Why it's hard to tell if a used car is a good deal

Stub — this is the seed of what might eventually become its own CompareThe piece, not specific to EVs. For now it lives alongside the electric-car research since that's where the problem showed up.

The problem

A used-car ad gives you a spec string, a price, and a few photos. Working out whether that price is actually good requires things the ad never gives you:

• Jargon decoding. "534ch Performance, 79kWh" isn't self-explanatory — it takes knowing the model line to translate a marketing spec into an actual trim and battery pack. See the worked example in 07-reading-a-real-listing.md.
• A fair-price benchmark. Without knowing what comparable trims at comparable mileage/age actually sell for, "€34,990" is just a number. This project had to go build that benchmark itself — see 05-secondhand-price-bands.html — before the listing in 07 could be judged at all.
• Condition signals an ad doesn't surface. Battery State of Health, accident/repair history, fast-charging habits — the single biggest driver of used-EV value (per 02-model-tco-survey.md) isn't in the ad and usually isn't offered up by the seller.
• Legal/market context specific to the country. Whether a seller is a private individual or a professional changes the legal protection you get (France's garantie légale de conformité for professional sellers) — a fact that changes how much risk a given price represents, and one a buyer has to already know to even think to check.
• Regional variation. The same car can be a good deal in one region and an average one in another, for reasons that have nothing to do with the car itself.

Why this isn't solved by just asking an AI

Worth naming directly, since it's part of the motivation: an AI assistant can decode the jargon and reason about what to check — that's genuinely useful and is most of what made the case study in 07 possible. But it doesn't have a fair-price benchmark sitting in its head, and it can't tell you the car's real condition from a photo. The benchmark had to be built — sourced, checked, corrected when wrong (see the trim-mismatch correction in 02-model-tco-survey.md) — before "is this a good deal" had any real answer. That gap between general reasoning ability and grounded, sourced comparison data is exactly the space CompareThe is trying to fill.

What a fuller version of this piece would need

• A worked-example library (more listings like 07, across more models/trims) rather than one data point.
• A clearer checklist format: what to ask/check before any used-car purchase, EV or not (Histovec, SOH tools, professional-vs-private-seller protections, regional pricing context).
• Whether "appreciation database" is even the right frame — probably closer to a living, sourced price-band dataset per model/trim, kept current, which is what 05-secondhand-price-bands.html is a first sketch of.

Not written yet. Flagging it here so it's not lost, and so 07 has somewhere to point.