A market-position endpoint for used vehicles.

Ask it about one car and it answers with that car's standing in its local market — not a page of rows for you to reduce yourself.

Endpoints

GET/v1/valuation Where one vehicle sits: percentile, rank, price band, days-supply, and a confidence object naming which dimension had to be relaxed to find the comparables.

GET/v1/comps The comparable vehicles behind that answer. Thin factual rows, each naming its dealership and linking to that dealership's own listing.

GET/v1/health When the archive was last updated, and per-county coverage. Fails loudly rather than serving stale numbers quietly.

What a valuation looks like

This is now the shape the engine actually returns, taken from a real reading rather than sketched. Nothing serves it over HTTP yet.

GET /v1/valuation?vin=3C4NJDCN3RT…&mileage=33098&zip=48307

{
  "ask": {
    "percentile": 28,
    "rank": 27,
    // What `rank` is out of, and NOT the same as `of_n`: the subject
    // is one of the cars being ranked and the comparables are the others,
    // so this is always `of_n` + 1. Reading rank against of_n produces
    // "10th lowest of 9".
    "rank_of": 103,
    "of_n": 102,
    "median": 23357,
    "p25": 22334,
    "p75": 24489,
    "median_at_mileage": 22928,
    "mileage_adjusted": true,
    // The subject is priced so far outside its comparables that it is
    // probably a different specification. The market figures stay true;
    // "over market by X" is withheld. See the refusals below.
    "subject_out_of_range": false
  },
  "confidence": {
    "tier": "high",
    "n": 102,
    "radius_mi": 20,
    "relaxed_dimension": null,
    // p75/p25. Near 1 is one specification; far above it is a mixed pile.
    "spread": 1.11,
    "mixed": false
  },
  "as_of": "2026-09-05",
  "disclaimer": "Asking prices from public dealer listings. Not sold prices, not a trade appraisal, not a transaction estimate."
}

days_on_market is not in this response and will not be until it is real. The archive began on 3 September 2026 and records every car that leaves from that date; publishing a median before enough of them have left would be the one number here that cannot be checked by anyone reading it.

Two things that are unusual, on purpose

A refusal is a 200, never a 404. There are three, and each names itself: insufficient_comps (fewer than five nearby), comparables_too_varied (they were found, but they span more than 2.5× from lower to upper quartile, so they are a base trim and a top trim in one pile rather than one comparable set), and subject_incomplete. A machine handles a structured refusal far better than an error, and "we do not know" is a real answer that deserves a real response body. A refusal carries no prices at all — just the dealerships, mileages and links — because handing a language model three prices gets them averaged and presented as an answer.
It answers a question rather than selling rows. Listing APIs hand you vehicles and leave the statistics to you. This one does the comparable selection, the mileage adjustment and the confidence assessment, and tells you what it relaxed to get there.

Terms, plainly

Derived figures — percentiles, medians, counts, indices — are yours to display to your own users on the tier that permits it.
Comparable rows carry attribution and a link, and must keep both wherever they are shown.
Photographs and dealers' written descriptions are never collected, so they are never served. Neither is anyone's personal information.

If you are building something that would use this — a CRM, a marketplace, an appraisal tool, an agent — write to hello@lotclock.io and say what shape you need. The contract is easier to change now than after anyone depends on it.