Fair play

How marks work

Every garage in every league is valued by one published method — the Dream Garage Mark. Same data, same math, every player.

Archive: 1,334 recorded sales(708 BaT · 626 Curated OVRLND)· deepens daily

On every lot the AI also predicts the hammer from the same archive — the Mark is the value; the prediction is your guide when you bid.

1.One method for everyone

Every car in every garage is valued by the same published method — the Dream Garage Mark. Nobody gets a better feed; the edge is in the buying, not the data.

2.Primary mark

The median of comparable sales: same model line, within ±3 model years, sold in the trailing 24 months, with at least 3 comparable sales. The archive holds Bring a Trailer auction results and — tagged by source — completed sales from Curated OVRLND, a curated dealership whose sold archive deepens the classic-truck families BaT covers thinly. A model line pools its variants — an SL500 and an SL600 are the same line (the numbers name the engine, not the car), while genuinely different lines (a 300SL, an SLS) never pool together.

3.The fallback ladder, always disclosed

When the primary set is too thin, the method widens in a fixed order — ±5 model years, then a 36-month window, then the marque-wide median, then the catalog reference value. Every rung needs at least 3 comparable sales, the marque rung included: with fewer than 3 sales anywhere in the marque, the mark falls through to the catalog reference rather than lean on one or two sales. Any mark from a fallback rung carries its basis in plain words beside the price — "From ⟨n⟩ real sales" when sales priced it, "Collector value · catalog reference" when the catalog reference did. And if any comp pool spans ten-to-one or more from its cheapest to its dearest sale, that pool is not one market, so the Mark steps past it rather than average two different worlds. The marque median must also sit within a third-to-three-times band of the car's catalog reference when one exists; a marque number wildly inconsistent with the reference abstains to it — and for a reference of $1M or more, that abstention yields no Mark at all rather than a pooled mispricing. The marque rung must also be relevant: it may only price a car when at least 3 of the sales in the marque pool are the car's own model line — with fewer than 3, the pool mostly describes other cars, so the rung abstains to the catalog reference instead of pricing one model off its cousins.

4.Sources, never blended silently

Every mark names its sources and comp count (e.g. "7 BaT sales, 24 mo", or "9 sales (6 BaT · 3 Curated OVRLND), 24 mo"). Provenance is stored on every archived record and carried through the comp pool to the label and the nightly integrity sweep. Two honesty rules ride along: a broker's displayed sold price may be its final asking price rather than a verified auction hammer, so broker sales are tagged as their own source and the method's at-least-3-sales medians keep any single record from carrying a price; and a live listing's asking price is never a comp at all — asking prices sit in the archive as reference data only, excluded from every valuation by type. The broker publishes no sale dates, so its sales are dated at listing (a sale can only complete after its listing); records whose dates were lost when the store changed platforms are archived dateless and — exactly like an undated auction result — never enter a trailing window. When a model has almost no sales history anywhere, it is flagged for review rather than quietly valued on air.

Reading a mark

“Dream Garage Mark — 7 BaT sales, 24 mo” is a primary mark: the median of 7 comparable sales inside the standard window. “(thin data)” means a fallback rung was used — the label always says which one and how many sales stand behind it. Predictions in the morning brief are estimates from the same archive, never guarantees.

How we keep prices and photos honest

One price, one method. Every value in the game derives from the Mark above — never from a listing's asking price, never from a prediction, and never typed in by hand. There is no screen anywhere in Dream Garage where a person can edit a price.

A photo has to prove it belongs. A picture may only appear as a car model's face — on cards, in the wizard, in the swipe deck, on your Dream List — after its source listing has been checked against the model it's filed under: the listing's own title and URL must agree with the car. Anything unverified steps aside for a designed placeholder tile. A photo of a different car never stands in.

A sweep walks the vaults nightly. Every night an automated integrity sweep re-verifies every photo, hunts for Mark outliers and predictions that should have abstained, checks every shelf against its tier's cap, and confirms every model in the catalog carries a reference value. Anything it finds goes to the commissioner's review queue with its evidence attached — prices there are never hand-edited; they resolve only by fixing the data the Mark reads.

And you can flag anything. See a wrong photo, a wrong price, or the wrong car entirely? The “Flag this” link on car pages and cards sends it straight to the same review queue. Flags are private — nobody else sees yours — and every one is read by a human commissioner.