We measured how often our own model is right. Here are the numbers.

Every property tool claims its predictions work. Almost none of them show you the denominator.

So here is ours. We took every monthly scoring cohort old enough to have a three year outcome, matched each district's score to what actually happened to prices afterwards, and counted. About 28,000 district windows across 13 cohorts. This is what a Praesago score has historically meant:

Score band	Beat their county median	Sample
60+	66%	1,422
50 to 60	53%	13,182
40 to 50	44%	12,404
under 40	36%	1,227

Read it from the bottom up, because that is where the honesty lives. A district scoring under 40 beat its county's price growth about a third of the time. The middle bands hover around a coin flip. The 60+ band gets it right two times out of three.

That gradient is the whole product. If the numbers were flat across the bands, the score would be decoration. They are not flat. Each step up the score means better odds, every time, in the same order. Statisticians call this calibration. We call it the reason the thing is worth paying for.

What 66% is not

It is not a guarantee, and we would rather lose a sale than pretend otherwise. One in three of our highest scoring districts still underperformed its county. If you buy one property in one high scoring district, you are taking one draw from a distribution, and distributions do not care about anecdotes.

What the score gives you is the same thing a card counter gets at a blackjack table. Not certainty on any hand. Better odds across many hands. That is why the product is built for screening: start with 2,292 districts, use the score to shortlist, then do the local work the data cannot do.

Where the model fails, in public

Our model has been through seven historical test windows. It failed two of them, and both failures are published on our validation page.

COVID inverted it completely. The 2017 to 2022 window caught the largest urban to rural shift in modern history and the model got it backwards. The 2022 rate shock produced a milder failure. The pattern is consistent: the model reads structural change in neighbourhoods. It does not read macro shocks. If you believe a pandemic or a rate spike is imminent, this model will not warn you.

The part nobody publishes

This month we tested two ideas for improving the model. One adjusted scores for districts with thinner data. One fed Bank of England rate moves into the regime detection.

Both failed our promotion test. Both were reverted and written into the changelog next to every other idea that did not survive contact with the data. Since 2026 we have rejected more signals than we have shipped, including several that looked great until we checked whether their predictive power was borrowed from signals we already had.

We also run a deliberately simpler version of the model in parallel every month, scored on the same districts, judged by the same outcomes. If the simple version keeps beating the complicated one, we will ship the simple one and say so. The next verdict lands in January.

None of this makes for good advertising. All of it makes for a number you can actually use.

Check it yourself

The full validation record, including the failures, is at praesago.com/validation. The score for any postcode district in England and Wales is on the live map. Look up somewhere you know well. That is the fastest way to decide whether the model sees what you see.