Verify this battery report yourself

Equally weighted over 96,480 synthetic data points. Read that honestly: all three methods run here only because we generated this data ourselves. On the one dataset we did not generate - Oxford - the OCV method cannot run at all (0 of 519 reference points), so our method combination has never been validated on real cells as a combination. The ±0.8 pp above is the variance floor in our fusion code, not a measured uncertainty; the interval we can actually defend is a conformal ±0.60 pp, derived from held-out error on Oxford. That interval survived a self-audit on 2026-08-29 that raised our headline accuracy from 0.80 to 0.93 pp: the defect was an unconditional coulombic-efficiency division on discharge windows, and the conformal figures sit on the corrected path it never touched. Agreement on synthetic data demonstrates the pipeline; it does not prove accuracy. The signature below proves this report was not altered.