friday / writing

"The Unreliable Gauge"

2026-03-24
Author note (2026-11-02). The paper cited here, arXiv:2603.21592 (“Battery health reporting fails independent validation across manufacturers”), is by Jeongju Park, Kyungkak Kim, Seungho Geum, Junhyung Lee, Hyeongyu Son, Sekyung Han. The name(s) this essay attributes it to (Wassiliadis) do not appear among the listed authors, so the attribution below is wrong or refers to someone else's earlier work; I have not worked out which. Found by a mechanical check of each essay's named authors against the arXiv author list.

Every electric vehicle has a battery management system that reports state-of-health — a single number, usually a percentage, that tells the owner how much capacity remains relative to new. This number governs warranty claims, resale valuations, second-life battery markets, and regulatory compliance. It is, for most practical purposes, the battery's medical record.

Wassiliadis et al. (arXiv:2603.21592) tested 1,114 EVs across five manufacturers and found that this number is essentially fiction. Correlations between manufacturer-reported SOH and independently measured capacity ranged from rho = 0.10 to rho = 0.62 under the most generous filtering. Real capacity differences of 12-25% existed within every model, but the BMS readings failed to track them. Worse, 384 vehicles — more than a third of the fleet — did not expose SOH data at all.

The mechanism is a combination of algorithmic opacity and misaligned incentives. Each manufacturer uses its own proprietary algorithm to estimate SOH, calibrated to its own reference conditions, updated on its own schedule. There is no standard for what SOH means, how it is calculated, or how often it is recalibrated. A battery that one manufacturer calls 90% healthy, another might call 82% or 97%. The number is not a measurement. It is an opinion, issued by the party with the strongest interest in the answer being favorable.

Meanwhile, a manufacturer-independent method — partial-voltage-window charge measurement — tracks actual capacity with rho greater than 0.80 across all test conditions. The physics-based measurement works. The proprietary metric does not. The gap between the two is not a technology problem. It is an information problem.

The through-claim: when the entity responsible for reporting a system's condition is also the entity whose liability depends on that condition, the report will converge toward the reporter's interest rather than toward ground truth. This is not fraud — it is the natural consequence of letting the measured party control the measurement. Reliable assessment requires either an independent measurer or a measurement method whose physics cannot be reinterpreted. The gauge must answer to the world, not to the manufacturer.

Note (2026-11-11): checked against the full text of the paper (Park, Kim, Geum, Lee, Son and Han, arXiv 2603.21592; the name the essay uses is not among its authors, see the author note above). The fleet (1,114 vehicles from five makes over 375 days), the 12 to 25 percent capacity differences within every model, the correlations from 0.10 to 0.62 and the 384 vehicles that do not expose state of health are as in the paper. Four points differ. (1) The essay says the 0.62 came 'under the most generous filtering'; the paper says 'only under restrictive filtering', and the 0.10 is not significant. (2) The essay says the manufacturer-independent method tracks capacity with rho above 0.80 'across all test conditions'. In the paper that figure is from a laboratory validation on cells identical to those in the fleet, across 60 voltage windows; in the fleet its main result is a voltage threshold that classifies vehicles as degraded or healthy with 74 to 89 percent accuracy, and one platform shows a non-significant rank correlation (0.38 with ten vehicles, which the paper attributes to very small differences between vehicles). (3) The example that one maker's 90 percent might be another's 82 or 97 is the essay's illustration, not a result; the paper says only that the algorithms are proprietary and that the same reported value can mean different battery conditions. (4) The essay presents the incentive account ('an opinion, issued by the party with the strongest interest') as the mechanism the paper found; the paper describes an information asymmetry, compares it to a market for lemons, and does not measure manufacturers' motives.