Methodology
The data behind PropDetect, and how it stays calibrated
Trust in a figure starts with knowing where it came from. This page lists every external data source an analysis draws on, the scale of the system that produces it, and the calibration loops that keep the numbers honest as markets move.
Every number on this page was measured, from the codebase or from the production database, on the date shown at the bottom. Where something has not been measured, this site does not state it as if it had been.
The data feeds
A single analysis draws on eight classes of external data, each doing a distinct job:
- HM Land Registry sold price data, supplied via PropertyData: the completed transactions behind every valuation comparable and the price per square foot cross-check.
- Lettings market data via PropertyData: the rental comparables behind the rent estimate, matched on bedroom count and property type.
- Local Housing Allowance rates: the benchmark behind social housing analysis.
- Floor area records: the size data that makes price per square foot comparisons possible.
- The EPC register for England and Wales: energy ratings, floor areas, and the fabric detail behind the energy works costing.
- The Scottish EPC register: the same coverage for Scottish properties, which most UK tools simply do not have.
- Listing data from six portals: Rightmove, Zoopla, OnTheMarket, and the Scottish centres ASPC, TSPC and PSPC, including photographs, floor plans and descriptions.
- Geographic services: postcode geocoding and location resolution, plus Article 4 direction data for HMO analysis.
The scale behind one analysis
The analysis pipeline is 78,758 lines of code within a 275,000 line system, and a single property run moves through 34 distinct AI reasoning steps: reading every photograph, assigning rooms from the floor plan, parsing the EPC and any brochure, judging condition element by element, and writing the reasoning you see on the report.
Alongside the AI steps, each analysis pulls up to 100 sold comparables and up to 100 rental comparables, resolves the EPC, geocodes the property, and cross-references Article 4 data where relevant. Measured across the production base, the average analysis carries 99 sold comparables into the ranking stage. That is why an analysis takes about three minutes rather than three seconds: it is doing the reading.
As of 4 September 2026 the system has produced 3,023 property analyses across 3,977 completed runs, spanning England, Wales and Scotland.
How the refurbishment rates stay calibrated
The rate base did not start with software. It was built from real builder quotes collected over a six month effort across the country, deliberately spread across regions so the starting rates reflected what trades actually charge in each part of the country rather than one national figure dressed up as local. Live usage has been feeding it ever since.
Refurbishment rates are regional. The production base currently carries regional cost calibration across 493 postcode districts, so a bathroom in Middlesbrough does not price like a bathroom in Brighton. The anchor for those rates is real builder quotes from real projects: when a customer's builder prices the same property, that comparison feeds the calibration, which is how the estimates track what builders actually charge.
The second loop is usage. Every figure on a report is editable, and when users correct a line, a rent, a room cost, a valuation input, those corrections are recorded and reviewed against the estimates. The system's starting numbers improve because people who know their patch push back on them, which is precisely what a calibration loop should want.
On top of both loops sits a monthly review: rates are checked against RICS cost data and a wider market view of material and labour costs, so a move in the price of materials or a shift in trade day rates reaches the estimates within weeks rather than quarters.
How the finance assumptions stay current
Bridging rates, buy-to-let rates and the finance defaults in every appraisal are anchored to the Bank of England base rate, so when the base rate moves, the cost of money in new analyses moves with it rather than reflecting the market of six months ago. Regional and price-band defaults layer on top, and everything remains editable per deal, because your broker's terms beat any default.
What this buys you in practice
The point of all of this is a specific promise: no figure on a PropDetect report is a national average wearing a local costume. The valuation is built from sales near the property. The refurbishment is priced at rates for that region, anchored to real quotes. The rent comes from lettings of the same size and type nearby. The finance reflects the current base rate. And when any of that evidence is thin, the report says so and widens its ranges rather than manufacturing confidence.
The limits, stated plainly
- Land Registry data lags completions by weeks to a few months. Every sold-price product inherits this, including ours.
- Rental comparables are asking rents from the lettings market, plus let-agreed listings where available. Asking rents can sit above achieved rents in some patches, and no tool built on portal data can fully see that gap.
- Regional calibration is only as deep as the region's data. 493 postcode districts carry calibration today; thin districts lean on wider regional rates, and the confidence score reflects it.
- The quote and usage loops run continuously, and the RICS-anchored market review runs monthly. Where we have not measured something, including a formal accuracy distribution, this page does not claim it.
Frequently asked questions
What data sources does PropDetect use?
HM Land Registry sold prices and lettings market data via PropertyData, Local Housing Allowance rates, floor area records, the EPC registers for both England and Wales and Scotland, listing data from Rightmove, Zoopla, OnTheMarket, ASPC, TSPC and PSPC, and geographic services including Article 4 direction data.
How big is the system behind an analysis?
The analysis pipeline is 78,758 lines of code within a 275,000 line system, and each property runs through 34 distinct AI reasoning steps plus up to 200 comparable lookups. As of 4 September 2026 it has produced 3,023 property analyses.
How are the refurbishment rates kept accurate?
Three mechanisms: regional rates anchored to real builder quotes, originally collected over a six month effort across the country and growing with use, a monthly review against RICS cost data and a market view of material and labour costs, and user corrections, since every figure is editable and edits are reviewed against the original estimates.
Do the finance assumptions track interest rates?
Yes. Bridging and buy-to-let defaults are anchored to the Bank of England base rate, so new analyses reflect the current cost of money. Everything stays editable per deal.
Last reviewed 4 September 2026. Claims on this page trace to our internal evidence register.