I SHOW MY WORK.

While a home is for sale, I show a likely sale price for it and keep it current as the market and the asking price move. When the home sells, I take the last likely sale price I showed while it was still for sale and compare it to the real selling price. The number above is the median gap across those homes: half land closer than that, and half land further.

One number for a region this large hides more than it tells. My estimates are closer in some cities than others, and in a few places they lean one way on average. So the table below breaks the same record out by city and by postal area, with the number of homes behind every figure.

Nothing is left out. Places with only a handful of graded sales are here too, marked, because a breakdown that quietly drops its smallest rows stops adding up to the total above it. The counts in the table sum to the count in the headline.

The last column is the way my estimates miss on average, not how far any one home missed. A place reading low is one where, on average, my estimate came in under the eventual sale price. Where I cannot tell a direction apart from ordinary scatter, it reads level rather than showing a number I cannot stand behind, and the range under each row says how sure I am.

I watch this myself. When a place's lean becomes both real and large, I treat it as something to fix in my pricing, and until I fix it I stop showing that place a headline accuracy figure anywhere else in the app.

I do not just publish these numbers. On a closed listing, the app shows my likely sale price next to what the home actually sold for, so you can check it one sale at a time. Browse listings.

I use the typical percentage gap, the standard accuracy measure for home-value estimates, and I leave out any home my pricing learned from, so every figure here comes from homes it had never seen. I grade the last estimate I published while the home was still for sale, against the price it actually sold for. The full methodology, including the chance-of-selling side of the Signal, is on the methodology page.

The accuracy figures on this page are public, for the region and for every place in it. The price a specific home sold for is reserved for registered consumers, because board rules treat closed sale prices as members-only. Create a free account to see the per-home comparisons.