Show the source. Show the limits.
A valuation is only as good as what sits underneath it. This is how observations are graded, what each one has to carry, and where price evidence normally goes wrong.
Not all prices are evidence
Five different things in this market are all called "the price". They are not interchangeable, and the gap between the top and bottom of this list is routinely larger than the disagreement people are arguing about.
Completed sale
A transaction that closed, with a known price, date and buyer type. This is the only class of evidence that records what somebody actually paid rather than what somebody was willing to say.
Accepted bid or confirmed hammer price
A price a seller accepted, before fees and transport are resolved. Close to a completed sale, but the final settled amount can still move.
Closing bid on an unsold lot
Useful as a floor: the market reached that level and no further. It is not a sale, and treating it as one drags every derived figure downwards.
Reserve or guide figure
A seller's expectation, often set commercially rather than analytically. Informative about intent, almost uninformative about value.
Asking price or listing price
What was hoped for. It is the most abundant price information in the market and the least reliable, which is exactly why it contaminates so many datasets.
Weaker classes are not discarded — a floor and a ceiling are useful. They are labelled, weighted down, and never allowed to masquerade as a completed sale in a result.
What every observation has to carry
A price without context is a rumour. An observation is only usable if it arrives with the fields that determine what it means, and keeps them for its whole life in the dataset.
Price type
Which rung of the hierarchy above the figure sits on. Permanently attached, never inferred later.
Date observed
When the figure was recorded, not when it was ingested. Salvage demand moves with metals prices, parts supply, currency and export routes.
Fee treatment
Whether the figure includes or excludes buyer premium, lot fees, storage and administration. A number that silently flips between gross and net makes a dataset unusable.
Transport basis
Whether collection cost sits with the buyer, and from where. On low-value salvage, transport can be a large share of the total outlay.
Currency and tax basis
The currency, the rate date, and whether the figure is stated under a margin scheme or with recoverable tax. Cross-border comparison fails first here.
Vehicle and damage state
Identity, mileage, recorded category, damage area, loss type and whether the vehicle was complete at the point of sale.
Provenance and reuse rights
Where the observation came from and whether it may legitimately be reused. Public visibility is not permission.
Where price evidence breaks
We do not name platforms, providers or competitors. These are structural problems in how damaged-vehicle price data is gathered and reused, and they apply broadly across the sector.
Survivorship in the visible record
The transactions that are easiest to observe are not a random sample. Lots that sell quickly, in volume, through channels that publish outcomes are over-represented. Unusual vehicles, private disposals, direct trade deals and negotiated sales are under-represented — and those are disproportionately the cases where a valuation is actually disputed.
Relisting duplicates inflate apparent depth
An unsold vehicle relisted several times can appear as several observations. A dataset counting lots rather than vehicles will report a comfortable comparable count for what is in fact one car that nobody bought. Deduplication by vehicle identity is not a nicety; without it, confidence is measuring the wrong thing.
Fee-inclusive and fee-exclusive figures get mixed
Different channels publish different bases, and the difference can be a fifth of the price on cheap lots. Once mixed, the error is not noise — it is a systematic bias that shifts with the channel mix in the sample.
Tax and margin-scheme treatment is dropped in transit
Whether a price is stated under a margin scheme, with recoverable tax, or gross materially changes what a buyer pays and what a seller nets. This distinction is routinely lost in aggregation, then rediscovered as an unexplained gap between two apparently identical markets.
Damage description is free text pretending to be structure
Damage is usually captured as a short phrase written under time pressure. Mapping that to a structured severity is an interpretation step, and interpretation steps get silently hardened into categorical facts. Where the mapping is uncertain, the uncertainty belongs in the output, not in a footnote.
Completeness is assumed
A vehicle may be missing parts, keys, documents or a battery pack at the point of sale, and the record often does not say. Two observations of the same model at very different prices may be entirely consistent once completeness is known — and entirely misleading when it is not.
Access is confused with permission
Data that can be seen is not automatically data that can be reused commercially. Building a reference on observations without a defensible basis for reuse is a legal exposure and a continuity risk: the dataset can disappear at the moment it is most relied upon.
Coverage gaps are filled rather than declared
When evidence runs out, the professional response is to say so. The commercial incentive is to interpolate from neighbouring models, older periods or other markets and present the result at the same apparent precision. That choice is rarely disclosed, and it is the single most common reason a valuation cannot be reproduced.
See how this evidence is turned into a figure on the method page.