Reality Gap methodology

Data sources

Sold prices: HM Land Registry Price Paid Data (standard residential sales only, category A; deleted records excluded), trailing 12 months from the computation date. Asking prices: TrueDeed active and under-offer sale listings at the computation date. Both sides are medians — never means — so single outliers cannot move a figure.

The two evidence tiers — never blended

Matched pairs track the same property from listing to confirmed Land Registry completion. High confidence, small sample: the gap is measured on identical properties. Area medians compare the median asking price of what is for sale now against the median sold price of what completed in the trailing 12 months — two different populations, larger sample, cruder signal. Every published figure is labelled with its tier and the two are never combined into one number.

Suppression thresholds (disclosed, exact)

A cell publishes only when its sample clears these thresholds; otherwise it is suppressed and shown as suppressed — never estimated, interpolated, or padded:

• Area-median cells: at least 20 asking prices AND at least 100 sold transactions.
• Matched-pair cells: at least 10 confirmed listing-to-sale pairs.

Suppressed cells are also excluded from the Truth League rankings and from the CSV download.

The no-cherry-picking rule

Every computed cell that clears its threshold is published — none are hand-picked, and none are held back because they are unflattering to TrueDeed or to any area. The refresh recomputes every national and district cell in one pass; there is no editorial step between computation and publication.

Known lag: Land Registry registration

Completions typically take around 3 months to appear in Price Paid Data, so the sold side of each edition is missing the most recent sales. This biases nothing systematically, but it means a quarter’s figures firm up over time — which is one reason each edition is recomputed rather than frozen on first publication.

Periods and versioning

Editions are quarterly, named YYYY-Qn (e.g. 2026-Q2). Every published cell carries methodology_version (currently 1); if the computation ever changes, the version increments and this page documents the change — figures computed under different versions are never mixed silently.