Prices a private-party listing against what an instant-offer buyer will actually pay, then subtracts the real cost of going and getting it. The ceiling is the point. ← simple view
Without a real offer the engine only screens — it will not quote a price, because a modelled ceiling is not a negotiating number.
A buyer’s offer is retail minus a margin minus a cost stack that does not scale with the car — roughly $2,900 of recon, auction fees and holding whether the car is worth $30,000 or $3,000. So the fraction they keep is not a constant. The orange line is what a single ratio does: it tracks at the top and invents money at the bottom, where this business actually operates.
Hand-extracted from public Reddit threads. Self-reported and second-hand, so they seed the model rather than calibrate it. Tier A carries a real retail anchor; tier B is shown hollow and excluded — two of those rows have an offer exceeding retail, the signature of a poster quoting KBB trade-in as if it were value. Fitting the 11 clean tier-A points gives alpha 0.806 against our guess of 0.80, and beta $2,409 against our $1,800.
Cost model: $0.62/mi, $45/hr at 45 mph door to door, one fixed hour for meeting and paperwork, $450 blind recon, $145 title and registration, $600 minimum margin, 6% haircut on an unverified offer. Every one of those is an assumption. Replacing them with measured numbers is the highest-value work available.