Methodology

How Property Priors builds a read

A read is not a guess dressed up as a number. It is a stack of recorded documents, classified, filtered, and adjusted for time, with a confidence tier that tells you how much evidence it stands on. Here is the method, in plain terms.

On this page
  1. The data layers
  2. We classify, we do not average
  3. Arm's-length filtering
  4. Time-adjustment
  5. Confidence tiers
  6. The held-out backtest
  7. On the appraiser objection
Step 1

The data layers

A read is assembled from layers, each one a stream of documents or tracked listings tied to a single parcel. We do not buy a black-box estimate and re-skin it. We read the record and compute from it.

Step 2

We classify, we do not average

The common shortcut is to take every sale in a zip, average the price per square foot, and multiply. That is how you get a number that is precise and wrong. A distressed transfer, a family gift deed, and an arm's-length retail sale all land in the same bucket, and the average washes out the very thing you care about.

We do the opposite. Every recorded event is classified before it is used: an arm's-length sale, a related-party transfer, a distressed sale, or a finished flip. A read for a rehab exit is built from finished flips and retail sales that resemble your subject in condition and size. A rent read is built from tracked leases in the same bedroom class. The subject sets the reference frame, and only events that belong in that frame get a vote.

The short version

Averaging every sale in a zip answers a question nobody asked. Classifying first, then reading from the events that match your deal, answers the one you did.

Step 3

Arm's-length filtering

Not every recorded sale is a market signal. Deeds between family members, transfers into and out of trusts and LLCs at token consideration, quitclaims, and estate settlements all show a price, and none of them tell you what a house is worth on the open market. Left in the pool, they drag the number toward zero and inflate the spread.

We screen those out before scoring. What survives into the comp set is the set of transactions that behaved like real, willing-buyer, willing-seller trades. This is the single biggest reason a raw public-record average and a Property Priors read disagree, and it is why the read shows you the surviving comps by name so you can see what was and was not counted.

Step 4

Time-adjustment

A sale from two years ago is evidence, but it is not today's price. Markets move, and a comp that closed before the market moved will mislead you if you treat it as current. We adjust each surviving comp for the time between its sale date and today, using the trend the market itself recorded, so an older sale is pulled forward to a present-day equivalent rather than used raw.

The read tells you the newest resale date behind each figure for exactly this reason. Freshness is part of the evidence, so we put the date on the page.

Step 5

Confidence tiers

A number with three comps behind it and a number with fifty comps behind it are not the same number, and pretending they are is how people get hurt. Every read carries a confidence tier that reflects how much matching, recent, arm's-length evidence stood behind it, along with the comp count and the newest resale date. The tier is a warning label, and it is shown, not buried.

High
A deep set of matching, recent, arm's-length comps. The read stands on a lot of evidence and the range is tight.
Moderate
Enough comps to read, but thinner or older. The range widens to reflect it, and the read says so.
Thin
Too few matching records to price honestly. We widen the range, flag it, or decline the number outright rather than guess.

When the evidence is not there, the honest move is a dash with a reason, not a confident-looking number. A read that will not manufacture a figure it cannot support is the whole point.

Step 6

The held-out backtest

A method is only trustworthy if it has been checked against outcomes it did not get to see. We hold out a set of parcels that later sold on the record, build the read as if we did not know the outcome, and compare the read to what the house actually resold for.

What the backtest shows, scoped honestly

On high-confidence reads, the median miss against actual resales is about 8 percent. That figure describes high-confidence reads only, checked against real resales; thinner tiers carry wider error by design, which is exactly why the tier is on the page. The method cleared this bar before we put it in front of anyone.

This is not a claim that every read is right to the dollar. It is a claim that we measured our own error, on data the method did not train on, and that we report the tier so you know which reads earned their tight range and which did not.

The objection

On the appraiser objection

A fair challenge we hear from appraisers: you do not baseline off raw public record, because raw public record is full of non-market transfers, stale dates, and mismatched properties. That is correct, and it is precisely the argument for everything above. We do not baseline off raw public record either.

The steps on this page, classify before you count, filter to arm's-length, adjust for time, tier by evidence, and check against held-out outcomes, are the same disciplines a good appraiser applies by hand, run consistently over every parcel and shown with the receipts. A Property Priors read is not an appraisal and does not replace one. It is a fast, evidence-first read that tells you where a deal stands and shows you the documents, so that when you order the appraisal you already know what it should say.

Not an appraisal

A Property Priors read is an information product built from public records and market listings. It is not an appraisal, not a broker price opinion, and not legal, tax, lending, or investment advice. It is not a consumer report, and it must not be used for tenant screening, credit, employment, or insurance decisions, or for any purpose covered by the Fair Credit Reporting Act. Verify anything that carries money against the source documents before you act on it.