Markets move. Our prices move with them.
A housing market is a living system. It surges in spring and summer, pushes through year-end, contracts when inventory dries up, floods when it overflows, and reshapes overnight around rate shifts, storms and job markets. Any estimate frozen at a single moment in time — however authoritative it looks — is quietly pretending none of that happens. We built our pricing the other way: a price is not a snapshot, it’s a claim about the market right now — so we measure it against the market continuously, and we publish the result. That measurement is Jitter.
Static numbers, living markets.
Institutions like to present property values as settled, static figures. The market underneath never stops moving: seasonal waves of listings and closings, inventory droughts and gluts that bend prices for months at a time, disaster events that redraw whole neighborhoods, lending conditions that change what buyers can pay. A stale price doesn’t feel wrong — it just quietly costs you money, in a negotiation, a listing, or an offer you can’t take back.
So our engine is built to follow the market, not a calendar. It prices from what the record shows today, and a self-correcting loop — Jitter — continuously re-checks those prices against the one source of truth that never lies: what homes actually sold for.
The ARED Estimate — a deterministic engine, not a guess.
Every property report carries an ARED Estimate produced by the platform’s own versioned pricing engine. It prices from the parcel’s public record: the county’s own appraisal and verified tax bills, deed-recorded sale prices, arm’s-length comparable sales, published portal estimates (always shown provider-neutral on our reports), and the neighborhood’s value trajectory.
The engine is deterministic — the same inputs produce the same number, every estimate carries its engine version, and no free-form guessing touches the price. When the record moves, the estimate moves; when it doesn’t, it doesn’t.
Jitter — the price check that never sleeps.
Each day the platform samples recently sold homes at random across all 50 states — a continuous, moving picture of the market rather than a quarterly report card. Only verified arm’s-length sales between unrelated parties, with a publicly confirmed sold price, enter the roster — never estimates, ranges, family transfers or auctions.
The engine appraises each parcel without seeing the sale, and the variance between our estimate and the actual sold price is measured. Every reading is de-drifted for the time gap between the sale and the appraisal, and outliers are clipped so one bad record can never skew a market’s number. A geography needs just a handful of verified sales to publish its own measured accuracy — the Jitter badge you see on our reports.
The number, published.
Sample-weighted mean of every published state reading — our estimates land, on average, within 6.6% of the prices homes actually sold for, in both directions, across 59 verified arm’s-length sales.
| State | Measured jitter | Verified sales |
|---|---|---|
| Arkansas | ±9.6% | 6 |
| Texas | ±4.4% | 5 |
| North Dakota | ±7.9% | 3 |
| Ohio | ±8.3% | 3 |
| South Dakota | ±8.9% | 3 |
| Rhode Island | ±9.0% | 3 |
| Maryland | ±1.7% | 3 |
| Delaware | ±6.5% | 3 |
| Georgia | ±9.9% | 3 |
| Arizona | ±6.5% | 3 |
| Massachusetts | ±9.9% | 3 |
| Virginia | ±4.3% | 3 |
| Minnesota | ±3.4% | 3 |
| Illinois | ±4.0% | 3 |
| New Hampshire | ±4.1% | 3 |
| Nebraska | ±6.7% | 3 |
| Iowa | ±6.1% | 3 |
| Oregon | ±5.6% | 3 |
Each reading is the mean absolute variance between our estimate and the actual sold price on verified arm’s-length sales, de-drifted for the sale-to-appraisal time gap. Measured continuously — the table tightens as the sample base grows. A geography publishes only while its reading sits within our published-accuracy standard; anything wider stays internal until the loop corrects it — we don’t publish a number we wouldn’t stand behind.
The self-correcting loop — a price that adapts with the market.
Measurement alone doesn’t move a market. What does is what happens next: the measured variance feeds a self-guiding, self-correcting loop that re-tunes our pricing to current conditions, pass after pass. Recent market truth carries the most weight — a 30-day recency half-life means this month’s sales steer far harder than last quarter’s — so when a market shifts, in a season or in a shock, the calibration follows it. Today’s conditions, not last year’s average.
And because a self-correcting system must be safe as well as fast, the loop is bounded by design: a minimum evidence floor before anything moves, hard caps per pass and lifetime, so it adapts quickly but can never chase a single sale. Corrections apply at the most local geography the evidence supports — ZIP first, then city, county and state — so a local market’s behavior corrects local prices, not the whole country’s. When the big portals’ own estimates on the same sales disagree with a local reading, that reading holds until the evidence agrees. The loop doesn’t guess; it follows proof.
Every estimate has variance. We audit ours — and show you.
No pricing model is perfect — ours included. The difference is what a platform does about it. We measure our variance daily, publish it on our reports as the Jitter badge, and let the loop shrink it: every correction is written to an append-only audit trail with one-click rollback, and every estimate carries its research timestamp, so any price we have ever published remains attributable to the factors in effect at the time.
That is what standing behind a price means. When we hand you a number, it isn’t a model’s opinion from some stale snapshot — it’s a price continuously checked against what the market is actually showing, adjusted by the most local verified evidence available, with the receipts to prove it.
What we don’t publish.
The engine’s internal anchors and weights are proprietary and are deliberately never published, on any surface. What is public is the measured outcome: every report carries its cited sources, its research timestamp, and its measured Jitter where the local sample base supports one.
Informational use only — the ARED Estimate is not an appraisal, brokerage opinion, or investment advice.