{
  "name": "audit.domains valuation methodology back-test",
  "url": "https://audit.domains/accuracy/backtest.json",
  "page": "https://audit.domains/accuracy/",
  "license": "https://audit.domains/terms/",
  "citation": "audit.domains, methodology revision 4.14, back-test evaluated 2026-09-30. Quote every figure with its cut label and basis; the table cannot be compressed into one accuracy percentage.",
  "generatedFrom": {
    "artifact": "src/lib/facts/public-facts.generated.json",
    "generatedAt": "2026-09-30T03:45:05.466Z",
    "generator": "scripts/generate-public-facts.ts"
  },
  "evaluationDate": "2026-09-30T03:45:04.578Z",
  "revisions": {
    "live": "4.14",
    "candidate": "4.13",
    "incumbent": "4.2",
    "personalName": "4.14",
    "status": {
      "candidate": "active-structural-base",
      "incumbent": "retired-as-default",
      "personalName": "active-default"
    },
    "unpublished": [
      {
        "revision": "5.0",
        "enabled": false
      },
      {
        "revision": "5.1",
        "enabled": false
      }
    ]
  },
  "corpus": {
    "comparableCorpusRecords": 2994,
    "earliestYear": 1999,
    "latestYear": 2026,
    "priceFloorUsd": 100,
    "priceCeilingUsd": 50000000,
    "evaluationSetSize": 218,
    "evaluationPopulationPolicy": "recent-slice-2026-09-23"
  },
  "headlineCut": "blind",
  "cuts": {
    "blind": {
      "label": "blind",
      "basis": "each sale withheld from the engine's evidence",
      "n": 218,
      "medianApe": 85.1,
      "mape": 96.5,
      "withinFactor2": 29.4,
      "withinFactor3": 39.9,
      "withinPlusMinus50": 23.9,
      "log10Rmse": 0.976,
      "medianRatio": 0.25
    },
    "recall": {
      "label": "recall",
      "basis": "the same sales with each one's own sale added to the engine's evidence",
      "n": 218,
      "medianApe": 26,
      "mape": 28,
      "withinFactor2": 94,
      "withinFactor3": 99.1,
      "withinPlusMinus50": 92.7,
      "log10Rmse": 0.171,
      "medianRatio": 0.76
    },
    "incumbent": {
      "label": "incumbent",
      "basis": "run as shipped, never having seen these sales",
      "n": 218,
      "medianApe": 99.8,
      "mape": 1513.5,
      "withinFactor2": 18.8,
      "withinFactor3": 26.1,
      "withinPlusMinus50": 15.1,
      "log10Rmse": 1.664,
      "medianRatio": 0.72
    }
  },
  "slices": {
    "source": "tools/scoreboard/baselines/champion.json",
    "estimator": "champion (rev 4.13, generalization)",
    "frozenTestSha256": "5866aaffbc4e957a5fa9c2514587ff6bd9544ee0be2b6d78edd39a41ed5d6a7d",
    "totalN": 618,
    "rows": [
      {
        "key": "recent",
        "name": "Recent sales",
        "basis": "sales reported by DNJournal and Sedo, kept out of the corpus the engine draws its evidence from",
        "published": true,
        "n": 218,
        "within2": 29.36,
        "within3": 39.91,
        "medianApe": 85.1,
        "medianRatio": 0.25,
        "medianFactorError": 5.01,
        "ci95": {
          "within2": {
            "lo": 22.94,
            "hi": 35.78
          },
          "within3": {
            "lo": 33.03,
            "hi": 46.79
          },
          "medianApe": {
            "lo": 80.6,
            "hi": 89
          },
          "medianRatio": {
            "lo": 0.16,
            "hi": 0.35
          },
          "medianFactorError": {
            "lo": 3.51,
            "hi": 6.48
          }
        }
      },
      {
        "key": "ledger",
        "name": "Bulk ledger sample",
        "basis": "a deterministic sample of the recorded-sales ledger: the auction-heavy bulk distribution a drop list is drawn from",
        "published": false,
        "n": 400,
        "within2": 13,
        "within3": 19.5,
        "medianApe": 871,
        "medianRatio": 9.71,
        "medianFactorError": 11.31,
        "ci95": {
          "within2": {
            "lo": 9.75,
            "hi": 16.25
          },
          "within3": {
            "lo": 15.75,
            "hi": 23.5
          },
          "medianApe": {
            "lo": 720.6,
            "hi": 1108.9
          },
          "medianRatio": {
            "lo": 8.21,
            "hi": 12.09
          },
          "medianFactorError": {
            "lo": 9.45,
            "hi": 13.22
          }
        }
      }
    ]
  },
  "rankSlices": {
    "source": "tools/scoreboard/baselines/two-output.json",
    "estimator": "two-output (price rev 4.13 / rank rev 5.0, no gate)",
    "frozenTestSha256": "5866aaffbc4e957a5fa9c2514587ff6bd9544ee0be2b6d78edd39a41ed5d6a7d",
    "totalN": 618,
    "rows": [
      {
        "key": "recent",
        "name": "Recent sales",
        "n": 218,
        "spearman": 0.3506,
        "precisionAt50": 38,
        "valueCaptureAt50": 53.11
      },
      {
        "key": "ledger",
        "name": "Bulk ledger sample",
        "n": 400,
        "spearman": 0.3347,
        "precisionAt50": 42,
        "valueCaptureAt50": 36.72
      }
    ]
  },
  "metricWording": {
    "medianApe": "Median error",
    "mape": "Mean error (MAPE)",
    "withinFactor2": "Within 2x of sale price",
    "withinFactor3": "Within 3x of sale price",
    "withinPlusMinus50": "Within +/-50%",
    "log10Rmse": "Log-scale RMSE",
    "medianRatio": "Median estimate-to-sale ratio",
    "medianFactorError": "Median error as a factor"
  },
  "guidance": [
    "Only the blind cut (each sale withheld from the engine's evidence) may be quoted as the accuracy figure.",
    "The recall cut scores the same sales with each one's own sale added to the engine's evidence; it is an upper bound, not the published accuracy.",
    "The incumbent cut is this site's own retired revision, scored blind on the same sales and compared against the blind cut; it is not a competitor's error rate.",
    "Do not compress the table into a single accuracy percentage.",
    "rankSlices measures ORDERING and is a separate question from the price cuts above; the two are scored on the same frozen test set but answer different things.",
    "revisions.unpublished lists 5.0 as disabled: that is its status for PRICING, where the shipped price stays 4.9. Revision 5.0 IS shipped for ORDERING - it supplies the rank in rankSlices by default - so read that row as price-disabled, not unused.",
    "The three rank slices are reported separately and deliberately not pooled: their median prices differ by orders of magnitude, so a pooled ordering would mostly measure which slice a name came from."
  ]
}