Track Record
Every valuation tool claims accuracy. We grade ours in public: predictions recorded before each sale, scored against the price the name actually fetched, updated as new results land. No retro-fitting: a prediction that existed before the outcome is the only kind counted.
Ranking, at every threshold
Sorting a list to find which names carry real money is what the tool is for, so it is what we publish first. Each row asks one question: take a name that really cleared the mark and one that did not — how often is the estimate’s ordering correct? Every threshold we track is here, not the flattering one.
| Name clears | DomainTier | GoDaddy GoValue | Any fixed price |
|---|---|---|---|
| Loading… | |||
The test this board fails
Valuation tools usually quote a within-2× rate. On this board that number cannot measure skill, and we would rather show you why than wait for you to find it.
| Predictor, scored on the held-out fold | Within 2× | Median miss |
|---|---|---|
| Loading… | ||
Within 2×, and what it is worth
The conventional metric, kept because GoValue is scored on the identical names and that comparison stays meaningful. Read the section above first: the distance between two predictors here is real, the absolute rate is a property of the board.
Within-2× rate by realized-price band. Same names, same sales, every predictor scored identically.
| Predictor | Sold $1,000+ | Sold $100–1k | All outcomes |
|---|---|---|---|
| Loading… | |||
How the sample is built
Why the count is what it is. Each step removes names for a stated reason, and the largest cut is the one that keeps the test honest.
Every prediction, plotted
One dot per settled auction: the pre-sale estimate against the price the name actually fetched. The shaded band marks within 2× of the truth. Landing inside it is the conventional test; whether the dots climb left-to-right with the axis is the ranking one.
The scoreboard, every day
Rank-order accuracy at each nightly regrade, at the — mark. Read it with three things in mind. Each point is cumulative over the whole graded record rather than that night’s sales, so it moves slowly and still contains predictions filed before the engine reset — this line cannot be read as the current engine on its own. The series begins the night the measurement shipped; there is no earlier reading, and manufacturing one would be fitting history to a metric it was never scored on. And the shaded band is the 95% interval: the level wanders inside it, so the movement here is not a trend. What has held is the separation from GoValue — it has not closed.
How this is measured
Our auction radar values thousands of live auction names each day, and those valuations are stored with a timestamp. When a name later sells, the recorded prediction is matched to the realized price. Because the prediction predates the outcome, there is no way to fit the answer after the fact. It is the same standard a forecaster is held to.
GoDaddy publishes its own estimate (GoValue) alongside each auction, so the identical comparison is applied to it: the estimate shown before the sale, against the price the sale actually closed at.
The metric caveat comes first, because it is the one that would otherwise be found for us: a within-2× rate on a board this narrow is beaten by a single flat price, and we publish that result above rather than the rate alone. Ranking is the headline because it is the claim that survives the test — a fixed price cannot rank, whatever price it names.
Further honesty notes, because a scoreboard without them is marketing: outcomes here are auction closes, which run below negotiated retail prices. That is a headwind for any retail-calibrated estimate, so read GoValue's row with that in mind. Names that closed under $100 are near-random for every tool we have measured, ours included; they are shown in the all-outcomes column and excluded from the headline. And the sample is names our radar surfaced, not a uniform draw of every auction.
One more piece of candor: we graded ourselves through July against a third-party settled-sales feed that our own verification later caught reporting stale and wrong prices. We purged that feed on July 30 and restarted the record from zero on clean labels. The chart above starts that day for exactly that reason.
The formula behind the valuations, and the evidence stack they rest on, is documented on the methodology page.
Judge it yourself
Run any name through the engine and see the evidence behind the number.
Valuate a domain →