{
  "version": "https://jsonfeed.org/version/1.1",
  "title": "audit.domains corrections feed",
  "home_page_url": "https://audit.domains/accuracy/",
  "feed_url": "https://audit.domains/accuracy/corrections.json",
  "description": "Every audit.domains methodology revision and what it changed in the published error table - including measured regressions, when a committed back-test recorded one - plus the current live revision's largest misses against real sales.",
  "license": "https://audit.domains/terms/",
  "items": [
    {
      "id": "methodology-rev-4.14",
      "url": "https://audit.domains/method/",
      "title": "Methodology revision 4.14 (live): The personal-name pricing path carried unchanged onto the rev 4.13 base.",
      "content_text": "The personal-name pricing path carried unchanged onto the rev 4.13 base.\n\n- The personal-name model itself is identical to revision 4.12.\n- Renumbered because comparable selection underneath changed in revision 4.13: the revision stamp scopes the valuation cache, so a result priced under the new selection must never be served under an earlier stamp.",
      "date_published": "2026-09-30T03:45:05.466Z",
      "tags": [
        "methodology",
        "rev-4.14"
      ]
    },
    {
      "id": "methodology-rev-4.13",
      "url": "https://audit.domains/method/",
      "title": "Methodology revision 4.13: Comparables must trade in the same market as the name; the pricing model is unchanged from revision 4.11.",
      "content_text": "Comparables must trade in the same market as the name; the pricing model is unchanged from revision 4.11.\n\n- A recorded sale is cited, and priced from, only when it shares the name's TLD family (.com; .ai/.io/.co/.app/.dev; .net/.org/.info/.biz; country codes; other extensions), its class (keyword, brandable, numeric, acronym), its word count and a similar length. Digit names need the same length. The same name or word family in another zone always qualifies.\n- Before, a sale could be cited on spelling or a shared word alone: rank.ai at $200,000 appeared as a comparable for airank.si, and .de sales for .com names. Where too few peers exist, the report now shows fewer comparables and reads as thin evidence instead.\n\nMeasured on the frozen 218-domain holdout against revision 4.2: median APE: 99.8 -> 85.1 (improved); MAPE: 1513.5 -> 96.5 (improved); within 2x: 18.8 -> 29.4 (improved); within 3x: 26.1 -> 39.9 (improved).",
      "date_published": "2026-09-30T03:45:05.466Z",
      "tags": [
        "methodology",
        "rev-4.13"
      ]
    },
    {
      "id": "methodology-rev-4.12",
      "url": "https://audit.domains/method/",
      "title": "Methodology revision 4.12: The personal-name pricing path carried unchanged onto the rev 4.11 base.",
      "content_text": "The personal-name pricing path carried unchanged onto the rev 4.11 base.\n\n- The personal-name model itself (detection, popularity curve, cohort discipline, tier floors) is identical to revision 4.10.\n- Renumbered because name recognition underneath changed in revision 4.11: the revision stamp scopes the valuation cache, so a result priced under the wider vocabulary must never be served under an earlier stamp.",
      "date_published": "2026-09-30T03:45:05.466Z",
      "tags": [
        "methodology",
        "rev-4.12"
      ]
    },
    {
      "id": "methodology-rev-4.11",
      "url": "https://audit.domains/method/",
      "title": "Methodology revision 4.11: Long ordinary English words are recognized as words; the pricing model and its gating are unchanged from revision 4.9.",
      "content_text": "Long ordinary English words are recognized as words; the pricing model and its gating are unchanged from revision 4.9.\n\n- The name classifier recognized only a hand-built list of about 3,200 common words, so long ordinary words (engineering, agriculture, bookkeeping, neighborhood) and compounds built from them read as unrecognized strings and priced like random letters. It now also recognizes the 7+ letter words of the SCOWL common-word list (size 35, about 28,000 words).\n- Shorter words were left out on evidence: adding them split ordinary brandable names into spurious compounds and cost accuracy on bulk aftermarket sales, while every measured misreading was a long word.",
      "date_published": "2026-09-30T03:45:05.466Z",
      "tags": [
        "methodology",
        "rev-4.11"
      ]
    },
    {
      "id": "methodology-rev-4.10",
      "url": "https://audit.domains/method/",
      "title": "Methodology revision 4.10: The personal-name pricing path carried unchanged onto the rev 4.9 base.",
      "content_text": "The personal-name pricing path carried unchanged onto the rev 4.9 base.\n\n- The personal-name model itself (detection, popularity curve, cohort discipline, tier floors) is identical to revision 4.8.\n- Renumbered because the class gating underneath changed in revision 4.9: the revision stamp scopes the valuation cache, so a result priced under the corrected gating must never be served under a pre-correction stamp.",
      "date_published": "2026-09-30T03:45:05.466Z",
      "tags": [
        "methodology",
        "rev-4.10"
      ]
    },
    {
      "id": "methodology-rev-4.9",
      "url": "https://audit.domains/method/",
      "title": "Methodology revision 4.9: Corrected which names the protection caps and the comparables pricer apply to; the pricing model itself is unchanged from revision 4.7.",
      "content_text": "Corrected which names the protection caps and the comparables pricer apply to; the pricing model itself is unchanged from revision 4.7.\n\n- Pronounceable brandable names (measured phonotactic gate: vowel balance, no long consonant runs, an early vowel) are exempt from the garbage-cap ladder. The garbage scorer's random-string signals mis-read this class - measured against the real-sales corpus, 28.4% of genuine recorded transactions would have been capped to the junk floor; the gate salvages the class while keyboard-mash names and confident brand typo-squats remain capped by their own signals.\n- Eligibility for the comparables overlay is decided by name class instead of by the incoming value: a legitimate name a protection layer mis-capped to a junk price is repriced by the overlay, while genuine garbage-class strings never reach the comparable anchor at any value.\n- Hyphenated all-numeric names take the pure-digit class ceiling on their stripped digits (123-456 prices as the discounted form of 123456), closing the gap where they escaped every numeric ceiling by classifying as hyphenated.",
      "date_published": "2026-09-30T03:45:05.466Z",
      "tags": [
        "methodology",
        "rev-4.9"
      ]
    },
    {
      "id": "methodology-rev-4.8",
      "url": "https://audit.domains/method/",
      "title": "Methodology revision 4.8: The personal-name pricing path carried unchanged onto the indexed-comparables rev 4.7 base.",
      "content_text": "The personal-name pricing path carried unchanged onto the indexed-comparables rev 4.7 base.\n\n- The personal-name model itself (detection, popularity curve, cohort discipline, tier floors) is identical to revision 4.6.\n- Renumbered because the base comparable machinery underneath changed in revision 4.7: the revision stamp scopes the valuation cache, so an indexed-anchor result must never be served under a pre-indexing stamp.",
      "date_published": "2026-09-30T03:45:05.466Z",
      "tags": [
        "methodology",
        "rev-4.8"
      ]
    },
    {
      "id": "methodology-rev-4.7",
      "url": "https://audit.domains/method/",
      "title": "Methodology revision 4.7: Indexed comparables: every matched sale is adjusted to the appraised name before it prices anything, so a comparable contributes the market premium it demonstrated rather than its raw price level.",
      "content_text": "Indexed comparables: every matched sale is adjusted to the appraised name before it prices anything, so a comparable contributes the market premium it demonstrated rather than its raw price level.\n\n- Each comparable's price is indexed to the subject on the published structural model - adjusted for zone, length, keyword demand and brandability differences - and the anchor is the similarity- and recency-weighted median of those indexed prices. A same-name sale in another zone now translates through the documented zone offsets instead of exporting its raw dollar figure.\n- Premiums transfer 1:1 up to two-and-a-half orders of magnitude beyond typical (the p95 of recorded premiums); only headline-grade deviations damp, because the held-out back-test rejected every whole-range compression.\n- Recent sales carry more anchor weight than decades-old prints: a ten-year half-life with a floor, on the fixed corpus-edition clock - never the wall clock, so determinism is unchanged.\n- The outlier screen and the dispersion that drives the band width both moved to the indexed scale: raw price spread across zones is expected and explained; residual disagreement is genuine uncertainty.\n- Similarity adds a word-family channel (testing / tester / tests are one market), damps spelling coincidence between distinct dictionary words and across structural classes, and scales dictionary-class matches by word-tier proximity.\n- Elite and known-entity reports, whose price is owned by the protection layers, now carry display-only methodology-scored comparables instead of an empty evidence panel; each report row and the analyst note publish the per-sale indexed figure.",
      "date_published": "2026-09-30T03:45:05.466Z",
      "tags": [
        "methodology",
        "rev-4.7"
      ]
    },
    {
      "id": "methodology-rev-4.6",
      "url": "https://audit.domains/method/",
      "title": "Methodology revision 4.6: The rev 4.4 personal-name pricing path carried unchanged onto the recalibrated rev 4.5 base.",
      "content_text": "The rev 4.4 personal-name pricing path carried unchanged onto the recalibrated rev 4.5 base.\n\n- The personal-name model itself (detection, popularity curve, cohort discipline, tier floors) is identical to revision 4.4.\n- Renumbered because the base structural path underneath changed in revision 4.5: the revision stamp scopes the valuation cache, so a recalibrated result must never be served under a pre-recalibration stamp.",
      "date_published": "2026-09-30T03:45:05.466Z",
      "tags": [
        "methodology",
        "rev-4.6"
      ]
    },
    {
      "id": "methodology-rev-4.5",
      "url": "https://audit.domains/method/",
      "title": "Methodology revision 4.5: Recalibrated the structural price level and the comparable-anchor blend against the held-out back-test, correcting a systematic ~3x undervaluation of recorded sales.",
      "content_text": "Recalibrated the structural price level and the comparable-anchor blend against the held-out back-test, correcting a systematic ~3x undervaluation of recorded sales.\n\n- The structural intercept rises by 0.3 log10 (2x): the rev 4.3 structural prior predicted a median 30% of the recorded price on held-out sales; slope coefficients are unchanged so relative ranking between names is preserved.\n- The comparable anchor may now carry up to 95% of the final log-price (was 84%): on comp-rich names the residual structural share was dragging estimates roughly 2x under their own matched-sales anchor.\n- The correction is deliberately partial: a larger lift, a relaxed thin-evidence gate, and a full least-squares refit each scored better still on the holdout but were rejected - they inflate speculative thin-evidence compounds out of their golden-set band, or collapse the length slope toward the corpus median (a survivorship artifact of a notable-sales corpus).",
      "date_published": "2026-09-30T03:45:05.466Z",
      "tags": [
        "methodology",
        "rev-4.5"
      ]
    },
    {
      "id": "methodology-rev-4.4",
      "url": "https://audit.domains/method/",
      "title": "Methodology revision 4.4: Added a personal-name pricing path so recognized human first names are valued as the liquid end-user .com class they are, instead of being garbage-floored or priced off unrelated comparables.",
      "content_text": "Added a personal-name pricing path so recognized human first names are valued as the liquid end-user .com class they are, instead of being garbage-floored or priced off unrelated comparables.\n\n- Recognized first names (a bundled, versioned given-name dataset with a popularity weight) are detected as a distinct value class - they are not dictionary words or keyword terms, so the structural model and comp matcher could not read them.\n- First names are priced on a small log-space curve in popularity and length, disciplined by a personal-name comparable cohort when one exists and floored by popularity tier, so a recognized name can never collapse to a near-zero value (aubree.com no longer reads ~$15).\n- The comparable matcher restricts a first name's cohort to other recognized first names, so an incidental, unrelated sale is never shown as its comparable; when no first-name sale exists in the curated corpus the report discloses the thin evidence with a wide band.\n- Every domain outside the personal-name class is priced exactly as in revision 4.3 and keeps the 4.3 stamp; the curated back-test corpus contains no first-name sales, so held-out accuracy is unchanged from 4.3.",
      "date_published": "2026-09-30T03:45:05.466Z",
      "tags": [
        "methodology",
        "rev-4.4"
      ]
    },
    {
      "id": "methodology-rev-4.3",
      "url": "https://audit.domains/method/",
      "title": "Methodology revision 4.3: Recalibrated scoring and pricing on the curated real-sales corpus, with a comparable-sales anchor blended into every estimate.",
      "content_text": "Recalibrated scoring and pricing on the curated real-sales corpus, with a comparable-sales anchor blended into every estimate.\n\n- Length follows a convex scarcity curve - one-to-three character names price an order of magnitude above mid-length names - replacing the rev 4.2 linear curve.\n- Every TLD zone carries its own market multiplier and liquidity weight, fit from recorded sales, instead of a single shared discount.\n- Keyword demand and search footprint read advertiser cost-per-click and monthly search volume from the bundled CPC dataset, not aggregate engine buckets.\n- The structural estimate is blended in log space with a similarity-weighted anchor from matched real sales; a recorded prior sale of the domain itself anchors the estimate directly.\n- Back-tested with a generalization cut that excludes the subject's own recorded sale, so the held-out numbers measure prediction rather than recall.",
      "date_published": "2026-09-30T03:45:05.466Z",
      "tags": [
        "methodology",
        "rev-4.3"
      ]
    },
    {
      "id": "methodology-rev-4.2",
      "url": "https://audit.domains/method/",
      "title": "Methodology revision 4.2: The prior default: a linear length curve with keyword demand read from aggregate engine buckets.",
      "content_text": "The prior default: a linear length curve with keyword demand read from aggregate engine buckets.\n\n- Length scored linearly in character count - 100 * (12 - n) / 12, floored.\n- Keyword demand and search footprint came from the engine's aggregate market and technical buckets.\n- No similarity-weighted comparable-sales anchor; estimates leaned on the structural model alone.\n\nMeasured on the frozen 218-domain holdout against revision 4.13: median APE: 85.1 -> 99.8 (worse); MAPE: 96.5 -> 1513.5 (worse); within 2x: 29.4 -> 18.8 (worse); within 3x: 39.9 -> 26.1 (worse).",
      "date_published": "2026-09-30T03:45:05.466Z",
      "tags": [
        "methodology",
        "rev-4.2"
      ]
    },
    {
      "id": "measured-comparison-personal-name-holdout",
      "url": "https://audit.domains/method/",
      "title": "Personal-name path held-out check: no regression on the 174-domain holdout",
      "content_text": "median APE 80.98% -> 80.98%, within-2x 18.97% -> 18.97%. Predictions identical on holdout: true. The curated corpus has no first-name sales, so rev 4.4 = rev 4.3 on the holdout; the personal-name path is a strict, scoped improvement on the first-name cohort, which the corpus does not (yet) measure.",
      "date_published": "2026-07-05T06:36:13.751Z",
      "tags": [
        "methodology",
        "measured-comparison"
      ]
    },
    {
      "id": "residual-v0.io",
      "url": "https://audit.domains/predictions/",
      "title": "Largest miss: v0.io - model said $32,700, sold for $2,000 (overestimated, 16.35x)",
      "content_text": "With v0.io's own recorded sale withheld from the comparable pool, the deterministic model estimated $32,700; it sold for $2,000 (estimate/actual ratio 16.35). One row of the published 218-domain blind holdout (median error 85.1%, 29.36% within 2x), methodology revision 4.14.",
      "date_published": "2026-09-30T03:45:04.578Z",
      "tags": [
        "residual",
        "miss"
      ]
    },
    {
      "id": "residual-qrod.com",
      "url": "https://audit.domains/predictions/",
      "title": "Largest miss: qrod.com - model said $48,000, sold for $4,000 (overestimated, 12x)",
      "content_text": "With qrod.com's own recorded sale withheld from the comparable pool, the deterministic model estimated $48,000; it sold for $4,000 (estimate/actual ratio 12). One row of the published 218-domain blind holdout (median error 85.1%, 29.36% within 2x), methodology revision 4.14.",
      "date_published": "2026-09-30T03:45:04.578Z",
      "tags": [
        "residual",
        "miss"
      ]
    },
    {
      "id": "residual-irip.com",
      "url": "https://audit.domains/predictions/",
      "title": "Largest miss: irip.com - model said $38,600, sold for $3,800 (overestimated, 10.1579x)",
      "content_text": "With irip.com's own recorded sale withheld from the comparable pool, the deterministic model estimated $38,600; it sold for $3,800 (estimate/actual ratio 10.1579). One row of the published 218-domain blind holdout (median error 85.1%, 29.36% within 2x), methodology revision 4.14.",
      "date_published": "2026-09-30T03:45:04.578Z",
      "tags": [
        "residual",
        "miss"
      ]
    },
    {
      "id": "residual-jru.com",
      "url": "https://audit.domains/predictions/",
      "title": "Largest miss: jru.com - model said $126,000, sold for $17,000 (overestimated, 7.4118x)",
      "content_text": "With jru.com's own recorded sale withheld from the comparable pool, the deterministic model estimated $126,000; it sold for $17,000 (estimate/actual ratio 7.4118). One row of the published 218-domain blind holdout (median error 85.1%, 29.36% within 2x), methodology revision 4.14.",
      "date_published": "2026-09-30T03:45:04.578Z",
      "tags": [
        "residual",
        "miss"
      ]
    },
    {
      "id": "residual-2live.com",
      "url": "https://audit.domains/predictions/",
      "title": "Largest miss: 2live.com - model said $18,000, sold for $3,500 (overestimated, 5.1429x)",
      "content_text": "With 2live.com's own recorded sale withheld from the comparable pool, the deterministic model estimated $18,000; it sold for $3,500 (estimate/actual ratio 5.1429). One row of the published 218-domain blind holdout (median error 85.1%, 29.36% within 2x), methodology revision 4.14.",
      "date_published": "2026-09-30T03:45:04.578Z",
      "tags": [
        "residual",
        "miss"
      ]
    },
    {
      "id": "residual-rocket.com",
      "url": "https://audit.domains/predictions/",
      "title": "Largest miss: rocket.com - model said $69,300, sold for $14,000,000 (underestimated, 0.005x)",
      "content_text": "With rocket.com's own recorded sale withheld from the comparable pool, the deterministic model estimated $69,300; it sold for $14,000,000 (estimate/actual ratio 0.005). One row of the published 218-domain blind holdout (median error 85.1%, 29.36% within 2x), methodology revision 4.14.",
      "date_published": "2026-09-30T03:45:04.578Z",
      "tags": [
        "residual",
        "miss"
      ]
    },
    {
      "id": "residual-tyres.co.uk",
      "url": "https://audit.domains/predictions/",
      "title": "Largest miss: tyres.co.uk - model said $1,650, sold for $375,000 (underestimated, 0.0044x)",
      "content_text": "With tyres.co.uk's own recorded sale withheld from the comparable pool, the deterministic model estimated $1,650; it sold for $375,000 (estimate/actual ratio 0.0044). One row of the published 218-domain blind holdout (median error 85.1%, 29.36% within 2x), methodology revision 4.14.",
      "date_published": "2026-09-30T03:45:04.578Z",
      "tags": [
        "residual",
        "miss"
      ]
    },
    {
      "id": "residual-fuse.com",
      "url": "https://audit.domains/predictions/",
      "title": "Largest miss: fuse.com - model said $7,150, sold for $2,129,509 (underestimated, 0.0034x)",
      "content_text": "With fuse.com's own recorded sale withheld from the comparable pool, the deterministic model estimated $7,150; it sold for $2,129,509 (estimate/actual ratio 0.0034). One row of the published 218-domain blind holdout (median error 85.1%, 29.36% within 2x), methodology revision 4.14.",
      "date_published": "2026-09-30T03:45:04.578Z",
      "tags": [
        "residual",
        "miss"
      ]
    },
    {
      "id": "residual-intelligence.xyz",
      "url": "https://audit.domains/predictions/",
      "title": "Largest miss: intelligence.xyz - model said $137, sold for $150,000 (underestimated, 0.0009x)",
      "content_text": "With intelligence.xyz's own recorded sale withheld from the comparable pool, the deterministic model estimated $137; it sold for $150,000 (estimate/actual ratio 0.0009). One row of the published 218-domain blind holdout (median error 85.1%, 29.36% within 2x), methodology revision 4.14.",
      "date_published": "2026-09-30T03:45:04.578Z",
      "tags": [
        "residual",
        "miss"
      ]
    },
    {
      "id": "residual-icon.com",
      "url": "https://audit.domains/predictions/",
      "title": "Largest miss: icon.com - model said $6,590, sold for $12,000,000 (underestimated, 0.0005x)",
      "content_text": "With icon.com's own recorded sale withheld from the comparable pool, the deterministic model estimated $6,590; it sold for $12,000,000 (estimate/actual ratio 0.0005). One row of the published 218-domain blind holdout (median error 85.1%, 29.36% within 2x), methodology revision 4.14.",
      "date_published": "2026-09-30T03:45:04.578Z",
      "tags": [
        "residual",
        "miss"
      ]
    }
  ]
}