Why Offrly publishes accuracy in the open
Almost no free UK valuation tool publishes a real accuracy figure. The ones that do tend to cite proprietary back-tests no one can verify. Offrly's stance is the opposite — every published valuation is frozen at quarter-end and auto-joined to the eventual HM Land Registry sold price for the same postcode, then fed into a public accuracy report. The figures aren't a marketing claim; they're a prediction-vs-outcome record you can read, with no human in the loop.
- Public accuracy report at /accuracy-report.
- Every figure traces to a real Offrly prediction and a real HM Land Registry sold price.
- Predictions are made strictly before the eventual sold date — no leakage.
- Pairing is fully automated — no admin chooses which pairs make the cut.
- Rentals are not in the report — there's no public sold-rent register to pair against.
The problem with "free valuation" accuracy
Most free UK valuation tools either:
- Don't publish an accuracy figure at all, or
- Cite a proprietary back-test ("our model is 93% accurate") that no third party can audit.
The second is worse than the first. A back-test on the model's own training period inflates the result. A back-test on data the model has already seen tells you nothing about how it performs on next month's homes. And a self-reported number with no auditable source can be set to whatever marketing wants.
What Offrly does instead
Every Offrly sale valuation is logged with its postcode, predicted price and timestamp. At quarter-end the BUY rows are frozen to an immutable snapshot, then six months later auto-joined to HM Land Registry sales on the same full postcode — sold date strictly after the valuation date. There is no human in the loop. The lifetime aggregate is published at /accuracy-report; the per-quarter sealed snapshot is at /benchmark.
The figures published:
- Hit rate within ±10% — the share of Offrly estimates that landed within 10% of the actual sold price.
- Hit rate within ±20%.
- Median absolute percentage error (MdAPE).
- Mean absolute error in £ (MAE).
- Per-property-type breakdowns (detached, semi-detached, terraced, flat).
Each figure traces to a real Offrly prediction made strictly before the eventual sold date. By construction, predictions never see future sold prices.
Why HM Land Registry?
HM Land Registry Price Paid Data is the authoritative record of every residential property transaction in England and Wales, published monthly under the Open Government Licence. It's the only licence-free primary-source sold-price dataset for those nations. Using it as the ground-truth means the accuracy claim is auditable by anyone with an internet connection.
Scotland uses Registers of Scotland; Northern Ireland uses Land & Property Services. Those datasets aren't yet integrated into the accuracy report — the auto-joined pair set is England-and-Wales for now.
What's NOT in the accuracy report
- Rentals. No public sold-rent register exists, so we don't publish a rental accuracy claim.
- Probate, mortgage and insurance valuations. Those are regulated processes; Offrly's estimate is indicative market guidance, not a binding figure. The accuracy report measures market-pricing accuracy, not regulated-valuation equivalence.
- Off-market or pocket-listed sales. HMLR captures completed transactions only; off-market chats don't show up.
And the practical implication
If you're choosing between Offrly and another free tool: read the accuracy report. If the other tool publishes one against an authoritative sold-price register, compare them honestly. If it doesn't, that's a meaningful signal in itself.
FAQ: Offrly vs
How is the accuracy figure computed?
Every BUY valuation Offrly issues in a given quarter is frozen to an immutable snapshot at quarter-end. Six months later, when HM Land Registry has published the matching sold-price data, each frozen valuation is auto-joined to sales on the same full postcode that completed after the valuation date. The metrics on /accuracy-report aggregate every quarterly snapshot; /benchmark shows a single quarter at a time.
What happens when a postcode has multiple sales in the matching window?
A full UK postcode unit holds 10–15 properties, so a single frozen valuation can sometimes match more than one HMLR sale. We resolve this with an explicit, documented rule: pick the sale whose price is closest to the prediction. The rule biases each pair toward the model and we surface that on the page rather than hide it. We chose transparency over an admin choosing which pair counts.
Can the headline figure get worse?
Yes. As fresh HMLR data lands and new quarterly snapshots resolve, the headline figure can move in either direction. We don't cherry-pick to keep the number flattering — the page recomputes on every load over every snapshot.
What about rentals?
Rentals are NOT in the accuracy report. There is no public sold-rent register equivalent for rentals — HM Land Registry covers sale prices only, in England and Wales. Publishing a rental accuracy claim without a public ground-truth would be a marketing claim, not an accuracy claim, so we don't.
Are there any internal accuracy claims Offrly makes that aren't published?
No. The figures on /accuracy-report and /benchmark are the only public accuracy claims Offrly makes. We don't cite model internals, proprietary benchmarks or cherry-picked cases.