Methodology
How VehicleGrade turns a listing into a score, a market value, and a set of known-issue and maintenance estimates. Every number below is produced by an explicit, rule-based calculation — there is no machine learning model deciding whether a car is a good deal.
Market value estimate
Each vehicle generation in our reference database has a base value and a reference mileage. A listing's estimated market value starts from that base value and is adjusted by percentage for: trim (relative to the base trim), mileage (higher mileage than the reference point lowers value, lower mileage raises it, capped at −35%/+15%), title status (clean, unknown, rebuilt, or salvage), and condition (excellent, good, fair, or poor). The result is an independent estimate of what the vehicle is worth — it deliberately does not look at the seller's asking price, seller rating, or how long the listing has been posted.
VehicleGrade Score (0–100)
The score starts at 50 and is adjusted by a fixed set of rules: how far the asking price sits below or above the estimated market value (the dominant factor), title status, how many known issues are already past their typical onset mileage, whether mileage is high or low for the vehicle's age, seller rating, and how long the listing has been posted. Every adjustment is shown to you as a plain-English reason with its point value, so the score can never say something the explanation doesn't back up.
Known issues & maintenance
Known issues are stored against a vehicle's generation in our reference database, each with a typical onset mileage. For a specific listing, we compare its actual mileage to that onset mileage to say whether an issue is not yet relevant, worth watching for, common at this mileage, or overdue. The underlying issue record never changes — only how it's presented changes based on the odometer reading in front of you.
Confidence score
Confidence starts at 100 and is deducted for concrete, checkable gaps: few or no comparable listings for the vehicle's generation, an unidentified trim, a generation with no known-issue data recorded yet, missing mileage, or (for manually entered listings) optional fields left blank. Every deduction is shown with its reason, and anything that represents missing information is also listed explicitly so you know what to verify yourself.
What VehicleGrade does not do
VehicleGrade does not use machine learning to judge reliability, market value, or deal quality. Where an AI language model is used, it is limited to turning numbers we've already computed into a plain-language summary, or to extracting structured fields from a listing description — it never invents a known issue, a repair cost, or a score. See the Data Sourcespage for where each report section's underlying numbers come from.