Data downloads
Aggregated UK used-car statistics derived from the TrothStone corpus. Researchers, journalists, and tooling authors: download the files below and cite the build date — the corpus refreshes, so figures move.
Files
| File | Format | Bytes | sha256 |
|---|---|---|---|
| /api/data/makes.json | JSON (per-make quantiles) | 7,552 | ab4c1bf415fd413e3602015ee92386b8acb314e3a777bbbdce0651c7d7f68271 |
| /api/data/model_aggregates.csv | CSV (make × model × year; n≥10) | 0 | — |
| /api/data/methodology.csv | CSV (evaluation metrics) | 428 | 4b7bea63b22dc610c0368325ee794c8a7d3b2b149c873c334907eb1d7e8cecdf |
sha256 stamps are computed at build time against the bytes the live endpoint serves; if a corpus refresh lands, the next build re-stamps and the figures above move.
Aggregate-only boundary (why no row-level export)
The corpus is built from a scrape of public UK listings (AutoTrader, cinch, Cazoo, drive) graded against the TrothStone model. Republishing the raw rows — even with seller names and images stripped — would convert an intelligence layer into a listings data provider, against the terms the source sites publish. Every file above is therefore derived statistics only: quantiles, ratios, counts, evaluation metrics. No per-listing URLs, no per-listing IDs, no seller names, no images.
The aggregate-only boundary is also why every bucket has an n≥10 suppression floor: a row that lists 3 listings of one make/model/year would re-identify them by hand. Anything below 10 listings is dropped from the CSV before the file is served.
How to cite
TrothStone UK used-car aggregates, snapshot 2026-10-02 (https://trothstone.com/data/) — CC-BY-4.0; cite the build date because the corpus refreshes.
What's in the files
makes.json
One row per make that meets n≥10: make, n, p25_price,
median_price, p75_price, median_truecost_mo. Built for quick
"what does this make cost" lookups and headline charts.
model_aggregates.csv
One row per (make, model, year) bucket with n≥10 listings:
make, model, year, n, p25_price, median_price, p75_price,
median_ratio, median_resale_36. median_ratio is
asking ÷ fair (the A–E grade scale); median_resale_36 is
the model's 36-month predicted resale in £.
methodology.csv
Evaluation metrics + rates as cited on the methodology page: holdout RMSLE, MAE, cohort count, lease-deal count, plus the rates table (cash 4%, loan 7.4%, HP 8.9%, PCP 7.9%).
Limitations (read before citing)
- UK-only corpus. No US/EU/AU listings in scope.
- Snapshot is build-day: the figures above move on every corpus refresh.
- Estimates, not financial advice. No credit intermediation; no FCA-regulated activity.
- No CORS on
/api/*: download server-side, or use a proxy. - Aggregate-only by design. If your analysis needs row-level context, the per-model insight pages show price-by-year tables and cheapest-to-own examples per make-model (live, with source links).
Generated at 2026-10-02T23:42:22.833Z · file bytes + sha256 stamped at build time.