YardBrain — part intelligence for salvage yards

What it is looking at right now

These figures are read from the running app, not typed into this page. They move every night as cars arrive and are crushed — roughly sixteen thousand cars turn over in a week.

$81.9M Inventory priced across 189 lots · 361,377 cars

$31.1M Profitable after fees & shipping ≈39% rests on estimated yard costs

hand the counter (inc. each yard's own tax)$18,501,574.67

it all sells for$81,918,846.74

eBay takes−$12,083,751.22

shipping takes−$10,284,744.20

kept, on the pulls that come out ahead$31,092,085.47

The last line is not the four above it subtracted. The cost lines count every part worth pulling; the total counts only the pulls that come out ahead, which is the number a person with a trolley and an afternoon can act on. Measured Oct 3, 2026 09:00 UTC, refreshed hourly.

189salvage lots walked

361.4Klive cars tracked

3.72Mpart records indexed

160.9Kpulls worth making

What these numbers are not

Not revenue, and not gains. Nothing here has been bought, pulled or sold. It is what the parts worth pulling would cost and would return — a priced opportunity across a whole market, measured continuously.
Not uniformly firm. About 39% of the profit figure rests on an estimated yard cost rather than a published price board. The app tracks that share as its own number so a mixed total can always say how much of itself is soft.
Not a fixed achievement. Cars arrive and are crushed daily. Two readings an hour apart legitimately differ by tens of thousands of dollars.

The problem

A self-service yard prices by part type, not by part. A door handle is a door handle: same shelf price whether it is a common one nobody wants or the one variant that is unobtainable new. The information that separates them is real, public and scattered across catalogues, interchange tables and completed sales — and nobody standing in a row of cars on a Saturday has time to assemble it.

The hard part was never finding parts. It was refusing to guess about the ones it could not price.

An early version estimated a part's value as a fraction of its list price. Measured against real research it was wrong in both directions — an ABS modulator estimated at $539 sells for about $55; a headlamp estimated at $205 fetches around $400. Replacing the guess with evidence moved one lot's headline from $10,251 to $4,522. That correction is the product's whole design philosophy, arrived at the expensive way.

What it actually does

●

Walks the lots

Every yard's live inventory is pulled on a nightly schedule, cars matched to catalogue entries, and cars that have been crushed retired — scoped per yard, so one yard's sweep can never declare another's cars gone.

■

Builds a real receipt

Counter price plus core charge plus that yard's own county tax, against the evidence price for the part, minus the marketplace fee and the shipping it will actually need. Money is never averaged across yards.

◆

Grades its own confidence

Three badges, decided server-side: a researched sold comp, a discounted active asking price, or an arithmetic estimate. An asking price can never be promoted to a sale, however many listings stand behind it.

▲

Watches for arrivals

A watchlist that rescans as new cars land, a pull log, a shelf inventory, and a go-bag checklist for the trip itself. Built phone-first, because that is where it is read — standing in the row.

The rules it is built on

Truth over optimismA well-sourced $55 beats a confident $500. Every price carries its confidence, its sample count and how it was matched — and the app says "I don't know" rather than flattering a number.

A zero is a claimAn unverified gate fee once rendered as $0.00 on 106 lots, which reads as free entry. Unknowns are dashes now. A fallback shown as money is a lie with a decimal point.

A guessed total says soAny figure mixing researched and estimated dollars keeps its exact value, wears a mark, and can state what share of itself is a guess — because "some of this" could mean a dollar or all of it.

Checked against itselfThe pricing logic is verified by replaying it over the live corpus — most recently 464,911 comparisons with zero divergence — because fixtures only prove the cases somebody thought of.

A wrong match counts like a right oneA metric that moved is not a metric that moved to the right place. An early matcher priced a headlamp as a tail light and the report called it a gain. Deltas are read against the actual part, never on their own.