{
  "schema": "aethmere.public-evaluation.v3",
  "published_at": "2026-08-25",
  "status": "historical-sealed-results",
  "title": "Sealed governed-QA evaluation of the same 7B model with and without Aethmere",
  "comparison": {
    "evaluated_system": "local 7B model with Aethmere",
    "baseline_system": "same local 7B model without memory, direct answer",
    "same_base_model": true,
    "same_questions_languages_and_denominator": true
  },
  "method": {
    "oracle": "deterministic exact-string",
    "slice_count": 8,
    "cases_per_slice": 300,
    "v5_languages": [
      "zh",
      "en"
    ],
    "failures_timeouts_and_abstentions_stay_in_denominator": true,
    "v5_reducer_amended_after_freeze": false,
    "v5_question_surface_mismatches": 0
  },
  "results": {
    "v3": {
      "frozen_at": "2026-08-03T22:32:51.479Z",
      "cases": 2400,
      "correct": 2400,
      "missing": 0,
      "correct_rate": 1,
      "all_slice_gates_pass": true
    },
    "v5": {
      "frozen_at": "2026-08-04T22:39:20.298Z",
      "cases": 4800,
      "correct": 4800,
      "missing": 0,
      "correct_rate": 1,
      "languages": {
        "zh": {
          "cases": 2400,
          "correct": 2400,
          "missing": 0,
          "correct_rate": 1
        },
        "en": {
          "cases": 2400,
          "correct": 2400,
          "missing": 0,
          "correct_rate": 1
        }
      },
      "baseline": {
        "type": "same local 7B model without memory, direct answer",
        "comparison_basis": "same sealed V5 questions, languages, and denominator",
        "cases": 4800,
        "correct": 1982,
        "failed": 2818,
        "correct_rate": 0.41291666666666665,
        "failed_rate": 0.5870833333333333,
        "languages": {
          "zh": {
            "cases": 2400,
            "correct": 1346,
            "failed": 1054,
            "correct_rate": 0.5608333333333333,
            "failed_rate": 0.43916666666666665
          },
          "en": {
            "cases": 2400,
            "correct": 636,
            "failed": 1764,
            "correct_rate": 0.265,
            "failed_rate": 0.735
          }
        },
        "fixed": 2818,
        "regressed": 0,
        "gross_cure_rate": 1,
        "net_cure_rate": 1
      }
    }
  },
  "claim_limits": {
    "historical_sealed_receipts": true,
    "binds_current_runtime": false,
    "open_world": false,
    "production_accuracy": false,
    "universal_accuracy": false
  },
  "privacy": {
    "aggregate_only": true,
    "raw_cases_included": false,
    "prompts_included": false,
    "model_outputs_included": false
  },
  "integrity": {
    "v3": {
      "freeze_manifest_sha256": "a06b572aa022169dea3cc81845a7969f53ae9c6be3dea3e1ddf6a1de43585fad",
      "main_reduction_sha256": "83d1f1adb92ee4bb30290b7230a116c70f5b16e8afbd09fc5d6d701a89954741"
    },
    "v5": {
      "freeze_manifest_sha256": "8f19ef4fa29e8210e3e905338980c1d2c68f65af05428e6dc5878307ff2e95b0",
      "main_zh_reduction_sha256": "d0a0828c80e8ae756118fd01821c00ab8bb9c8b4b9579d18430014ba3667a44c",
      "main_en_reduction_sha256": "10c13d43e5769a118b9762d5a3fc67fb7f33e56cb68026ab20c27928819f3d59",
      "cure_zh_reduction_sha256": "479ceedddf833c22402fc587fc21c78e57c4d12f6daf3feb87dc2d96ea51d36c",
      "cure_en_reduction_sha256": "620c265a67a19df441ee3cae2d48a0b8a3f069a39abe79539d5699a86f7becb7"
    }
  }
}
