Methods

How we collect, compare and publish UK property data.

What the system is designed to do

Property evidence arrives in incompatible shapes: an official sale, a certificate, a planning entity, an advertised rent, a map result or a dated calendar. Upgraded Intelligence preserves what each source actually proves, then makes the records comparable enough to study together.

We make the available evidence useful while keeping its origin, date and limits visible. The data cannot support one definitive number for every property.

  1. DiscoverSearch defined geographic areas and record the source, search boundary and time of observation.
  2. NormaliseStandardise dates, currencies, coordinates, postcodes, property attributes and geographic identifiers without rewriting the underlying evidence.
  3. Resolve identityCompare location, imagery and property attributes. A possible match remains a possible match until enough independent signals agree.
  4. Observe changeKeep dated price and availability states so the market can be studied as a sequence, not a one-off screenshot.
  5. Publish aggregatesRemove operational detail, group observations to an appropriate geography and state the sample, date and definition beside the result.

Market replay

Market replay shows monthly official rents and house prices on their published evidence dates. Missing readings remain gaps. The observation view uses dated release counts, which describe observed rental records and can change as collection coverage changes.

Observation points must use the current collection method and have at least 90% collection completion for the selected area. A change calculation also requires at least 20 records in the baseline and no more than five percentage points of difference in completion. A change in observed records does not establish a change in the number of homes or the supply available to rent.

Similar places

Similar places compares areas using their positions in the UK distribution for the same evidence period of each selected measure. The rents view uses average rent and annual rent growth. The sale prices view uses average price, annual growth and two year growth. The overall view uses both groups where the selected area has the required evidence.

Each included measure has equal weight. Areas are ordered by the root mean square difference between their percentile ranks. A candidate must have every included measure for the same evidence periods as the selected area. A country filter changes the candidate places while keeping the UK benchmark. Missing values are not estimated.

Similarity describes the published area statistics. It does not measure neighbourhood quality or establish the prospects of an individual property.

Rent budget map

The rent budget map uses the bedroom rent averages already published in the UK Property Atlas. An area qualifies when its published monthly average for the chosen bedroom count is at or below the entered budget. The four bedroom group includes homes with four or more bedrooms.

The figures cover new and existing tenancies in the official rent series. Missing readings stay unavailable. The map colours describe area averages. A home currently advertised in that area can have a different rent. Nearby comparisons use distances between area centres.

Download the bedroom figures with their evidence month and source. The entered budget changes the comparison and leaves the underlying figures intact.

Asking rent cuts

The asking rent cuts explorer compares the prices on the same Rightmove advert exactly 30 calendar days apart, using UTC dates. We take the last saved card on the day we first recorded the advert and the last saved card 30 days later. We check each card against the advert identifier and the saved search page. Both must have a comparable monthly asking rent in pounds. We exclude commercial and student search channels and cards identified as rooms, shares, holiday or short lets.

We looked at adverts first recorded from 15 September 2026 to 16 October 2026. Some adverts may already have been live when we first saw them. We count each advert once. If a starting advert has no comparable price on day 30, it stays in the starting count. Its price change is unknown. A cut means the rent was lower at the second check. The typical cut is the middle reduction among adverts that got cheaper.

Area results use adverts whose recorded locations fall in the same local authority or region on both dates. Bedroom results use adverts with the same recorded bedroom count on both dates. The overall result also includes comparable adverts whose bedroom count or location was unclear. A result needs at least 10 usable comparisons. The typical cut needs at least 5 lower prices. The figures describe the prices people advertised and leave the rent agreed with a tenant unknown.

The statistical range is a 95% Wilson interval for the share of comparable adverts with a cut. It reflects the precision of the count within the sample. Which adverts we were able to capture twice also influences the results.

Read the first edition for results from the smaller October sample.

The advert price replay compares positive GBP asking prices for the same advert, transaction type and price period. It retains the latest eligible capture on an observed day and excludes price thresholds and incomplete histories. The current bounded selection starts with up to 1,000 adverts from the latest 5,000 retained observations and examines up to 90 days of history.

A fixed panel contains exactly the same adverts on every displayed date. Each advert starts at an index of 100 against its own first panel price; the chart shows the median of those individual indices. That measure differs from the median asking price in pounds. Missing dates remain absent. Wider first to latest summaries can have different observation intervals and remain separate from the fixed panel. Small groups below ten are withheld.

The sample is an observed group rather than a national market index. Asking prices do not establish achieved rents, completed sales, tenancies or changes in housing supply.

The publication outcome note examines the latest 64 recorded nonpublic observations with retained observations of the same adverts from other dates. Its 101 dated observations cover 64 adverts from 20 September to 8 November 2026. Eighteen adverts link an earlier observed public profile to a later unavailable record, with 10 to 49 calendar days between observation dates. That gap does not measure time on market. The reason for the later status is unknown, and the pairs do not establish daily status or a completed transaction. The extract does not provide a complete collection error history. Read the dated evidence.

Market discoveries

The market discoveries comparison pairs positive average rents from the same evidence month. The default window is 1.5%, calculated as the absolute rent difference divided by the mean of the two readings. Sale prices must also share an evidence month. Pairs are ordered by their price ratio, then rent difference and area code. Wider windows remain explicit.

The growth view keeps rising, falling and unchanged readings distinct. Rent and house price growth can cover different months, which remain visible. These area averages describe different mixes of properties and do not compare the returns of matched homes.

Historical sale prices

The housing time machine keeps UK House Price Index averages separate from recorded Price Paid transactions. The index adjusts for the mix of homes sold. Annual Price Paid medians and quartiles describe the transactions recorded in that transfer year, including Standard and Additional records. Prices are nominal, current years are partial and late registrations can revise the results.

Energy assessments

The home energy atlas uses the government annual tables published on 29 October 2026. The publisher retains the latest valid assessment per dwelling within April 2016 to March 2026. These assessments cover a selected part of the housing stock and can predate subsequent work.

The recorded energy changes comparison instead links usable domestic certificates by the same official property reference. A home must have at least two distinct usable dates. Conflicting grades on the same date are excluded. The first and latest grades define its recorded transition, and the latest certificate supplies its area. Homes with no latest area remain in the national group. Small groups below ten are withheld.

The assessment interval differs between homes. Changes in grades can reflect changes in a building, assessment methods or recorded information, so this comparison does not establish the cause of an improvement or represent all homes.

Recorded heating, wall and roof assessments use the official five level scale from very poor to very good. Their comparisons keep the same first and latest certificate dates as the band analysis. Both endpoint feature ratings must be known. Missing, not applicable and conflicting endpoint values remain explicit, and the endpoint dates are never shifted to obtain a favourable comparison. Feature assessments can be modelled or assumed and do not prove that insulation or equipment was installed.

Official property histories

The property history explorer joins sales to energy assessments only through source supplied official property references. Its current sales associations come from the September 2026 HM Land Registry transaction to property lookup. That monthly file covers newly published Price Paid records and does not backfill the complete sale history of every home.

A transaction linked to several property references retains its whole recorded consideration and all its associations. The amount is not allocated to an individual property. Missing, conflicting or withdrawn price records remain explicit. Energy assessments retain their recorded dates and certificate types; conflicting domestic grades on the same date remain visible and break the grade chart.

The public files contain licensed official identifiers and non address facts. They exclude address fields and private advert joins. A missing property or assessment means it is absent from this public cohort, not that the home or certificate does not exist.

Stay price comparisons

The stay price explorer compares saved stay quotes for an identical listing, arrival and departure dates, currency, guest counts and request settings. Only a generated impression identifier is ignored. The primary comparison requires explicit adult, child, infant and pet counts and retains the latest saved quote on each UTC date.

Every request in the primary comparison is for three nights and two adults, with no children, infants or pets. We retain the original quoted stay total and calculate its rounded average per night. Multiplying that rounded average by three can differ from the original total. Different requests can concern the same listing, so request counts are not counts of distinct properties.

The archive review covers 278,786 retained observations across 185,099 listings. The primary comparison includes 5,225 original verified quotes for 2,554 requests with at least two capture days before arrival. Single day contexts, unavailable responses, failed requests and missing context stay outside the change comparison.

Primary directions compare quoted stay totals. Nightly amounts are rounded averages of those totals. The first and latest included quotes were one to six days apart, with a median gap of two days. The explorer shows thirty illustrated histories; its headline counts and complete downloads cover the full primary comparison.

The earlier displayed price dataset has 11,102 original whole pound totals for 5,297 repeated saved requests. Each original main price section must explicitly show one unambiguous GBP total on the same price basis. Unknown guest fields remain unknown. These results are reported separately from the primary quoted totals.

Read the full report, download the primary CSV or the earlier displayed price CSV. The original illustrative edition remains available.

Observed availability

The availability calendar uses the latest retained calendar state per listing and target date, with source observations no older than five days. The existing 150 metre spatial grouping convention forms observation groups rather than resolved homes. A date needs at least ten qualifying groups.

The 30 date summary starts on the observation date and includes exactly 30 target dates. The 90 date summary includes exactly 90. Each is the mean of its qualifying daily unavailable shares. Daily samples can change across the horizon. An unavailable date does not establish a booking, and price coverage is qualified separately. Older dated editions retain their original windows and evidence.

Short let market observations

The public map shows aggregate listing records placed into sampled map cells. A record is evidence that a listing version was observed in a defined search, not proof of a unique home. The same property can be advertised more than once or appear through more than one channel.

Dated calendar observations preserve the price and availability state shown at collection time. They allow observed calendar change to be studied, but the cause of a state change is not supplied by the marketplace.

Unavailable does not automatically mean booked.

A night may have been booked, blocked by the operator or made unavailable for another reason. We describe the observed state and only use “booked” where the method and evidence justify it.

Observation frame and release controls

Collection uses a fixed national frame of 3,847 sampled cells. Public consumer marketplace result pages and listing pages are observed against that frame, then retained only when the record has enough current location evidence for the published geography. The map does not expose collection endpoints, account details or source-specific operational controls.

Each approved national release reports completed, capped and total cells together with geolocation, allocation and grid-completion rates. Definitions remain fixed within a release. Local change is withheld when coverage moves too far between the compared dates, because a collection shift can otherwise look like a market shift.

Provider identities and operational collection details are withheld to protect source security. Coverage, completion, allocation and uncertainty measures remain public so the effect of those limits can still be inspected.

Rates and prices

Advertised prices can move with date, party size, length of stay, fees and platform presentation. A quoted nightly rate is therefore tied to the search configuration and observation date. Advertised rent is not the same thing as an achieved rent; official achieved-rent series are kept distinct.

Geography

Search grids are a collection tool, not a claim about neighbourhood boundaries. Public area names describe the labels attached to sampled cells. Official postcode and statistical geographies are used where a formal boundary is required.

Area estimates can use verified approximate advert positions and postcode centres where a record has no existing coordinates. These positions do not establish an exact address. Comparisons are withheld when a change in location evidence affects the measurement.

UK Property Atlas

Atlas profiles use the Office for National Statistics December 2025 local authority boundaries and the 4 November 2026 public observation release. House prices, rents and observations retain their own evidence dates. Yield uses rent and house prices from the same month. Each area profile shows the period beside its measures.

Prices, rents and the gross-yield proxy

Average prices and annual price change come from the mix-adjusted UK House Price Index. Average rents and annual rent change come from the Price Index of Private Rents, a modelled stock measure covering new and existing tenancies. The gross-yield proxy is calculated as twelve times the area-average monthly rent divided by the area-average sale price.

The proxy is not a property-level investment return.

The rent and price populations are not matched to the same homes. The calculation does not deduct voids, finance, management, maintenance, tax or transaction costs.

Sales activity

The twelve-month registered-sales figure sums the records present in the latest twelve published HPI volume months. Those recent volumes are provisional and rise as registrations complete, so the Atlas presents them as source evidence rather than a final estimate of market activity. It does not use them to infer momentum.

Two-year sale-price growth compares the latest mix-adjusted HPI average price with the same area 24 months earlier. It uses the price-index period shown in the profile and remains subject to the publisher's normal revisions.

Percentiles and benchmarks

Percentiles are calculated independently for each measure across local authorities with a non-missing comparable value. Country and UK benchmarks are medians, not weighted national averages. A high percentile is a position in a distribution, not a positive rating.

Allocating observation cells

The national observation grid is coarser than many local authorities. Atlas roll-ups approximate each cell's boundary share using an evenly spaced sub-grid, preserve the cell total with integer allocation, and use a published district centroid for very small or narrow coastal fragments that fall between sample points. Density divides the allocated records by official boundary area. The resulting figure is an area-level observation measure, not a precise dwelling count.

A correction on 20 September 2026 fixes the Atlas data description. It previously said records were assigned by cell centre; the published calculation uses sampled boundary shares. Figures are unchanged.

Deliberate missingness

England and Wales rents are published at local-authority level. Scotland and Northern Ireland use rental geographies that do not map directly onto the Atlas boundary set, so their local-authority rent and yield cells remain blank. They are not filled with regional averages.

Identity resolution

A marketplace advert, a UPRN and a sale record are different objects. Joining them too quickly creates confident-looking errors. Candidate matches are scored using available evidence such as location proximity, consistent property attributes and image similarity. Ambiguous candidates remain unjoined.

Published aggregates may include listing versions before identity resolution. Where that is true, the metric is labelled as adverts, listings or records rather than homes or properties.

Published technical standards

The general method is supported by versioned notes for decisions that materially affect interpretation.

Testing simulation results

The housing stress atlas compares conditional model responses across England and Wales. Its time windows use elapsed model months. It retains missing and undefined comparisons and reports variation across paired runs. Separate comparisons across three population sizes assess sensitivity within the fixed model and do not establish convergence or forecast accuracy.

Our housing model size study compares 64 completed simulations across four populations through one million synthetic households. It reports paired responses, uncertainty and the continuing drift of unshocked baselines. A larger simulation does not by itself establish empirical calibration or a reliable market forecast.

The mortgage repayment quote study examines how a higher quoted rate changes hypothetical repayments across England and Wales. It keeps assumptions about loan amounts, incomes and terms separate, shows missing source weight and distinguishes the mean of individual payment ratios from the ratio of payment and income totals.

Known limitations

Coverage is uneven

Collection depth varies by place, source and date. Sparse areas on the map are left sparse. They should not be interpreted as proof that no market exists.

Listings can duplicate

One home may generate several advert versions or appear on several websites. Until identity is resolved, aggregate record counts are not unique-property counts.

Source interfaces change

Marketplaces can change search behaviour, availability displays and result limits. Collection methods are reviewed when a source changes and historical definitions are not silently rewritten.

Official data has publication lag

Transactions, indices and administrative records are updated on the publisher's timetable. The coverage ledger states the last loaded period rather than implying real-time completeness.

Corrections and reproducibility

Area name correction, 9 November 2026

We corrected 17 local authority names in the repeat EPC comparisons and corrected or filled area names in the property histories using their official area codes. The counts and 8 November evidence cutoff did not change. The corrected EPC comparison, feature comparison and property histories are published as a 9 November edition. The original dated files remain available.

Material public figures carry a snapshot or release date. If a source correction changes a published conclusion, the affected output receives a dated explanation. Immutable release files are not silently overwritten. Data extracts are published with selected research where disclosure is safe and licensing allows it.

The research governance policy explains editorial responsibility, commercial separation and correction handling. Questions about a definition or correction can be sent to hello@theupgradeauthority.co.uk.