{"live":{"label":"LIVE","chained_since":"2026-07-05","chain":{"count":0,"head_hash":"0000000000000000000000000000000000000000000000000000000000000000","last_snapshot_ts":null},"anchor":"The chain head (docs/receipts-chain-head.txt) is committed to a public GitHub repository on every refresh; the commit history is the independent timestamp witness that each snapshot existed before the sale it is later settled against.","settlement":"Live receipts settle against HM Land Registry Price Paid Data as sales complete (typically 3-6 months after a snapshot). Settled LIVE results are published here as they accrue — misses included.","settled":{"label":"LIVE-SETTLED","n":0,"method":"Each settled row compares the estimate Nestraq pre-committed to its public hash chain with the price the home actually sold for afterwards (HM Land Registry Price Paid Data). Estimates were locked before the sales were knowable; every miss is included.","match_rule":"A snapshot settles only against a UNIQUE candidate: exactly one Land Registry sale in the same postcode district, completing 2-10 months after the snapshot, same property type where both are known, within 25% of the asking price at snapshot time. Zero candidates stay unsettled; two or more are ambiguous and dropped. High precision over recall - the match rate is published alongside the results.","method_version":"1","attribution":"Contains HM Land Registry data © Crown copyright and database right. Licensed under the Open Government Licence v3.0."}},"backtest":{"label":"BACKTEST","backtest_year":2025,"n":0,"method":"Backtest over HM Land Registry Price Paid Data (England & Wales). For each home sold in the backtest year we rebuilt the estimate Nestraq's free-data method would have produced the day before the sale, using ONLY transactions completed strictly before that sale date (pre-sale-data cutoff enforced in code): same-postcode comparables, same-type stratification, log-MAD outlier trim, median point estimate with a 10th–90th percentile band, and the live method's confidence formula — only estimates that would have qualified as medium/high confidence are settled (low-confidence estimates are never published live either). Each estimate is then settled against the actual sold price. All misses are included; no hand-picked examples. No time-indexing is applied in the backtest, so the live method's error is, if anything, overstated here.","method_version":"1","cutoff_rule":"Every comparable used in a backtest estimate completed strictly BEFORE the subject's sale date. Enforced twice in code: a date filter, then a fail-closed assertion that aborts the estimate if any on-or-after-sale data slips through.","attribution":"Contains HM Land Registry data © Crown copyright and database right. Licensed under the Open Government Licence v3.0."},"provenance":{"method":"deterministic — no LLM anywhere in the numbers path","publication_rule":"Distributions with all misses included; regions below the n gate are omitted entirely; no per-property results are published (ASA rule — no hero anecdotes).","attribution":"Contains HM Land Registry data © Crown copyright and database right. Licensed under the Open Government Licence v3.0.","not_a_valuation":"Estimates are indicative statistical figures from public sold-price data - not a survey, mortgage valuation or advice."}}