We don't publish surveys. We publish math. Here's exactly how raw failure data becomes the objective Risk Scores you use to make better buying decisions.
Every score on our platform is the output of a four-stage data pipeline. No surveys. No guesswork. No advertiser influence at any stage.
Millions of individual data points on repair, warranty, and recalls by component and model.
Raw failure counts are normalized against vehicle population size, model year production volume, and class peer benchmarks to ensure fair apples-to-apples comparisons.
Each normalized record feeds two independent algorithms: our Frequency model and our Severity model. Catastrophic failures receive exponentially higher Severity weights.
Final normalized scores are published. A score of 1.0 = class average. Below 1.0 = outperformer. Above 1.0 = underperformer. Benchmarks recalculated quarterly.
We separate every failure event into two mathematically independent dimensions. Understanding both is critical to interpreting your Risk Score correctly.
Measures how often a vehicle visits the shop compared to its exact class peers. Normalized against fleet size and production volume so a model with 500,000 units is evaluated on the same terms as one with 50,000.
A high Frequency score is costly over time, but doesn't necessarily mean the car is dangerous. It might just mean you'll be in the waiting room a lot.
Measures the financial and mechanical danger of failures. Engine failure, transmission failure, and drivetrain collapse receive exponentially higher weights than nuisance failures like infotainment glitches or interior trim issues.
A car can visit the shop often for minor issues and still score low on Severity. A car with rare but catastrophic failures scores very high. This distinction can save you thousands.
Rarely breaks. When it does, it's minor. This is your target.
In the shop often, but bills stay manageable. Annoying, not ruinous.
Rarely breaks — but when it does, it's catastrophic. Dangerous gamble.
Breaks often and expensively. No negotiation justifies this risk profile.
Traditional automotive reviewers rely on subjective owner surveys (“Did you like your car?”) or short-term initial quality metrics — issues that occurred in the first 90 days of ownership. We use objective failure data spanning the entire operational lifespan of each vehicle. We also apply a severity weighting system: a broken infotainment screen and a failed transmission are fundamentally different events. On our platform, they don’t count equally — because in the real world, they aren’t.
We separate all failure data into two mathematically independent dimensions. Frequency (The Hassle Factor) measures how often a vehicle visits the shop compared to its class peers — a score of 1.0 means exactly average, 0.5 means half the failure rate, 2.0 means twice the rate. Severity (The Pain Factor) measures the financial and mechanical danger of those failures. Catastrophic events like engine and transmission failures are weighted exponentially higher than nuisance failures. A car can visit the shop often for minor issues and still score low on Severity.
A Risk Score of 1.0 means the vehicle performs exactly at its class average — no better, no worse. It is not a rating out of 10. It is a normalized ratio against a dynamic, rolling class benchmark. A score of 0.5 means the vehicle has half the failure rate of the average (a strong outperformer). A score of 2.0 means twice the failure rate (a significant underperformer). The benchmark updates quarterly as new data enters the system, so scores reflect current real-world performance, not outdated snapshots.
Yes — and powertrain isolation is one of our most important differentiators. An Internal Combustion Engine version of a model and a Hybrid version of the model are treated as entirely separate vehicles with entirely separate reliability profiles. Their powertrains are mechanically different. Their failure modes are different. Their component-level risk profiles are different. Blending their data would produce a meaningless average. We don’t do meaningless averages.
Absolutely. Armed with component-level severity data, you can walk into any dealership with specific, data-backed questions. “I see this model year has an electrical system severity score 2.8x the class average. Can you provide documentation of a pre-sale electrical inspection?” Dealers operate on information asymmetry — the assumption that you don’t know what they know. Our data closes that gap, in your favor, before you sign anything.
Generative AI is excellent at summarizing qualitative information at scale — blog posts, owner forums, review articles. That is genuinely useful for research context, and you should use it. However, LLMs are language models, not statistical engines. They cannot tell you that the 2019 model year has a transmission failure rate 3.2x higher than the 2020 due to a specific production run defect. They summarize narratives. We produce quantitative, standardized, reproducible metrics. Use both tools — but understand precisely what each one is actually doing.
We aggregate from multiple independent, objective sources: NHTSA complaint databases, manufacturer Technical Service Bulletins (TSBs). Every data point is cross-referenced against vehicle production numbers to normalize for fleet size — a model with 500,000 units on the road is held to a different absolute standard than one with 50,000. We do not accept data sponsorships from automakers or dealership groups.
Our dataset is refreshed on a quarterly cycle as new failure records are processed and validated. The benchmark averages used for normalization are also recalculated each quarter, so scores always reflect the current competitive landscape, not historical baselines.
Because the platform focuses on real-world data from vehicles in the general population, it takes time to get an adequate sample set to do the analysis. This means that the very latest year models will take time for their data to "trickle" in.
We do not accept advertising from automakers, dealer groups, or any entity with a financial interest in vehicle scores. Revenue comes exclusively from subscriber access fees. Full stop.
Every score is the output of a documented statistical algorithm. No editorial opinion, no subjective ratings, no expert panels. Numbers in, numbers out.
Data and benchmarks are recalculated every quarter. Scores reflect current real-world performance, not historical snapshots that may no longer represent the vehicles on the road today.
Every vehicle is evaluated relative to its class peers and normalized against its own production volume. Popular models aren’t penalized for scale. Rare models aren’t given a free pass.
Access every Risk Score, component breakdown, and year-over-year comparison across 3,200+ models going back 21 years.