About Scale Prognostics
We build physics-based battery degradation prediction for engineering teams that need answers they can stake real decisions on — procurement commitments, warranty terms, fleet deployment schedules.
The Founder
Jason brings over a decade of experience working with electrical infrastructure and real-world degradation across extreme operating environments. That fieldwork built something most battery modelers never develop: an intuition for how cyclic thermal and mechanical stress compounds over years of service — and where conventional models quietly get it wrong.
The core differentiator behind Scale Prognostics came from observing a gap in commercial aging models. Batteries don't just degrade from charge-discharge cycling — they degrade from the temperature oscillations that real-world deployment creates, and that mechanism is mathematically independent of the cycling itself. Jensen's inequality on the convex Arrhenius rate function guarantees that any constant-temperature model will systematically underpredict degradation. This is a mathematical theorem; we build the correction into every simulation by tracking thermal variance as an independent state variable.
The Engine
Scale Prognostics v18.3 couples ten degradation mechanisms — SEI growth, lithium plating (transport-limited and resistance-triggered cascade), particle fracture, porosity evolution, electrolyte depletion, cathode structural degradation, thermal fatigue, contact loss, crack-SEI feedback, and graphite exfoliation — in a 24-element state vector with 39 physics-derived parameters.
Performance: calibrated against 8 published silicon-graphite datasets — 6 at A+ (RMSE 0.13–0.32 pp) and 2 at A on knee-prone cells (RMSE 1.01–1.23 pp); mean 0.44 pp. Held-out forward-prediction RMSE on knee-prone cells is materially higher (~12 pp at 70% retention cutoff) and is the focus of ongoing architecture-sprint work — we publish that alongside the calibrated number because it's the metric that matters for the cycle-life forecasting use case. ~30ms simulation time per 1,000 cycles. Physics-first architecture, calibrated against literature; predictions adjust when conditions change because the mechanisms adjust.
Our Approach
We don't compete on data volume or training set size. We compete on physics. Our model extrapolates because the mechanisms are real — they don't stop working outside the training distribution. When you change the temperature profile, the C-rate, or the silicon content, the predictions adjust because the physics adjusts.
This is the difference between a model that tells you what happened to batteries like yours and a model that tells you what will happen to yours.
AI-Augmented Development
Scale Prognostics is built with AI at every stage of the development cycle. AI-assisted derivation and symbolic verification of coupled degradation equations ensures mathematical rigor. Continuous automated cross-validation against published datasets catches regressions before they reach production.
The result: a model that improves faster than any single researcher could push it, with mathematical rigor that no pure-ML approach can match.
Talk to Us
Questions about the engine, pricing, or enterprise deployment? We'd love to hear from you.
Scale Prognostics LLC — Atlanta, GA