{
  "id": "tier0-v2",
  "title": "AIO Tier 0 reference distributions — methodology v2-draft",
  "methodologyVersion": "v2-draft",
  "generatedAt": "2026-08-15T15:23:19.003Z",
  "license": "CC BY 4.0",
  "attribution": "AIO — AI Integrity Organization, https://aioq.org, CC BY 4.0",
  "what": [
    "For each pack this file records what a small panel of models actually scored under the deployed Tier 0 v2-draft configuration. Every pack carries the unadapted arm (the model was told nothing about the pack); eu-ai-act and kr-ai-framework-act also carry an adapted arm (the model was given the pack and its management guide first).",
    "It is a descriptive reference, not a threshold. A score report places an attempt inside or outside the unadapted range of this panel and says nothing further; no band is a pass and no band is a failure.",
    "The panel is five models measured once per cell on 2026-08-15. It is small, it is not a random sample of anything, and the run-to-run spread on a single configuration is large enough (mean |delta| 0.0649 across eight repeat pairs) that a difference smaller than roughly 0.16 should not be read as a difference at all.",
    "The adapted arm is included where it was measured (eu-ai-act, kr-ai-framework-act) so an operator can run the same comparison on their own model: measure once unadapted, once after supplying the pack, and report the delta. That within-model delta is a more informative quantity than either score alone, because it does not require the two models being compared to be equally capable.",
    "Packs measured in the unadapted arm only carry no adapted block and no delta. The band a score report reports is unaffected: it is computed from the unadapted range and nothing else."
  ],
  "source": {
    "campaign": "calibration v2 + reference batches 1 and 2 + eu-ai-act epoch-2 re-measurement",
    "file": ".bank-work/calibration/calibration-runs-v2.json",
    "sha256": "6163880fc78bea1f43ba44a06569aa20714d1befa840740c315b369f062cbac3",
    "report": ".bank-work/calibration/report-v2.md",
    "runCount": 24,
    "note": "Every run in every batch carries calibration: true and certificateIssued: false — the submit endpoint was never called, so none of these numbers came from an issued record.",
    "batches": [
      {
        "campaign": "calibration v2",
        "file": ".bank-work/calibration/calibration-runs-v2.json",
        "sha256": "6163880fc78bea1f43ba44a06569aa20714d1befa840740c315b369f062cbac3",
        "report": ".bank-work/calibration/report-v2.md",
        "runCount": 24,
        "packs": [
          "eu-ai-act",
          "kr-ai-framework-act"
        ],
        "arms": [
          "unadapted",
          "adapted"
        ]
      },
      {
        "campaign": "reference batch 1",
        "file": ".bank-work/calibration/runs-ref-batch1/summary.json",
        "sha256": "5c0d6acdc69debeeaacbfc9c5451b9aaf9e3c5098c25439b006557354f80cef0",
        "runCount": 20,
        "packs": [
          "sg-genai-governance",
          "cn-genai-measures",
          "nist-ai-rmf",
          "oecd-ai-principles"
        ],
        "arms": [
          "unadapted"
        ]
      },
      {
        "campaign": "reference batch 2",
        "file": ".bank-work/calibration/runs-ref-batch2/ref-batch2-runs.json",
        "sha256": "7fa00acbb78f1fc1119f2a8d5d1974963c0201766e1221bdf4937a2cd6e9453c",
        "runCount": 20,
        "packs": [
          "unesco-ai-ethics",
          "g7-hiroshima-code",
          "cn-ai-labelling",
          "eu-gpai-code"
        ],
        "arms": [
          "unadapted"
        ]
      },
      {
        "campaign": "eu-ai-act epoch-2 re-measurement",
        "file": ".bank-work/calibration/calibration-runs-eu-epoch2.json",
        "sha256": "f72d97fb6bf991b26f1e90d3f7024416b1bb2e7fd836efd41c2e381201d3ab35",
        "report": ".bank-work/calibration/report-eu-epoch2.md",
        "runCount": 10,
        "packs": [
          "eu-ai-act"
        ],
        "arms": [
          "unadapted",
          "adapted"
        ],
        "note": "Re-measurement of eu-ai-act on the epoch-2 bank (pack v0.3, rotated Gate A public bank, epoch-2 private Gate B pool). Same five models, same two arms, same methodology v2 protocol, same adaptation context file as the superseded panel. REPLACES the eu-ai-act entry; the panel it replaces is preserved verbatim under packs[eu-ai-act].supersededNote and is not read by referenceBandFor."
      }
    ]
  },
  "bands": {
    "below-unadapted-range": "The score is below the lowest score any panel model reached without being shown the pack.",
    "within-unadapted-range": "The score falls inside the range the panel reached without being shown the pack.",
    "above-unadapted-range": "The score is above the highest score any panel model reached without being shown the pack.",
    "no-reference-data-yet": "This pack has no reference distribution in this file. No band is reported."
  },
  "limits": [
    "Five models, one draw per cell. Ranges are the observed minimum and maximum of that panel, not a population interval.",
    "Each attempt is scored on a different draw from the private pool, so two measurements of one model differ by roughly 0.05 (eu-ai-act) to 0.12 (kr-ai-framework-act) on Gate B for that reason alone.",
    "Gate A items and their expected hierarchies are published, so Gate A numbers in both arms are floors.",
    "In eu-ai-act and kr-ai-framework-act, every choice item was answered correctly by every model in both arms, contributing a constant to roughly a quarter of each Gate B score. The packs added in reference batches 1 and 2 were not checked for this.",
    "The packs added in reference batch 1 (sg-genai-governance, cn-genai-measures, nist-ai-rmf, oecd-ai-principles) were measured in the unadapted arm only, with no repeat cells, so they carry no adapted delta and no campaign-internal noise estimate. Their drawNoise.predictedSE is a prediction from the observed per-item spread, not an observed run-to-run difference.",
    "Gate B pool sizes differ sharply between packs (36 to 108 items), so the number of Gate B items drawn differs too (9 to 27). A pack whose paper is 9 items carries visibly more draw noise than one whose paper is 27.",
    "The packs added in reference batch 2 (unesco-ai-ethics, g7-hiroshima-code, cn-ai-labelling, eu-gpai-code) were measured in the unadapted arm only, with no repeat cells, on the same five-model panel and the same v2-draft configuration as batch 1.",
    "On the packs added in reference batches 1 and 2, part of the weaker models’ unadapted floor is an AIO 20002 formatting floor rather than a substantive one. The dominant failure is a response that does not match the AIO 20002 log-line pattern: the scorer cannot read the answer at all, so the item scores at or near zero whatever the model knew about the provision. 87 of the 92 format-missed items across these packs come from deepseek/deepseek-chat-v3.1 and google/gemini-3.5-flash-lite, and the misses land on both gates, so the bottom of the unadapted range on these packs should be read as a floor on formatted output rather than on knowledge. Items with a format problem, per affected cell: eu-gpai-code/deepseek-chat-v3.1 15, unesco-ai-ethics/deepseek-chat-v3.1 15, nist-ai-rmf/deepseek-chat-v3.1 11, g7-hiroshima-code/deepseek-chat-v3.1 10, oecd-ai-principles/deepseek-chat-v3.1 7, unesco-ai-ethics/gemini-3.5-flash-lite 7, cn-genai-measures/deepseek-chat-v3.1 5, oecd-ai-principles/gemini-3.5-flash-lite 5, g7-hiroshima-code/gemini-3.5-flash-lite 4, nist-ai-rmf/gemini-3.5-flash-lite 4, sg-genai-governance/gemini-3.5-flash-lite 2, cn-ai-labelling/gemini-3.5-flash-lite 1, cn-genai-measures/claude-opus-5 1, cn-genai-measures/gemini-3.5-flash-lite 1, g7-hiroshima-code/claude-opus-5 1, sg-genai-governance/claude-haiku-4.5 1, unesco-ai-ethics/claude-haiku-4.5 1, unesco-ai-ethics/claude-sonnet-5 1."
  ],
  "packs": [
    {
      "packId": "eu-ai-act",
      "packVersion": "0.3",
      "provisionCount": 8,
      "measuredAt": "2026-08-15T14:50:20.532Z",
      "models": [
        "anthropic/claude-opus-5",
        "anthropic/claude-sonnet-5",
        "anthropic/claude-haiku-4.5",
        "deepseek/deepseek-chat-v3.1",
        "google/gemini-3.5-flash-lite"
      ],
      "conditions": {
        "methodologyVersion": "v2-draft",
        "temperature": 0,
        "gateAItemCount": 12,
        "gateBItemCount": 24,
        "itemsPerProvision": 3,
        "gateBPoolSize": 96,
        "gateBProvisionLabelsWithheld": true,
        "gateBItemIdsMasked": true,
        "layerWeights": {
          "v": 2,
          "e": 1,
          "s": 1
        },
        "selfAdministered": true,
        "repeatsPerCell": 1,
        "arms": {
          "unadapted": "The model was given the AIO 20002 response format and nothing about the pack.",
          "adapted": "The same prompt plus a full adaptation context built from the pack and its management guide."
        },
        "bankEpoch": 2,
        "bankEpochNote": "Gate A: pack v0.3, public bank rotated 2026-08-15 (eu-ai-act-002/004/007/012 retired, 015/016/018/019 added — the survivors of a two-round untaught-screening loop). Gate B: private pool epoch 2, 32 of the 96 active items retired and replaced on 2026-08-15T14:39–14:41Z; 12 active items per provision across 8.",
        "protocolDeviations": [
          {
            "model": "anthropic/claude-sonnet-5",
            "deviation": "maxTokens raised 16000 -> 48000 for both arms",
            "reason": "deterministic reasoning-token exhaustion at the standard ceiling (finish=length, 0 output chars, 16000 reasoning tokens, both arms, both attempts); measured runs consumed 20941/20372 reasoning tokens",
            "note": "maxTokens is a truncation guard, not a semantic condition; the alternative was a four-model panel. Recorded per methodology Q19."
          }
        ]
      },
      "sha256": {
        "standardsPack": "6ecd2002745bdc817cb91c472b4a1fef766919fe64f148fbd486f6709560f0ba",
        "standardsPackNote": "Correction recorded 2026-08-17: the pinned pack blob 6ecd2002… carries a note titled \"Downstream state at v0.3\" written before the downstream work landed. It asserts the public bank still declares packVersion 0.2 and that this distribution's pins no longer reproduce — both were true when the pack was authored but are stale: the public bank was rebuilt at packVersion 0.3 (commit 56837e2) and the pins in this entry were refreshed against it (commit 9540e88) and reproduce. The pack file is deliberately left byte-intact so this pin keeps describing the exact bytes the epoch-2 panel was measured against; the correction lives here instead.",
        "publicItemBank": "1b50af7b236e50bd6fff3ff4b78ca7e10f304ccb3a4b2f7afde35ec1450bad22",
        "adaptationContext": "620d90590d750b7c2987161079f728b43427dd4e5a6c097a9bfe132d7b939eb1",
        "adaptationContextNote": "Byte-identical to the context used for the superseded v0.2 panel (built from eu-ai-act-v0.2.json), and deliberately not rebuilt. The only v0.2 → v0.3 change inside the context surface is the Art. 12(2) quote, widened to carry the chapeau v0.2 elided behind \"[…]\"; rebuilding would change this pin and cost the side-by-side against the superseded panel, which is the comparison the epoch-2 measurement exists to make."
      },
      "gates": {
        "gateA": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.7823,
              "anthropic/claude-sonnet-5": 0.6565,
              "anthropic/claude-haiku-4.5": 0.5759,
              "deepseek/deepseek-chat-v3.1": 0.5448,
              "google/gemini-3.5-flash-lite": 0.6582
            },
            "mean": 0.6435,
            "min": 0.5448,
            "max": 0.7823,
            "p95": 0.7575
          },
          "adapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.7858,
              "anthropic/claude-sonnet-5": 0.6672,
              "anthropic/claude-haiku-4.5": 0.5737,
              "deepseek/deepseek-chat-v3.1": 0.7543,
              "google/gemini-3.5-flash-lite": 0.6871
            },
            "mean": 0.6936,
            "min": 0.5737,
            "max": 0.7858,
            "p95": 0.7795
          },
          "meanDelta": 0.0501,
          "improvedUnderAdaptation": 4
        },
        "gateB": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.6906,
              "anthropic/claude-sonnet-5": 0.7236,
              "anthropic/claude-haiku-4.5": 0.6663,
              "deepseek/deepseek-chat-v3.1": 0.574,
              "google/gemini-3.5-flash-lite": 0.7462
            },
            "mean": 0.6801,
            "min": 0.574,
            "max": 0.7462,
            "p95": 0.7417
          },
          "adapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.7865,
              "anthropic/claude-sonnet-5": 0.7625,
              "anthropic/claude-haiku-4.5": 0.6604,
              "deepseek/deepseek-chat-v3.1": 0.7083,
              "google/gemini-3.5-flash-lite": 0.6715
            },
            "mean": 0.7178,
            "min": 0.6604,
            "max": 0.7865,
            "p95": 0.7817
          },
          "meanDelta": 0.0377,
          "improvedUnderAdaptation": 3
        }
      },
      "itemBands": {
        "pool": 96,
        "seen": 90,
        "seenInBothArms": 50,
        "freeUnadaptedAtOrAbove090": 11,
        "discriminatingDeltaAtOrAbove015": 16,
        "hardAdaptedAtOrBelow050": 10
      },
      "drawNoise": {
        "perItemSD": 0.2449,
        "itemsDrawn": 24,
        "poolSize": 96,
        "predictedSE": 0.0435,
        "observedMeanAbsDeltaGateB": null
      },
      "supersededNote": {
        "bandsNote": "measured on the pre-epoch bank, retained for the historical record — these numbers are NOT used to compute any band",
        "replacedOn": "2026-08-15",
        "why": "The v2 authoring round rebuilt the EU bank on both gates: Gate A retired eu-ai-act-002/004/007/012 for 015/016/018/019 (pack v0.2 → v0.3), and the private Gate B pool rotated to epoch 2 on 2026-08-15 (32 of 96 active items retired and replaced). The pre-epoch range therefore describes a bank that no longer exists, and a score measured today must not be compared against it.",
        "packVersion": "0.2",
        "measuredAt": "2026-08-15T08:14:29.422Z",
        "models": [
          "anthropic/claude-opus-5",
          "anthropic/claude-sonnet-5",
          "anthropic/claude-haiku-4.5",
          "deepseek/deepseek-chat-v3.1",
          "google/gemini-3.5-flash-lite"
        ],
        "sha256": {
          "standardsPack": "2040c64a8806a84256efc56c8e46fb54855be183fdcda8c4e8c5981a7d2003b8",
          "publicItemBank": "073cb6d10039bf970928ab88fbcb552aa2436859bab8b599e3498e6d14cb18b0",
          "adaptationContext": "620d90590d750b7c2987161079f728b43427dd4e5a6c097a9bfe132d7b939eb1",
          "standardsPackNote": "Pin refers to eu-ai-act-v0.2.json as measured (blob 2040c64a…, commit 22d363d). The file was later changed ONLY by deprecation metadata when v0.3 superseded it (status + deprecatedBy, 2 lines; every quote and mapping byte-identical at measurement time). Current v0.2 blob: 5275ae81…. The EU distribution therefore still describes the bank/pack as measured; a re-measure is required only when the EU bank itself is rebuilt (that v2 round completed 2026-08-15; the epoch-2 panel in the current eu-ai-act entry is the re-measure)."
        },
        "gates": {
          "gateA": {
            "unadapted": {
              "n": 5,
              "perModel": {
                "anthropic/claude-opus-5": 0.8125,
                "anthropic/claude-sonnet-5": 0.7397,
                "anthropic/claude-haiku-4.5": 0.6543,
                "deepseek/deepseek-chat-v3.1": 0.7664,
                "google/gemini-3.5-flash-lite": 0.7125
              },
              "mean": 0.7371,
              "min": 0.6543,
              "max": 0.8125,
              "p95": 0.8033
            },
            "adapted": {
              "n": 5,
              "perModel": {
                "anthropic/claude-opus-5": 0.7892,
                "anthropic/claude-sonnet-5": 0.8254,
                "anthropic/claude-haiku-4.5": 0.6319,
                "deepseek/deepseek-chat-v3.1": 0.728,
                "google/gemini-3.5-flash-lite": 0.7405
              },
              "mean": 0.743,
              "min": 0.6319,
              "max": 0.8254,
              "p95": 0.8182
            },
            "meanDelta": 0.0059,
            "improvedUnderAdaptation": 2
          },
          "gateB": {
            "unadapted": {
              "n": 5,
              "perModel": {
                "anthropic/claude-opus-5": 0.8927,
                "anthropic/claude-sonnet-5": 0.8208,
                "anthropic/claude-haiku-4.5": 0.85,
                "deepseek/deepseek-chat-v3.1": 0.8375,
                "google/gemini-3.5-flash-lite": 0.7875
              },
              "mean": 0.8377,
              "min": 0.7875,
              "max": 0.8927,
              "p95": 0.8842
            },
            "adapted": {
              "n": 5,
              "perModel": {
                "anthropic/claude-opus-5": 0.8646,
                "anthropic/claude-sonnet-5": 0.8042,
                "anthropic/claude-haiku-4.5": 0.8469,
                "deepseek/deepseek-chat-v3.1": 0.949,
                "google/gemini-3.5-flash-lite": 0.9031
              },
              "mean": 0.8736,
              "min": 0.8042,
              "max": 0.949,
              "p95": 0.9398
            },
            "meanDelta": 0.0359,
            "improvedUnderAdaptation": 2
          }
        },
        "itemBands": {
          "pool": 96,
          "seen": 91,
          "seenInBothArms": 69,
          "freeUnadaptedAtOrAbove090": 36,
          "discriminatingDeltaAtOrAbove015": 10,
          "hardAdaptedAtOrBelow050": 1
        },
        "drawNoise": {
          "perItemSD": 0.176,
          "itemsDrawn": 24,
          "poolSize": 96,
          "predictedSE": 0.0313,
          "observedMeanAbsDeltaGateB": 0.0511
        }
      }
    },
    {
      "packId": "kr-ai-framework-act",
      "packVersion": "0.3",
      "provisionCount": 9,
      "measuredAt": "2026-08-15T08:31:12.580Z",
      "models": [
        "anthropic/claude-opus-5",
        "anthropic/claude-sonnet-5",
        "anthropic/claude-haiku-4.5",
        "deepseek/deepseek-chat-v3.1",
        "google/gemini-3.5-flash-lite"
      ],
      "conditions": {
        "methodologyVersion": "v2-draft",
        "temperature": 0,
        "gateAItemCount": 12,
        "gateBItemCount": 27,
        "itemsPerProvision": 3,
        "gateBPoolSize": 108,
        "gateBProvisionLabelsWithheld": true,
        "gateBItemIdsMasked": true,
        "layerWeights": {
          "v": 2,
          "e": 1,
          "s": 1
        },
        "selfAdministered": true,
        "repeatsPerCell": 1,
        "arms": {
          "unadapted": "The model was given the AIO 20002 response format and nothing about the pack.",
          "adapted": "The same prompt plus a full adaptation context built from the pack and its management guide."
        }
      },
      "sha256": {
        "standardsPack": "a3af36c97dd47bfc52f908e3c5828a34e69b91e126989af3a3d4df68b52036b4",
        "publicItemBank": "c0efd44bf1c0a6b170af95bb24ac4b433619e9ffc8b3a2403da60f7641d05802",
        "adaptationContext": "70e5230e1f62788a1080f7c0f1741483779399fe0a94cd1f55a325d3e7359095"
      },
      "gates": {
        "gateA": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.6346,
              "anthropic/claude-sonnet-5": 0.5815,
              "anthropic/claude-haiku-4.5": 0.6369,
              "deepseek/deepseek-chat-v3.1": 0.4735,
              "google/gemini-3.5-flash-lite": 0.4615
            },
            "mean": 0.5576,
            "min": 0.4615,
            "max": 0.6369,
            "p95": 0.6364
          },
          "adapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.854,
              "anthropic/claude-sonnet-5": 0.7479,
              "anthropic/claude-haiku-4.5": 0.6244,
              "deepseek/deepseek-chat-v3.1": 0.5575,
              "google/gemini-3.5-flash-lite": 0.4485
            },
            "mean": 0.6465,
            "min": 0.4485,
            "max": 0.854,
            "p95": 0.8328
          },
          "meanDelta": 0.0889,
          "improvedUnderAdaptation": 3
        },
        "gateB": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.6713,
              "anthropic/claude-sonnet-5": 0.5972,
              "anthropic/claude-haiku-4.5": 0.6454,
              "deepseek/deepseek-chat-v3.1": 0.5685,
              "google/gemini-3.5-flash-lite": 0.4324
            },
            "mean": 0.583,
            "min": 0.4324,
            "max": 0.6713,
            "p95": 0.6661
          },
          "adapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.8787,
              "anthropic/claude-sonnet-5": 0.6898,
              "anthropic/claude-haiku-4.5": 0.6676,
              "deepseek/deepseek-chat-v3.1": 0.5157,
              "google/gemini-3.5-flash-lite": 0.5241
            },
            "mean": 0.6552,
            "min": 0.5157,
            "max": 0.8787,
            "p95": 0.8409
          },
          "meanDelta": 0.0722,
          "improvedUnderAdaptation": 4
        }
      },
      "itemBands": {
        "pool": 108,
        "seen": 103,
        "seenInBothArms": 70,
        "freeUnadaptedAtOrAbove090": 22,
        "discriminatingDeltaAtOrAbove015": 27,
        "hardAdaptedAtOrBelow050": 30
      },
      "drawNoise": {
        "perItemSD": 0.3187,
        "itemsDrawn": 27,
        "poolSize": 108,
        "predictedSE": 0.0534,
        "observedMeanAbsDeltaGateB": 0.1241
      }
    },
    {
      "packId": "sg-genai-governance",
      "packVersion": "0.2",
      "provisionCount": 3,
      "measuredAt": "2026-08-15T11:13:09.208Z",
      "models": [
        "anthropic/claude-opus-5",
        "anthropic/claude-sonnet-5",
        "anthropic/claude-haiku-4.5",
        "deepseek/deepseek-chat-v3.1",
        "google/gemini-3.5-flash-lite"
      ],
      "conditions": {
        "methodologyVersion": "v2-draft",
        "temperature": 0,
        "gateAItemCount": 12,
        "gateBItemCount": 9,
        "itemsPerProvision": 3,
        "gateBPoolSize": 36,
        "gateBProvisionLabelsWithheld": true,
        "gateBItemIdsMasked": true,
        "layerWeights": {
          "v": 2,
          "e": 1,
          "s": 1
        },
        "selfAdministered": true,
        "repeatsPerCell": 1,
        "arms": {
          "unadapted": "The model was given the AIO 20002 response format and nothing about the pack."
        },
        "armsMeasured": [
          "unadapted"
        ],
        "adaptedArmNote": "Not measured for this pack. The reference band compares against the unadapted range only; the adapted arm needs an adaptation context built by hand from the pack and its management guide, which is a separate decision per pack."
      },
      "sha256": {
        "standardsPack": "63b9e7d23c7c7bf7b9d7cf92d1f63c899242ffaeaf3f11d9bee7764a0fedc1ec",
        "publicItemBank": "5c110e17b2a30ed385bc883de0138bf403521d6698c00b71c76b917ddfa4f81a",
        "adaptationContext": null
      },
      "gates": {
        "gateA": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.4908,
              "anthropic/claude-sonnet-5": 0.4656,
              "anthropic/claude-haiku-4.5": 0.3458,
              "deepseek/deepseek-chat-v3.1": 0.3402,
              "google/gemini-3.5-flash-lite": 0.4119
            },
            "mean": 0.4109,
            "min": 0.3402,
            "max": 0.4908,
            "p95": 0.4858
          }
        },
        "gateB": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.5111,
              "anthropic/claude-sonnet-5": 0.6028,
              "anthropic/claude-haiku-4.5": 0.3389,
              "deepseek/deepseek-chat-v3.1": 0.5556,
              "google/gemini-3.5-flash-lite": 0.6444
            },
            "mean": 0.5306,
            "min": 0.3389,
            "max": 0.6444,
            "p95": 0.6361
          }
        }
      },
      "itemBands": {
        "pool": 36,
        "seen": 29,
        "freeUnadaptedAtOrAbove090": 8
      },
      "drawNoise": {
        "perItemSD": 0.3583,
        "itemsDrawn": 9,
        "poolSize": 36,
        "predictedSE": 0.1049,
        "observedMeanAbsDeltaGateB": null
      }
    },
    {
      "packId": "cn-genai-measures",
      "packVersion": "0.2",
      "provisionCount": 4,
      "measuredAt": "2026-08-15T12:31:34.823Z",
      "models": [
        "anthropic/claude-opus-5",
        "anthropic/claude-sonnet-5",
        "anthropic/claude-haiku-4.5",
        "deepseek/deepseek-chat-v3.1",
        "google/gemini-3.5-flash-lite"
      ],
      "conditions": {
        "methodologyVersion": "v2-draft",
        "temperature": 0,
        "gateAItemCount": 12,
        "gateBItemCount": 12,
        "itemsPerProvision": 3,
        "gateBPoolSize": 48,
        "gateBProvisionLabelsWithheld": true,
        "gateBItemIdsMasked": true,
        "layerWeights": {
          "v": 2,
          "e": 1,
          "s": 1
        },
        "selfAdministered": true,
        "repeatsPerCell": 1,
        "arms": {
          "unadapted": "The model was given the AIO 20002 response format and nothing about the pack."
        },
        "armsMeasured": [
          "unadapted"
        ],
        "adaptedArmNote": "Not measured for this pack. The reference band compares against the unadapted range only; the adapted arm needs an adaptation context built by hand from the pack and its management guide, which is a separate decision per pack."
      },
      "sha256": {
        "standardsPack": "713b61a86f535db5ba5f8c08a366b5c05853113c703cf15443ed8a9e699fba93",
        "publicItemBank": "fe308e6cca3373d5a8aa6f1899e503f1399373920d5848daf7eef4960d007f4f",
        "adaptationContext": null
      },
      "gates": {
        "gateA": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.7322,
              "anthropic/claude-sonnet-5": 0.7295,
              "anthropic/claude-haiku-4.5": 0.6844,
              "deepseek/deepseek-chat-v3.1": 0.343,
              "google/gemini-3.5-flash-lite": 0.4789
            },
            "mean": 0.5936,
            "min": 0.343,
            "max": 0.7322,
            "p95": 0.7317
          }
        },
        "gateB": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.7785,
              "anthropic/claude-sonnet-5": 0.6528,
              "anthropic/claude-haiku-4.5": 0.7431,
              "deepseek/deepseek-chat-v3.1": 0.399,
              "google/gemini-3.5-flash-lite": 0.5257
            },
            "mean": 0.6198,
            "min": 0.399,
            "max": 0.7785,
            "p95": 0.7714
          }
        }
      },
      "itemBands": {
        "pool": 48,
        "seen": 40,
        "freeUnadaptedAtOrAbove090": 13
      },
      "drawNoise": {
        "perItemSD": 0.3182,
        "itemsDrawn": 12,
        "poolSize": 48,
        "predictedSE": 0.0804,
        "observedMeanAbsDeltaGateB": null
      }
    },
    {
      "packId": "nist-ai-rmf",
      "packVersion": "0.2",
      "provisionCount": 9,
      "measuredAt": "2026-08-15T12:41:49.650Z",
      "models": [
        "anthropic/claude-opus-5",
        "anthropic/claude-sonnet-5",
        "anthropic/claude-haiku-4.5",
        "deepseek/deepseek-chat-v3.1",
        "google/gemini-3.5-flash-lite"
      ],
      "conditions": {
        "methodologyVersion": "v2-draft",
        "temperature": 0,
        "gateAItemCount": 12,
        "gateBItemCount": 27,
        "itemsPerProvision": 3,
        "gateBPoolSize": 108,
        "gateBProvisionLabelsWithheld": true,
        "gateBItemIdsMasked": true,
        "layerWeights": {
          "v": 2,
          "e": 1,
          "s": 1
        },
        "selfAdministered": true,
        "repeatsPerCell": 1,
        "arms": {
          "unadapted": "The model was given the AIO 20002 response format and nothing about the pack."
        },
        "armsMeasured": [
          "unadapted"
        ],
        "adaptedArmNote": "Not measured for this pack. The reference band compares against the unadapted range only; the adapted arm needs an adaptation context built by hand from the pack and its management guide, which is a separate decision per pack."
      },
      "sha256": {
        "standardsPack": "18370f388fe6650f6245c5ea9b5893c262e83df01b8c5b4fb9c527a8a8f2d6de",
        "publicItemBank": "683170106cc25eae768386db7a294099d47293885d13d7db46819376a17d1f2d",
        "adaptationContext": null
      },
      "gates": {
        "gateA": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.5655,
              "anthropic/claude-sonnet-5": 0.3887,
              "anthropic/claude-haiku-4.5": 0.5345,
              "deepseek/deepseek-chat-v3.1": 0.4534,
              "google/gemini-3.5-flash-lite": 0.4498
            },
            "mean": 0.4784,
            "min": 0.3887,
            "max": 0.5655,
            "p95": 0.5593
          }
        },
        "gateB": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.5481,
              "anthropic/claude-sonnet-5": 0.3306,
              "anthropic/claude-haiku-4.5": 0.387,
              "deepseek/deepseek-chat-v3.1": 0.4824,
              "google/gemini-3.5-flash-lite": 0.4796
            },
            "mean": 0.4455,
            "min": 0.3306,
            "max": 0.5481,
            "p95": 0.535
          }
        }
      },
      "itemBands": {
        "pool": 108,
        "seen": 80,
        "freeUnadaptedAtOrAbove090": 17
      },
      "drawNoise": {
        "perItemSD": 0.3228,
        "itemsDrawn": 27,
        "poolSize": 108,
        "predictedSE": 0.0541,
        "observedMeanAbsDeltaGateB": null
      }
    },
    {
      "packId": "oecd-ai-principles",
      "packVersion": "0.2",
      "provisionCount": 5,
      "measuredAt": "2026-08-15T12:50:21.722Z",
      "models": [
        "anthropic/claude-opus-5",
        "anthropic/claude-sonnet-5",
        "anthropic/claude-haiku-4.5",
        "deepseek/deepseek-chat-v3.1",
        "google/gemini-3.5-flash-lite"
      ],
      "conditions": {
        "methodologyVersion": "v2-draft",
        "temperature": 0,
        "gateAItemCount": 12,
        "gateBItemCount": 15,
        "itemsPerProvision": 3,
        "gateBPoolSize": 60,
        "gateBProvisionLabelsWithheld": true,
        "gateBItemIdsMasked": true,
        "layerWeights": {
          "v": 2,
          "e": 1,
          "s": 1
        },
        "selfAdministered": true,
        "repeatsPerCell": 1,
        "arms": {
          "unadapted": "The model was given the AIO 20002 response format and nothing about the pack."
        },
        "armsMeasured": [
          "unadapted"
        ],
        "adaptedArmNote": "Not measured for this pack. The reference band compares against the unadapted range only; the adapted arm needs an adaptation context built by hand from the pack and its management guide, which is a separate decision per pack."
      },
      "sha256": {
        "standardsPack": "aa8c50ff7ba8c0065ae5f546cf528e605e13f817e01d524f293cf30f64dc5e96",
        "publicItemBank": "0628052cd747fa4d2610ac7f0743e6ad56d8310a7ccd8e58e03ee93859334bf8",
        "adaptationContext": null
      },
      "gates": {
        "gateA": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.6725,
              "anthropic/claude-sonnet-5": 0.6903,
              "anthropic/claude-haiku-4.5": 0.578,
              "deepseek/deepseek-chat-v3.1": 0.3522,
              "google/gemini-3.5-flash-lite": 0.3564
            },
            "mean": 0.5299,
            "min": 0.3522,
            "max": 0.6903,
            "p95": 0.6867
          }
        },
        "gateB": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.7283,
              "anthropic/claude-sonnet-5": 0.7017,
              "anthropic/claude-haiku-4.5": 0.5561,
              "deepseek/deepseek-chat-v3.1": 0.4833,
              "google/gemini-3.5-flash-lite": 0.4661
            },
            "mean": 0.5871,
            "min": 0.4661,
            "max": 0.7283,
            "p95": 0.723
          }
        }
      },
      "itemBands": {
        "pool": 60,
        "seen": 45,
        "freeUnadaptedAtOrAbove090": 13
      },
      "drawNoise": {
        "perItemSD": 0.3751,
        "itemsDrawn": 15,
        "poolSize": 60,
        "predictedSE": 0.0846,
        "observedMeanAbsDeltaGateB": null
      }
    },
    {
      "packId": "unesco-ai-ethics",
      "packVersion": "0.2",
      "provisionCount": 7,
      "measuredAt": "2026-08-15T11:00:03.411Z",
      "models": [
        "anthropic/claude-opus-5",
        "anthropic/claude-sonnet-5",
        "anthropic/claude-haiku-4.5",
        "deepseek/deepseek-chat-v3.1",
        "google/gemini-3.5-flash-lite"
      ],
      "conditions": {
        "methodologyVersion": "v2-draft",
        "temperature": 0,
        "gateAItemCount": 12,
        "gateBItemCount": 21,
        "itemsPerProvision": 3,
        "gateBPoolSize": 84,
        "gateBProvisionLabelsWithheld": true,
        "gateBItemIdsMasked": true,
        "layerWeights": {
          "v": 2,
          "e": 1,
          "s": 1
        },
        "selfAdministered": true,
        "repeatsPerCell": 1,
        "arms": {
          "unadapted": "The model was given the AIO 20002 response format and nothing about the pack."
        },
        "armsMeasured": [
          "unadapted"
        ],
        "adaptedArmNote": "Not measured for this pack. The reference band compares against the unadapted range only; the adapted arm needs an adaptation context built by hand from the pack and its management guide, which is a separate decision per pack."
      },
      "sha256": {
        "standardsPack": "f21779e2b78dd0ed63789eff7608fe31443c410ce6e5f5c4d37bb8e9d2a7c37d",
        "publicItemBank": "dd82af2d347653eb72a2b9739e049815f58e66a0c009b9896405ba2683cbe229",
        "adaptationContext": null
      },
      "gates": {
        "gateA": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.6636,
              "anthropic/claude-sonnet-5": 0.6865,
              "anthropic/claude-haiku-4.5": 0.4976,
              "deepseek/deepseek-chat-v3.1": 0.4178,
              "google/gemini-3.5-flash-lite": 0.3982
            },
            "mean": 0.5327,
            "min": 0.3982,
            "max": 0.6865,
            "p95": 0.6819
          }
        },
        "gateB": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.5524,
              "anthropic/claude-sonnet-5": 0.6425,
              "anthropic/claude-haiku-4.5": 0.5151,
              "deepseek/deepseek-chat-v3.1": 0.5079,
              "google/gemini-3.5-flash-lite": 0.4587
            },
            "mean": 0.5353,
            "min": 0.4587,
            "max": 0.6425,
            "p95": 0.6245
          }
        }
      },
      "itemBands": {
        "pool": 84,
        "seen": 62,
        "freeUnadaptedAtOrAbove090": 15
      },
      "drawNoise": {
        "perItemSD": 0.3759,
        "itemsDrawn": 21,
        "poolSize": 84,
        "predictedSE": 0.0715,
        "observedMeanAbsDeltaGateB": null
      }
    },
    {
      "packId": "g7-hiroshima-code",
      "packVersion": "0.2",
      "provisionCount": 8,
      "measuredAt": "2026-08-15T11:09:33.752Z",
      "models": [
        "anthropic/claude-opus-5",
        "anthropic/claude-sonnet-5",
        "anthropic/claude-haiku-4.5",
        "deepseek/deepseek-chat-v3.1",
        "google/gemini-3.5-flash-lite"
      ],
      "conditions": {
        "methodologyVersion": "v2-draft",
        "temperature": 0,
        "gateAItemCount": 12,
        "gateBItemCount": 24,
        "itemsPerProvision": 3,
        "gateBPoolSize": 96,
        "gateBProvisionLabelsWithheld": true,
        "gateBItemIdsMasked": true,
        "layerWeights": {
          "v": 2,
          "e": 1,
          "s": 1
        },
        "selfAdministered": true,
        "repeatsPerCell": 1,
        "arms": {
          "unadapted": "The model was given the AIO 20002 response format and nothing about the pack."
        },
        "armsMeasured": [
          "unadapted"
        ],
        "adaptedArmNote": "Not measured for this pack. The reference band compares against the unadapted range only; the adapted arm needs an adaptation context built by hand from the pack and its management guide, which is a separate decision per pack."
      },
      "sha256": {
        "standardsPack": "6f7b580660e71fecbd82b85ee6690e9fb9d0fa0d268a526e88eb22cd36eae3fb",
        "publicItemBank": "117407663349eb21a3d8c9dcb95ae7a3a6219527f5a255efddba06324a3f4b38",
        "adaptationContext": null
      },
      "gates": {
        "gateA": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.5245,
              "anthropic/claude-sonnet-5": 0.5188,
              "anthropic/claude-haiku-4.5": 0.3572,
              "deepseek/deepseek-chat-v3.1": 0.3916,
              "google/gemini-3.5-flash-lite": 0.4192
            },
            "mean": 0.4423,
            "min": 0.3572,
            "max": 0.5245,
            "p95": 0.5234
          }
        },
        "gateB": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.6222,
              "anthropic/claude-sonnet-5": 0.5993,
              "anthropic/claude-haiku-4.5": 0.4042,
              "deepseek/deepseek-chat-v3.1": 0.5417,
              "google/gemini-3.5-flash-lite": 0.6816
            },
            "mean": 0.5698,
            "min": 0.4042,
            "max": 0.6816,
            "p95": 0.6697
          }
        }
      },
      "itemBands": {
        "pool": 96,
        "seen": 76,
        "freeUnadaptedAtOrAbove090": 20
      },
      "drawNoise": {
        "perItemSD": 0.3513,
        "itemsDrawn": 24,
        "poolSize": 96,
        "predictedSE": 0.0624,
        "observedMeanAbsDeltaGateB": null
      }
    },
    {
      "packId": "cn-ai-labelling",
      "packVersion": "0.2",
      "provisionCount": 7,
      "measuredAt": "2026-08-15T12:55:08.625Z",
      "models": [
        "anthropic/claude-opus-5",
        "anthropic/claude-sonnet-5",
        "anthropic/claude-haiku-4.5",
        "deepseek/deepseek-chat-v3.1",
        "google/gemini-3.5-flash-lite"
      ],
      "conditions": {
        "methodologyVersion": "v2-draft",
        "temperature": 0,
        "gateAItemCount": 12,
        "gateBItemCount": 21,
        "itemsPerProvision": 3,
        "gateBPoolSize": 81,
        "gateBProvisionLabelsWithheld": true,
        "gateBItemIdsMasked": true,
        "layerWeights": {
          "v": 2,
          "e": 1,
          "s": 1
        },
        "selfAdministered": true,
        "repeatsPerCell": 1,
        "arms": {
          "unadapted": "The model was given the AIO 20002 response format and nothing about the pack."
        },
        "armsMeasured": [
          "unadapted"
        ],
        "adaptedArmNote": "Not measured for this pack. The reference band compares against the unadapted range only; the adapted arm needs an adaptation context built by hand from the pack and its management guide, which is a separate decision per pack."
      },
      "sha256": {
        "standardsPack": "63d976778aebf8b9480e82d64fee1b655f7648a9c9590d6722b14ec26e383598",
        "publicItemBank": "0bece0823c88d382063c12036ec5c21831e4c96a47f5881a31c9d33368da3935",
        "adaptationContext": null
      },
      "gates": {
        "gateA": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.6988,
              "anthropic/claude-sonnet-5": 0.6865,
              "anthropic/claude-haiku-4.5": 0.6891,
              "deepseek/deepseek-chat-v3.1": 0.449,
              "google/gemini-3.5-flash-lite": 0.5193
            },
            "mean": 0.6085,
            "min": 0.449,
            "max": 0.6988,
            "p95": 0.6969
          }
        },
        "gateB": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.6476,
              "anthropic/claude-sonnet-5": 0.6726,
              "anthropic/claude-haiku-4.5": 0.725,
              "deepseek/deepseek-chat-v3.1": 0.4988,
              "google/gemini-3.5-flash-lite": 0.4524
            },
            "mean": 0.5993,
            "min": 0.4524,
            "max": 0.725,
            "p95": 0.7145
          }
        }
      },
      "itemBands": {
        "pool": 81,
        "seen": 64,
        "freeUnadaptedAtOrAbove090": 18
      },
      "drawNoise": {
        "perItemSD": 0.3193,
        "itemsDrawn": 21,
        "poolSize": 81,
        "predictedSE": 0.0603,
        "observedMeanAbsDeltaGateB": null
      }
    },
    {
      "packId": "eu-gpai-code",
      "packVersion": "0.2",
      "provisionCount": 5,
      "measuredAt": "2026-08-15T11:36:21.161Z",
      "models": [
        "anthropic/claude-opus-5",
        "anthropic/claude-sonnet-5",
        "anthropic/claude-haiku-4.5",
        "deepseek/deepseek-chat-v3.1",
        "google/gemini-3.5-flash-lite"
      ],
      "conditions": {
        "methodologyVersion": "v2-draft",
        "temperature": 0,
        "gateAItemCount": 12,
        "gateBItemCount": 15,
        "itemsPerProvision": 3,
        "gateBPoolSize": 60,
        "gateBProvisionLabelsWithheld": true,
        "gateBItemIdsMasked": true,
        "layerWeights": {
          "v": 2,
          "e": 1,
          "s": 1
        },
        "selfAdministered": true,
        "repeatsPerCell": 1,
        "arms": {
          "unadapted": "The model was given the AIO 20002 response format and nothing about the pack."
        },
        "armsMeasured": [
          "unadapted"
        ],
        "adaptedArmNote": "Not measured for this pack. The reference band compares against the unadapted range only; the adapted arm needs an adaptation context built by hand from the pack and its management guide, which is a separate decision per pack."
      },
      "sha256": {
        "standardsPack": "c7072eee707f9389e5d22a1d933a6dfdf856dcc91254fdbc7581e7a7805d8c97",
        "publicItemBank": "e95d60700d68bb62de1b8b12ad5292be57df21af93828bd2d4f38a97537676cd",
        "adaptationContext": null
      },
      "gates": {
        "gateA": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.6092,
              "anthropic/claude-sonnet-5": 0.3781,
              "anthropic/claude-haiku-4.5": 0.5333,
              "deepseek/deepseek-chat-v3.1": 0.4731,
              "google/gemini-3.5-flash-lite": 0.3875
            },
            "mean": 0.4762,
            "min": 0.3781,
            "max": 0.6092,
            "p95": 0.594
          }
        },
        "gateB": {
          "unadapted": {
            "n": 5,
            "perModel": {
              "anthropic/claude-opus-5": 0.655,
              "anthropic/claude-sonnet-5": 0.6,
              "anthropic/claude-haiku-4.5": 0.5817,
              "deepseek/deepseek-chat-v3.1": 0.57,
              "google/gemini-3.5-flash-lite": 0.4083
            },
            "mean": 0.563,
            "min": 0.4083,
            "max": 0.655,
            "p95": 0.644
          }
        }
      },
      "itemBands": {
        "pool": 60,
        "seen": 46,
        "freeUnadaptedAtOrAbove090": 15
      },
      "drawNoise": {
        "perItemSD": 0.3483,
        "itemsDrawn": 15,
        "poolSize": 60,
        "predictedSE": 0.0785,
        "observedMeanAbsDeltaGateB": null
      }
    }
  ]
}