd": 0.15630390000000016,
        "net_saved_fraction": 0.04940767585833028,
        "context_before": 986277,
        "context_after": 906785,
        "context_saved_fraction": 0.08059804699896678
      },
      "accepted": {
        "tasks": 43,
        "pairs": 129,
        "resolved": 109,
        "unresolved": 20,
        "correct_before": 109,
        "correct_after": 106,
        "regressions": 3,
        "regression_cases": [
          "longbench/multi_news/5021490f86fe35b2f55ffaa25a755dabfe084d68d55c55e5",
          "longbench/multi_news/cef9105b35619a56715bd8971c08f04183c9884ea5315029"
        ],
        "before_usd": 1.000075,
        "after_usd": 0.8435251,
        "compression_usd": 0.0844821,
        "net_saved_usd": 0.15654990000000002,
        "net_saved_fraction": 0.1565381596380272,
        "context_before": 290204,
        "context_after": 210780,
        "context_saved_fraction": 0.2736833399953137
      },
      "guard_rejected": {
        "tasks": 56,
        "pairs": 168,
        "resolved": 156,
        "unresolved": 12,
        "correct_before": 156,
        "correct_after": 155,
        "regressions": 1,
        "regression_cases": [
          "longbench/musique/d1ff18285ee3c6148bb4f8466a3d65d965432b694aba3ac7"
        ],
        "before_usd": 2.152049,
        "after_usd": 2.15226,
        "compression_usd": 0.0,
        "net_saved_usd": -0.00021100000000018326,
        "net_saved_fraction": -9.804609467534142e-05,
        "context_before": 692643,
        "context_after": 692575,
        "context_saved_fraction": 9.817467295558657e-05
      },
      "intentional_skip": {
        "tasks": 1,
        "pairs": 3,
        "resolved": 3,
        "unresolved": 0,
        "correct_before": 3,
        "correct_after": 3,
        "regressions": 0,
        "regression_cases": [],
        "before_usd": 0.011431,
        "after_usd": 0.011465999999999999,
        "compression_usd": 0.0,
        "net_saved_usd": -3.4999999999998574e-05,
        "net_saved_fraction": -0.0030618493570115923,
        "context_before": 3430,
        "context_after": 3430,
        "context_saved_fraction": 0.0
      }
    },
    "datasets": {
      "hotpotqa": {
        "tasks": 20,
        "pairs": 60,
        "resolved": 57,
        "unresolved": 3,
        "correct_before": 57,
        "correct_after": 57,
        "regressions": 0,
        "regression_cases": [],
        "before_usd": 0.76336,
        "after_usd": 0.7310928,
        "compression_usd": 0.015895799999999998,
        "net_saved_usd": 0.03226720000000005,
        "net_saved_fraction": 0.04226996436805708,
        "context_before": 251038,
        "context_after": 234872,
        "context_saved_fraction": 0.06439662521211931
      },
      "multi_news": {
        "tasks": 20,
        "pairs": 60,
        "resolved": 52,
        "unresolved": 8,
        "correct_before": 52,
        "correct_after": 49,
        "regressions": 3,
        "regression_cases": [
          "longbench/multi_news/5021490f86fe35b2f55ffaa25a755dabfe084d68d55c55e5",
          "longbench/multi_news/cef9105b35619a56715bd8971c08f04183c9884ea5315029"
        ],
        "before_usd": 0.25824199999999997,
        "after_usd": 0.2356643,
        "compression_usd": 0.0104733,
        "net_saved_usd": 0.02257769999999998,
        "net_saved_fraction": 0.08742845857761317,
        "context_before": 44823,
        "context_after": 34347,
        "context_saved_fraction": 0.23371929589719564
      },
      "musique": {
        "tasks": 20,
        "pairs": 60,
        "resolved": 56,
        "unresolved": 4,
        "correct_before": 56,
        "correct_after": 55,
        "regressions": 1,
        "regression_cases": [
          "longbench/musique/d1ff18285ee3c6148bb4f8466a3d65d965432b694aba3ac7"
        ],
        "before_usd": 0.944604,
        "after_usd": 0.9173688,
        "compression_usd": 0.0130698,
        "net_saved_usd": 0.027235200000000015,
        "net_saved_fraction": 0.028832399608724968,
        "context_before": 309466,
        "context_after": 296021,
        "context_saved_fraction": 0.04344580664758002
      },
      "qasper": {
        "tasks": 20,
        "pairs": 60,
        "resolved": 48,
        "unresolved": 12,
        "correct_before": 48,
        "correct_after": 48,
        "regressions": 0,
        "regression_cases": [],
        "before_usd": 0.306243,
        "after_usd": 0.29372159999999997,
        "compression_usd": 0.0068616,
        "net_saved_usd": 0.012521400000000016,
        "net_saved_fraction": 0.04088713864480176,
        "context_before": 96044,
        "context_after": 89638,
        "context_saved_fraction": 0.06669859647661491
      },
      "qmsum": {
        "tasks": 20,
        "pairs": 60,
        "resolved": 55,
        "unresolved": 5,
        "correct_before": 55,
        "correct_after": 55,
        "regressions": 0,
        "regression_cases": [],
        "before_usd": 0.891106,
        "after_usd": 0.8294036,
        "compression_usd": 0.0381816,
        "net_saved_usd": 0.061702399999999935,
        "net_saved_fraction": 0.06924249191454213,
        "context_before": 284906,
        "context_after": 251907,
        "context_saved_fraction": 0.11582416656721861
      }
    },
    "followups": [
      {
        "level": 0.5,
        "overall_cost_saved": 0.1351961520282482,
        "accepted_cost_saved": 0.1351961520282482,
        "vx_context_saved": 0.29949944382647387,
        "correct_before": 6,
        "correct_after": 2,
        "partial": 4,
        "unresolved": 0,
        "answers": 6,
        "accepted_cases": 2,
        "protected_cases": 0,
        "period": "initial targeted follow-up"
      },
      {
        "level": 0.45,
        "overall_cost_saved": 0.069349681652626,
        "accepted_cost_saved": 0.13525596725320288,
        "vx_context_saved": 0.267797552836485,
        "correct_before": 6,
        "correct_after": 4,
        "partial": 2,
        "unresolved": 0,
        "answers": 6,
        "accepted_cases": 1,
        "protected_cases": 1,
        "period": "initial targeted follow-up"
      },
      {
        "level": 0.3,
        "overall_cost_saved": 0.04096254900325291,
        "accepted_cost_saved": 0.07167614693530378,
        "vx_context_saved": 0.18437152391546163,
        "correct_before": 6,
        "correct_after": 6,
        "partial": 0,
        "unresolved": 0,
        "answers": 6,
        "accepted_cases": 1,
        "protected_cases": 1,
        "period": "initial targeted follow-up"
      },
      {
        "level": 0.25,
        "overall_cost_saved": 0.03923872708872134,
        "accepted_cost_saved": 0.06590511587232184,
        "vx_context_saved": 0.15322580645161288,
        "correct_before": 6,
        "correct_after": 6,
        "partial": 0,
        "unresolved": 0,
        "answers": 6,
        "accepted_cases": 1,
        "protected_cases": 1,
        "period": "later targeted follow-up"
      },
      {
        "level": 0.15,
        "overall_cost_saved": 0.010390567428150366,
        "accepted_cost_saved": 0.020506594558799218,
        "vx_context_saved": 0.09705228031145718,
        "correct_before": 6,
        "correct_after": 6,
        "partial": 0,
        "unresolved": 0,
        "answers": 6,
        "accepted_cases": 1,
        "protected_cases": 1,
        "period": "later targeted follow-up"
      }
    ],
    "preservation": {
      "total_inputs": 57,
      "by_reason": [
        {
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "inputs": 25,
          "decision": "guard_rejected"
        },
        {
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "inputs": 25,
          "decision": "guard_rejected"
        },
        {
          "recorded_reason": "output_rejected_integrity_constraint",
          "label": "Constraints",
          "inputs": 3,
          "decision": "guard_rejected"
        },
        {
          "recorded_reason": "output_rejected_integrity_url",
          "label": "Links",
          "inputs": 2,
          "decision": "guard_rejected"
        },
        {
          "recorded_reason": "llmlingua_skipped_exact_output_context",
          "label": "Exact-output policy",
          "inputs": 1,
          "decision": "intentional_skip"
        },
        {
          "recorded_reason": "output_rejected_integrity_inline_code",
          "label": "Code",
          "inputs": 1,
          "decision": "guard_rejected"
        }
      ],
      "inputs": [
        {
          "case_id": "longbench/hotpotqa/3f17f12cfde4c677f3af81362ce1df42f949bee1ceae40f4",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/hotpotqa/4636487a1855382398fe1e54284cb7dd3d7b8d178ac4a476",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/hotpotqa/4c34b33d7a545c4eba3848d54fc4568fdf186e849897aa6e",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/hotpotqa/6c64a81fa526cb96d0a191d9cbd85a7e917b5d4f03b316d7",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/hotpotqa/72a769d6d0fd1c5c90974e60ab97e88e6853b9115a7fb85e",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/hotpotqa/9729ff4cf6a13382c88896cf5a5b176d082cc1137c66b38b",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/hotpotqa/97b31d6d7a9137c220726ad51a6dca3e957e2c94ce5e65b8",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/hotpotqa/9a19f9594aa73716dd38bcbd5ff1c04e57a6649800570f80",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/hotpotqa/a18c0ece7c2ae88d2f9f3ffc5eaa13297a1b5d0125a172ad",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/hotpotqa/b785963089bd7f81ee294751c688af85bfe2863dba623373",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_url",
          "label": "Links",
          "description": "A URL or link failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/hotpotqa/c9ee71b5dc7f5a4538d7727c95f80c174d3c9819dfbeac5a",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/hotpotqa/cf16ba255ad4b7b0fdae56093ac7d5fb58487dc24f2659e2",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/hotpotqa/e0a5da9581af9a4a573423cbdcf207c5436dc35ff8a9d380",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/hotpotqa/ea58013717bdbe9f0bad1991ed03d1afe3208bf514d9a2cd",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/multi_news/277ce57c2b27852c85a4fd20556d56f1320a4c387e772c11",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/multi_news/9bf2345cf303a45d45bc2e3692f6277785f850e0be2e1ecd",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/multi_news/da9e45ebc6f33da54bcae73bb2a8237ead6889d37ae18dd8",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_url",
          "label": "Links",
          "description": "A URL or link failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/musique/037dbe8ccac208d33dfbd28d6fdc319f88b1c2c2bd4ba060",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/musique/0da908a928c375084638f1e0ab10c329679eaacae7e4ffe0",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/musique/16de423a865eb739d3ec248429096218820feb176cb1b5af",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/musique/2b182c3ce4bb7bb2c28fc1f86d7a76591488e7d2aef1bb35",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/musique/2e723619f6cb5b2fe9cffb3e419841f96bbfd79a13f07f18",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/musique/3770e097bb92f794adf6bd2a09ec1f40441a5d76820f71f7",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/musique/442e302cee202c213a085d6aa1caf87e3aebf6b954c690b0",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/musique/72265eb89876baa2a8ab23c14ad61073f5d9f26c471e1ff9",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/musique/7b7acfdf920adc600a78d5e07adbb139447a4f4c05cdeb97",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/musique/a3ba409a8a3c8bad40a1fa415b981a8227c7d21936a06f17",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/musique/a4c716e09434218a16bdfed465322e369606bd38f23d815a",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/musique/a6e0032c62ece331c56364ced53e1c4076c303e4ae287074",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/musique/d1ff18285ee3c6148bb4f8466a3d65d965432b694aba3ac7",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/musique/db14f1320cc3442cb2416f65c437cd0ee74449baf687602b",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/musique/e3d8c2e853ea5a5ce78f9285d01c0b9b6ba832bd1027699d",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/musique/ebd4b995b91cadca20e7b463a7eeb03130f2aa06a4ac30a8",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/musique/f11250d4b392a30f6d22566d5167dee1464544f525c2975c",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/qasper/03ac54754b9162f6632212ca481920db17fc953e73af3a1f",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/qasper/072d3de1a7122730a13a31db3eede4113af2d920814f0aaa",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/qasper/3ca65b23b3cb316653709b469b4a9b42b790e0350c76ae1a",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/qasper/677dad703988e8f26b322211803fad38ffe9e58d1725f51a",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/qasper/806667777a33e5aede1bd3a62cbb5562f99b2223f68b98fe",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/qasper/8595332098adaabcfd8ae199f754a9b06cdde08cdd4cc64a",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/qasper/bee3efb92996570d990ab04bfdf054835ef720b9dd1c6f5a",
          "decision": "intentional_skip",
          "recorded_reason": "llmlingua_skipped_exact_output_context",
          "label": "Exact-output policy",
          "description": "An exact-output phrase triggered a conservative policy skip; the model did not run."
        },
        {
          "case_id": "longbench/qasper/c624b6d8a5c2cbf6ad3c24de6e42d7b1b6504fe608ee3733",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/qasper/c8ddf1d2e1192893ee5bf9e0ffdeb6762a7a2f719d299c28",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/qasper/d022dfe02fb2a55b4baa40fe436f616aecd3f3ced1a58d7c",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_inline_code",
          "label": "Code",
          "description": "Inline or fenced code failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/qasper/e419f2bff9d2ab7c3b60b3250caccd2d9ae1285ec3e8e818",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/qasper/ea4db579b5224fe0fa413ee436a5f4f56777dfa3e25d99d7",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/qmsum/20306b2da31b6657b21ec6ea54742ac7e414bbe24c0aff96",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/qmsum/2cc7738467d6098ff04980af3f7cd9ef3610f97d63a8c211",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/qmsum/2e714029680f4f3e0e47e17fecc1114de8bd4fe86681f2c4",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/qmsum/56a700676625c22e773b7f7631ef99982ba56e2cb64b2586",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_constraint",
          "label": "Constraints",
          "description": "A requirement or critical clause failed exact preservation; retain its wording."
        },
        {
          "case_id": "longbench/qmsum/5770240f11b9eae7d4a9a8a8de1e70bd557229796d9324f0",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/qmsum/585f667a0717f95cc0c7398ab295818570d6c307c7ed1ade",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_protected_span",
          "label": "Protected content",
          "description": "Content covered by exact-preservation rules failed a check; use fallback text."
        },
        {
          "case_id": "longbench/qmsum/78f6c42056bd08025eeed815449c24526a051751d7c415fd",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_constraint",
          "label": "Constraints",
          "description": "A requirement or critical clause failed exact preservation; retain its wording."
        },
        {
          "case_id": "longbench/qmsum/901c8c1975e2bafb77640f8854a3bcefcde762e1db3b5db7",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_constraint",
          "label": "Constraints",
          "description": "A requirement or critical clause failed exact preservation; retain its wording."
        },
        {
          "case_id": "longbench/qmsum/9510dd17d2c5a97bbc0407a2bcf8332b7120244a7ff22863",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/qmsum/a81acf4510dd3154c2f73c46b86b077b281307fcabe0a4bd",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        },
        {
          "case_id": "longbench/qmsum/caab9dadbf5f025e1a6b63ccda2809c85999e3edd0f0cc47",
          "decision": "guard_rejected",
          "recorded_reason": "output_rejected_integrity_identifier",
          "label": "Identifiers",
          "description": "An identifier-like value failed exact preservation; retain the safe version."
        }
      ],
      "counting": "First recorded reason per input; guard fallbacks and explicit policy skips remain distinct. Rejected drafts are not retained, so these are not counts of proven answer errors prevented."
    },
    "structured_json": {
      "tasks": 10,
      "distinct_inputs": 9,
      "conditions": {
        "json_minified": {
          "tasks": 10,
          "pairs": 30,
          "resolved": 30,
          "unresolved": 0,
          "correct_before": 30,
          "correct_after": 30,
          "regressions": 0,
          "regression_cases": [],
          "before_usd": 0.179877,
          "after_usd": 0.128328,
          "compression_usd": 0.0,
          "net_saved_usd": 0.05154900000000001,
          "net_saved_fraction": 0.2865791624276589,
          "context_before": 54865,
          "context_after": 37024,
          "context_saved_fraction": 0.32517998724141073
        },
        "deterministic_only": {
          "tasks": 10,
          "pairs": 30,
          "resolved": 30,
          "unresolved": 0,
          "correct_before": 30,
          "correct_after": 30,
          "regressions": 0,
          "regression_cases": [],
          "before_usd": 0.179877,
          "after_usd": 0.1202753,
          "compression_usd": 0.0141993,
          "net_saved_usd": 0.05960170000000001,
          "net_saved_fraction": 0.3313469759891482,
          "context_before": 54865,
          "context_after": 30941,
          "context_saved_fraction": 0.4360521279504238
        },
        "full": {
          "tasks": 10,
          "pairs": 30,
          "resolved": 30,
          "unresolved": 0,
          "correct_before": 30,
          "correct_after": 30,
          "regressions": 0,
          "regression_cases": [],
          "before_usd": 0.179877,
          "after_usd": 0.1235753,
          "compression_usd": 0.0141993,
          "net_saved_usd": 0.05630170000000001,
          "net_saved_fraction": 0.3130011063115351,
          "context_before": 54865,
          "context_after": 30941,
          "context_saved_fraction": 0.4360521279504238
        }
      },
      "net_cost_saved_vs_minified": 0.03703556511439432,
      "context_saved_vs_minified": 0.16429883318928262,
      "official_correct_full": 30,
      "official_correct_by_condition": {
        "json_minified": 30,
        "deterministic_only": 30,
        "full": 30
      },
      "model_stage_skipped_inputs": 10,
      "basis": "Existing ten-task structured JSON cohort, revalidated separately. Not pooled with expanded long text. Structural savings remain included when the lossy model stage is skipped.",
      "roundtrip": {
        "status": "passed",
        "cases": 10,
        "representations_checked": 40,
        "transformed_representations_checked": 30,
        "decoder_source": "data/public-benchmarks/sources/toon/packages/toon/src/index.ts",
        "rows": [
          {
            "case_id": "toon/analytics/q75",
            "condition": "unchanged",
            "structure_and_values_preserved": true,
            "representation_sha256": "6b040cc29c38ec6079e2c99fe33c4711d1c6e214a61fabc2fdebdfd89ee0b879"
          },
          {
            "case_id": "toon/analytics/q75",
            "condition": "json_minified",
            "structure_and_values_preserved": true,
            "representation_sha256": "3d3f6065eb002d5ce9a7a4fa21705525504a2f39cad1b2be431aaa5a57589490"
          },
          {
            "case_id": "toon/analytics/q75",
            "condition": "deterministic_only",
            "structure_and_values_preserved": true,
            "representation_sha256": "ef9c60cf34458aa2d185b2992f917bd9d308d2440d19c700287c184d99a7fbc9"
          },
          {
            "case_id": "toon/analytics/q75",
            "condition": "full",
            "structure_and_values_preserved": true,
            "representation_sha256": "ef9c60cf34458aa2d185b2992f917bd9d308d2440d19c700287c184d99a7fbc9"
          },
          {
            "case_id": "toon/event-logs/q128",
            "condition": "unchanged",
            "structure_and_values_preserved": true,
            "representation_sha256": "a64c0567b4d5a7b81e3d9079d33f874be0029c2d019ae3370aa21b679094ea98"
          },
          {
            "case_id": "toon/event-logs/q128",
            "condition": "json_minified",
            "structure_and_values_preserved": true,
            "representation_sha256": "50f4306dc984415d0d5fd9bcd218220d81ee66f13801fa7e42009fcbcb67b8ac"
          },
          {
            "case_id": "toon/event-logs/q128",
            "condition": "deterministic_only",
            "structure_and_values_preserved": true,
            "representation_sha256": "50f4306dc984415d0d5fd9bcd218220d81ee66f13801fa7e42009fcbcb67b8ac"
          },
          {
            "case_id": "toon/event-logs/q128",
            "condition": "full",
            "structure_and_values_preserved": true,
            "representation_sha256": "50f4306dc984415d0d5fd9bcd218220d81ee66f13801fa7e42009fcbcb67b8ac"
          },
          {
            "case_id": "toon/github/q109",
            "condition": "unchanged",
            "structure_and_values_preserved": true,
            "representation_sha256": "6d26a496aecaea30e578b66c294a6665cc5ad22ba87b6d4e76380eadc4503468"
          },
          {
            "case_id": "toon/github/q109",
            "condition": "json_minified",
            "structure_and_values_preserved": true,
            "representation_sha256": "1835a5f3cf40810d777b07b8539e6a00fed7c200e3990666a1e60f5258b52ddf"
          },
          {
            "case_id": "toon/github/q109",
            "condition": "deterministic_only",
            "structure_and_values_preserved": true,
            "representation_sha256": "afe38d04ceaebf6c90995f4f0ff9a2511e8789b44981a12233d9b45386d4af89"
          },
          {
            "case_id": "toon/github/q109",
            "condition": "full",
            "structure_and_values_preserved": true,
            "representation_sha256": "afe38d04ceaebf6c90995f4f0ff9a2511e8789b44981a12233d9b45386d4af89"
          },
          {
            "case_id": "toon/keyed/q182",
            "condition": "unchanged",
            "structure_and_values_preserved": true,
            "representation_sha256": "3553397e0d2b0e72e1606560a951e757a5a7aca9c3f21b34aa95a5aa3f589e38"
          },
          {
            "case_id": "toon/keyed/q182",
            "condition": "json_minified",
            "structure_and_values_preserved": true,
            "representation_sha256": "dff6a4642c47ed7ebd95318aa69947fc3993ca6f266669aba8191f73bb149742"
          },
          {
            "case_id": "toon/keyed/q182",
            "condition": "deterministic_only",
            "structure_and_values_preserved": true,
            "representation_sha256": "dff6a4642c47ed7ebd95318aa69947fc3993ca6f266669aba8191f73bb149742"
          },
          {
            "case_id": "toon/keyed/q182",
            "condition": "full",
            "structure_and_values_preserved": true,
            "representation_sha256": "dff6a4642c47ed7ebd95318aa69947fc3993ca6f266669aba8191f73bb149742"
          },
          {
            "case_id": "toon/nested/q41",
            "condition": "unchanged",
            "structure_and_values_preserved": true,
            "representation_sha256": "64ab765be3cbb2d41c635a062301532b01986045269d4a0ef03d1917420b43b9"
          },
          {
            "case_id": "toon/nested/q41",
            "condition": "json_minified",
            "structure_and_values_preserved": true,
            "representation_sha256": "9073a3a9991fe71a969849a7af0c112af3dc42a31e51b4a33ae9be729a932598"
          },
          {
            "case_id": "toon/nested/q41",
            "condition": "deterministic_only",
            "structure_and_values_preserved": true,
            "representation_sha256": "9073a3a9991fe71a969849a7af0c112af3dc42a31e51b4a33ae9be729a932598"
          },
          {
            "case_id": "toon/nested/q41",
            "condition": "full",
            "structure_and_values_preserved": true,
            "representation_sha256": "9073a3a9991fe71a969849a7af0c112af3dc42a31e51b4a33ae9be729a932598"
          },
          {
            "case_id": "toon/nested-config/q153",
            "condition": "unchanged",
            "structure_and_values_preserved": true,
            "representation_sha256": "7f2b6bd130fd4756b79cf4b00a3a216f0f9ac5a960db9c4ec1e3afc82280556b"
          },
          {
            "case_id": "toon/nested-config/q153",
            "condition": "json_minified",
            "structure_and_values_preserved": true,
            "representation_sha256": "81b786a8567bd7458e4519e3175b3848188b38a95c5e140dd083288221ed47aa"
          },
          {
            "case_id": "toon/nested-config/q153",
            "condition": "deterministic_only",
            "structure_and_values_preserved": true,
            "representation_sha256": "81b786a8567bd7458e4519e3175b3848188b38a95c5e140dd083288221ed47aa"
          },
          {
            "case_id": "toon/nested-config/q153",
            "condition": "full",
            "structure_and_values_preserved": true,
            "representation_sha256": "81b786a8567bd7458e4519e3175b3848188b38a95c5e140dd083288221ed47aa"
          },
          {
            "case_id": "toon/nested-group/q202",
            "condition": "unchanged",
            "structure_and_values_preserved": true,
            "representation_sha256": "4292bb15ba061cfd5e20a94b67bebf64eb70b1eac1ade7c07163c5b569ba7c81"
          },
          {
            "case_id": "toon/nested-group/q202",
            "condition": "json_minified",
            "structure_and_values_preserved": true,
            "representation_sha256": "1d106e70014128e6b8f2b5d40dec9ae356a60ef69d14681fe6e874ed73aeab9e"
          },
          {
            "case_id": "toon/nested-group/q202",
            "condition": "deterministic_only",
            "structure_and_values_preserved": true,
            "representation_sha256": "1d106e70014128e6b8f2b5d40dec9ae356a60ef69d14681fe6e874ed73aeab9e"
          },
          {
            "case_id": "toon/nested-group/q202",
            "condition": "full",
            "structure_and_values_preserved": true,
            "representation_sha256": "1d106e70014128e6b8f2b5d40dec9ae356a60ef69d14681fe6e874ed73aeab9e"
          },
          {
            "case_id": "toon/structural-validation-control/q240",
            "condition": "unchanged",
            "structure_and_values_preserved": true,
            "representation_sha256": "5af278dc412df5612829e5f8bf1713958bd12ea8557f96816ab782eb356a4e3c"
          },
          {
            "case_id": "toon/structural-validation-control/q240",
            "condition": "json_minified",
            "structure_and_values_preserved": true,
            "representation_sha256": "21e621be37f65c02e73ce54d977e15466c4da31b5303708e69457f28f7a4b944"
          },
          {
            "case_id": "toon/structural-validation-control/q240",
            "condition": "deterministic_only",
            "structure_and_values_preserved": true,
            "representation_sha256": "44a5d477f6581b70aac6dc9130f0c89c73a16933475429eb5a796a232d1cd912"
          },
          {
            "case_id": "toon/structural-validation-control/q240",
            "condition": "full",
            "structure_and_values_preserved": true,
            "representation_sha256": "44a5d477f6581b70aac6dc9130f0c89c73a16933475429eb5a796a232d1cd912"
          },
          {
            "case_id": "toon/tabular/q217",
            "condition": "unchanged",
            "structure_and_values_preserved": true,
            "representation_sha256": "5a72c52f524742b5cd66d34f11a21bf3e0d86a4720b7a59c7df3aa012ba52f18"
          },
          {
            "case_id": "toon/tabular/q217",
            "condition": "json_minified",
            "structure_and_values_preserved": true,
            "representation_sha256": "a6481d4bcb01f91d3beb1cdccd792f10f231fe32c94491701f9479ad268b7dd7"
          },
          {
            "case_id": "toon/tabular/q217",
            "condition": "deterministic_only",
            "structure_and_values_preserved": true,
            "representation_sha256": "50c446b40550a6fe6b2fabd1fec8b941989a8dac4266bdcfa9507eca2a533d01"
          },
          {
            "case_id": "toon/tabular/q217",
            "condition": "full",
            "structure_and_values_preserved": true,
            "representation_sha256": "50c446b40550a6fe6b2fabd1fec8b941989a8dac4266bdcfa9507eca2a533d01"
          },
          {
            "case_id": "toon/analytics/q77",
            "condition": "unchanged",
            "structure_and_values_preserved": true,
            "representation_sha256": "6b040cc29c38ec6079e2c99fe33c4711d1c6e214a61fabc2fdebdfd89ee0b879"
          },
          {
            "case_id": "toon/analytics/q77",
            "condition": "json_minified",
            "structure_and_values_preserved": true,
            "representation_sha256": "3d3f6065eb002d5ce9a7a4fa21705525504a2f39cad1b2be431aaa5a57589490"
          },
          {
            "case_id": "toon/analytics/q77",
            "condition": "deterministic_only",
            "structure_and_values_preserved": true,
            "representation_sha256": "ef9c60cf34458aa2d185b2992f917bd9d308d2440d19c700287c184d99a7fbc9"
          },
          {
            "case_id": "toon/analytics/q77",
            "condition": "full",
            "structure_and_values_preserved": true,
            "representation_sha256": "ef9c60cf34458aa2d185b2992f917bd9d308d2440d19c700287c184d99a7fbc9"
          }
        ]
      },
      "revalidation": {
        "review_calls": 22,
        "mapped_results": 120,
        "unique_answer_reviews": 10,
        "validation_status_counts": {
          "agreed_correct": 120
        },
        "complete": true
      }
    }
  },
  "source_sha256": {
    "outputs\\longtext-expanded-benchmark-20261108\\report.json": "5222740f9a8c765143cf300f3c7331f01ab2c647a55b5a8e179251641c560607",
    "outputs\\longtext-expanded-benchmark-20261108\\all-results.zip": "bf3cb53ece38bfccc0cd1e7b383c0d4b8c45609c52898f4bfef4c810126efcb5",
    "outputs\\terra-full-benchmark-20261108-195352\\report.json": "cf2eab0dc2bfb8dfd7ff98e33b941e33d47a6c0f76821efd59622b1b78d5c160",
    "outputs\\terra-full-benchmark-20261108-195352\\all-results.zip": "e8ce4100632b3b304c3895f83e7464cd887aeece8f346cd453db28c26590101c",
    "data\\public-benchmarks\\structured-json-inclusion-20261109\\review\\semantic-review.json": "39d27e8540bb57234205adfa1722e74b2dc4970ce7ed0a726dec2d7cea4fe163",
    "data\\public-benchmarks\\structured-json-inclusion-20261109\\roundtrip.json": "cafd6d769eed55aa453aaba4d37c001aaba705c8b8dd0651b9920e9f2665a2bd",
    "outputs\\longtext-aggressiveness-025-20261109\\detailed-report.json": "9c7e8a4f4401103c5be9e869693f9ec36df5b6b758c6b908d17e249735d9b363",
    "outputs\\longtext-aggressiveness-025-20261109\\results.json": "6670897e6f75ec2bde08fff8d1637dc6187d855c470d8baf3ed4e7c6a4134804",
    "outputs\\longtext-aggressiveness-025-20261109\\all-results.zip": "c384e9d57860bb158e53f74f43ce2c4dd3b62f97a0c36008e292d22114d9ecc6",
    "outputs\\longtext-aggressiveness-015-20261109\\detailed-report.json": "8e8a824ffa823f62252d55a0bee115760a97b4581aaff6b181c1a1f98867c689",
    "outputs\\longtext-aggressiveness-015-20261109\\results.json": "afa0db3b34aff177ffa698d8cb660c355be75c27d1ede7142b29c382218b02f0",
    "outputs\\longtext-aggressiveness-015-20261109\\all-results.zip": "a5cffd888fe87af99cb26cf066b64bf4d7df0662b9209776a1427b02b6163240"
  },
  "policy": "JSON and long text have separate cohort denominators and savings. Structural savings remain included when the JSON model stage is skipped. Long-text savings include all accepted inputs, including answer regressions. Unresolved pairs stay explicit. Source damage is secondary. Follow-ups are exploratory, never pooled into the main cohort."
}