Method
What the words on this site mean, what we can and cannot see, and every gap we know about. Every figure here is read as the page renders, never typed in.
Read 11 Sept 2026, 05:03 am AWST. Counts are live; the price-basis figures come off a snapshot rebuilt twice a day, last at 10 Sept 2026, 08:25 pm.
What an observation is
Observation
12,526,810recorded readingsOne reading of one advertisement at one moment: price, odometer, photo count and the raw page. We read the same advertisements again and again and write a new row each time; nothing is overwritten or deleted. The first was 10 Aug 2026, 08:46 pm AWST, which is as far back as any claim here can go.
Sources
1,011enabledDealer websites and public datasets. We identify ourselves as SpotlotBot, obey robots.txt and its crawl delay, and take one page at a time per host. A site that declines automated access is left alone and listed as a gap; nothing is routed around. Coverage lists every one and its last pass.
A listing is not a car
Listings against cars
1.48listings per car, liveAn advertisement is a page; a car is a thing. One car is often on several pages at once, and counting pages where you meant cars overstates the market by that ratio: right now 229,415 live listings for 155,112 cars. Every figure on this site says which it counts.
How we know two listings are one car
136,750cars with a VINA VIN, where the seller publishes one — 61% of cars. Otherwise a fingerprint of make, model, badge, year, odometer and colour, scoped to one seller, because two different cars at two yards can look identical on paper.
Merged
5,420car records reconciledWhen a VIN proves two records are one car, the newer is retired into the older. Likewise sources (33) that publish under two addresses and listings (5,418) reachable by two URLs. A merge corrects our records, never the market.
What “sold” means, and what it does not
Nobody in Australia publishes what a used car actually sold for; a site claiming to is inferring. Here “sold” means only that a car left the market and the seller’s own page said so, and the price shown is the last one asked, not the one paid.
Sold: the seller said so
30,611cars · 73% of exitsThe advertisement changed to say SOLD, or the seller’s own feed marked it sold. The strongest evidence there is.
Sold: the page is gone
2,165cars · 5% of exitsThe advertisement now answers 404 or 410. 747 were 410, the stronger of the two: not found against deliberately removed.
Gone, unproved
9,350cars · 22% of exitsThe page still answers but no longer describes that car. The listing is gone; whether anyone bought the car, a page shape cannot say. Never counted as sold anywhere on this site. At some yards this is every exit, and folding them in would flatter exactly those dealers.
And in every case
A car is gone only once no listing of it is live anywhere we watch. Leaving one shopfront while advertised at another is leaving the shopfront, not the market.
Lapsed
23,025listingsWe lost the ability to watch it: a marketplace window moved on, or a source was withdrawn. Our failure, never an exit, and named as a series break below because a batch of lapses looks exactly like a fall in stock.
Prices are not one unit
Price basis
40%of live prices do not sayA price is quoted drive-away or excluding government charges. Of 228,768 live prices as at 10 Sept 2026, 08:25 pm AWST: 94,453 drive-away, 43,560 excluding government charges, and 90,755 that do not say. The gap shows only on a car carrying both quotes, which 66 cars do: the middle one differs by $2,727, with the middle half between $1,000 and $4,673. Too few to correct by, so we publish the mix instead.
A median across them mixes two units and an unknown. We publish it as the best available answer, with the mix beside it every time.
A median across them mixes two units and an unknown. We publish it as the best available answer, with the mix beside it every time.
Asking prices, not settlements
Every price is what a seller advertised on a day we read it. A cut is the difference between two readings of the same advertisement: the seller’s decision, not the market’s.
When a chart is allowed to be drawn
The evidence ladder
1,501cohorts assessedA chart does not draw because it can. Each of the 7 chart types declares the evidence it needs — cars, spread, days on the books — as rungs (13 in all, none above 2), and a cohort draws to the highest rung it earns. One that earns none gets a sentence saying what is missing, not an empty axis. Promotion needs two measurements; demotion is immediate.
Ladder movements recorded
254promotions and regressionsEvery time a chart appears or disappears: which cohort, which way, and which measure decided it.
What we have got wrong
Generated from the checks themselves, so it is the list we look at. The full register of what has gone wrong and what stops it recurring is on the data-quality page. Every change to the site, in the words it was made with, is on the changelog.
33 invariants, last run 11 Sept 2026, 04:49 am AWST: 0 failing, 5 warning.
- warn
no VIN maps to two different answers about what the car is — 168 pair(s) one VIN says are one car and the taxonomy says are two - warn
no dealer listing unseen for over a week — 1187 stale - warn
no dealer price has gone a fortnight unconfirmed — 345 dealer asks unconfirmed for 14+ days; oldest 30.9d (dealer:rockdalecitychery) - warn
no listing records more price changes than it has had prices — 58 listing(s) oscillating, 2724 events, e.g. dealer:autopact https://www.autopact.com.au/stock/details/OAG-AD-25253996/2025-foton-tunland-new - warn
no listing has been two cars — 328 listing(s) changed which car they describe, 357 pair(s), 642 price changes written on them, e.g. dealer:traralgontoyota https://www.traralgontoyota.com.au/cars/demo-ash-slate-2026-toyota-hilux-13001930 (6 cars)
Series breaks
1recordedA day the basis of a series changed rather than the market. Every figure spanning one carries the explanation in the same panel.
Pages that changed which car they describe
328listingsSome dealer URLs serve a different car over time. Where the VIN proves it we stop comparing readings across the swap, and 187 price changes written across one are marked as what they are. The readings stay; the inference was false.
Using these figures
Quote anything here with attribution and a link, and carry the qualifier that came with it: the sample size, the date, the basis. Spotlot is free, carries no advertising and sells no data. If a figure looks wrong, say so: taj@dealerloop.com.au.