1. Five principles
Segment-relative, never absolute
A city car and a supercar are never scored on the same axis. Every car is ranked only against its peers: same body type × same fuel type × same price band (quintiles within the body–fuel group). If a segment has fewer than 8 members, it widens — first dropping the price band, then the fuel type — so no car is ever "best of one".
Percentile-in-class normalization
Each criterion becomes a percentile rank inside the segment (the approach recommended by the OECD–JRC Handbook on Constructing Composite Indicators). That is why a score can honestly say "top 12% of its class" — the number IS the rank. Percentiles are also immune to spec-sheet outliers.
Missing data is never guessed
If a car is missing one criterion (e.g. no boot-capacity figure), the remaining criteria of that dimension are re-weighted to sum to 100% — the gap is not filled with an estimate. If an entire dimension has no data, it is excluded and the overall re-weights across what remains. A car with no usable data gets NO score, never a fabricated one.
Weakest-link cap
Borrowed from Euro NCAP and Green NCAP: a car cannot reach the top of the scale if a core dimension is bottom-tier. If Efficiency or Value sits below the 20th percentile of its segment, the overall score is capped at 4.2. A 700 hp car with terrible economy cannot out-score everything anyway.
Popularity is never an input
Page views, search demand and brand popularity have zero weight. A viral car is not automatically a good car. The score is computed from technical specifications only.