Are model outputs guarantees?
No. The published figures describe model performance on the stated evaluation data. Each estimate has limits and should be read with its evidence, date, version and confidence information.
Model evidence
Judge the numbers, not the claim.
Every figure on this page is published in full in our documentation, including the segments where the model performs worst. That disclosure is the point — a validation you cannot audit is a badge, not evidence.
7.6%
Median error (MdAPE)
84%
Within 20% (PPE20)
16%
Land outside 20% — published
228,443
Held-out sales tested
v7
Served model · Jul 2026
Median error (MdAPE) by price band
Under £100k21.9%
£100–150k11.2%
£150–175k8.9%
£175–250k7.4%
£250–400k6.2%
£400–700k7.1%
£700k–5m10.0%
Sub-£100k stock is a known weakness and we say so. At those prices condition drives the sale more than any data can see, so those cases are flagged low-confidence, carry a deliberately wide range, and are routed for human review rather than presented as a firm number.
Where every prediction lands
Within 10% of the achieved price61%
Between 10% and 20%23%
Outside 20% — we publish this16%
Open, verifiable sources
HM Land RegistryUK House Price IndexEPC registerONS / ASHE indicesPlanning registersFlood & IMDLHA rates
Comparables are HPI-indexed to the valuation date by district and property type, so a valuation is dated and current rather than quietly ageing as the market moves.
Held-out evaluation
Scored only on sales the model never trained on, so an accuracy claim means what it says rather than measuring the model against its own homework.
Expected loss, not optimism
Every modelled downside is probability-weighted and subtracted so the resulting risk-adjusted profit can be reviewed alongside its assumptions.
Explainable by construction
Every prediction attributes to specific input features. No black box: the inputs and limitations can be reviewed by the people already involved.
Versioned and documented
Data lineage, retraining controls, acceptance gates and audit trails are held as controlled documentation.
Maths where maths belongs
SDLT, stress scenarios and refinance arithmetic are deterministic and exact to the penny. We do not approximate with a model what a formula answers precisely.
Weaknesses published
Sub-£100k stock is where the model is weakest. It says so on the page, in the report, and on the estimate itself.
Helpful context
How to read the evidence
Can an earlier result be understood later?
The workflow is designed to retain the model version, relevant inputs and evidence available at the time, so a historic output can be inspected rather than treated as an unexplained score.
Great deals deserve independent evidence.
Run a real address through the model, see the working, and decide for yourself. Two quick checks are on us.
Join the waitlist
We are onboarding in waves ahead of the public launch. Tell us which side of the table you sit on and we will start you in the right place.
No spam, no countdown timers, no urgency tactics. One email when your wave opens. Already have an account? Sign in.