How we score every listing.

Every day we scrape public used-car marketplaces across UAE and KSA: Dubizzle, CarSwitch, haraj.com.sa among others. For each listing we capture make, model, year, mileage, location, asking price, and a fingerprint that identifies relistings of the same car.

We track price history per listing so we can see when a seller drops their price (a strong buy signal) and when a listing goes inactive (the car likely sold).

Listings are messy. "TOYOTA CAMRY 2020", "Toyota Camry 2020 GLI" and "كامري 2020" all describe similar cars, but a naive average over this raw data is meaningless. We normalise make and model with a curated alias dictionary plus fuzzy matching, parse mileage and price in a locale-aware way (Arabic numerals, AED vs SAR), and group cars into peer segments by make / model / year / mileage band.

We also flag fantasy listings: prices so far below peers that they're likely typos or scams, and exclude them before computing market statistics.

For each peer segment we compute the median price rather than the mean, because used-car distributions are skewed: a few overpriced flagship trims pull the average up. The median is the middle price; half the segment is more expensive, half is cheaper.

Around the median we measure the Median Absolute Deviation (MAD): a robust measure of spread that ignores outliers. Together they give us a stable centre and a stable scale.

For a given listing we compute its modified z-score:

This tells us how far the price is from the segment median, in robust standard deviations. Negative means cheaper than typical; positive means more expensive.

A car priced more than 30% below its segment median typically scores below −2 and earns the excellent_deal label. A car priced 15% below earns good_deal. A car priced 20% above earns above_market.

A score is only as trustworthy as the sample behind it. We classify confidence by the number of comparables we have for the segment:

When we have very few comparables (less than 3), we don't publish a score at all: we'd rather say "we don't know" than make something up.

Our score is statistical, not mechanical. We see asking prices, not sold prices. We don't (yet) inspect the car, read the service history, or value the trim. So:

• A "Good Deal" can still be a bad car if it's been crashed.
• Premium trims (e.g. Toyota Camry XLE) can score "Above Market" against the median, even when fairly priced for the trim.
• Modified or non-GCC-spec cars are flagged via warnings: read those carefully.

Always pre-purchase inspect, always check the service book, always run the chassis number through the local registry. Our score is the starting point: not the conclusion.