EV Battery Scores: Modeled or Measured?

Dealers relying on model-based scores are vouching for a guess.

The battery is the most expensive component of an EV, so when buying a used EV, the first question is often: what’s the health of the battery? Increasingly, dealers answer that question with an independent third-party score on the Vehicle Detail Page. The problem is the integrity of the data behind that score.

Many dealers are relying on model-based estimates: a statistical guess at the health of the battery based on aggregated data from similar vehicles. The result is a “score” inferred without ever measuring data from the vehicle itself.

We analyzed battery degradation from 150 vehicles using Lyteflo’s battery certification test, measured directly from the vehicle’s onboard diagnostic data, and compared the result to the exact same vehicle’s “range score” from the battery report of a well-known model-based product.

Model-based approach overstates scores on every cohort.

When comparing the direct-measurement method to the model-based method on 150 vehicles, we found that the model-based estimate overstated the score by 6.2 points on average.

Furthermore, the model-based estimate read higher than the direct-measurement on 92% of vehicles and exhibited a maximum overstatement of 18 points: issuing a score of 95 on a 2022 Tesla Model Y that actually reported a score of 77 using the direct-measurement method.

Reading the table. Each cohort is cumulative (e.g. “2022 & older” includes all vehicles model year 2022 and earlier). Scores are 0 to 100; average overstatement is the model-based score minus the direct-measured score, averaged across the cohort. The sample is approximately 57% Tesla.

A range estimate, read as a health score.

The model-based estimate is presented as a “range score,” while the direct measurement is presented as a state of health, representing a measured indicator of the actual degradation of the battery’s cells.

Dealerships and shoppers, however, may not fully understand the distinction. Many will interpret the model-based estimate as a battery state of health representing the degradation of the battery, when in fact it is a prediction of driving range built from external assumptions.

Cell-level condition is invisible to a model.

Battery health problems usually begin at the cell level, a single weak cell or growing imbalance, long before they appear as lost range. Cell voltages and cell-to-cell variance can only be read directly from the battery management system. They cannot be derived from mileage, age, or climate, so a model built on those inputs is blind to the exact faults that produce post-sale failures.

An optimistic guess becomes the dealer’s exposure.

A model-based score is a prediction, not a measurement, and across 150 vehicles it ran 6.2 points optimistic, with gaps as high as 18 points. Such an overstatement is not trivial. It flows directly into how a vehicle is priced and represented to the buyer.

The exposure shows up in three places: