Physical AI for the battery economy
Know what an EV battery is actually worth.
The battery is the collateral in an EV loan, the payout in a claim, and the price of a resale, and all three are priced blind today. Ionscore reads pack telemetry and returns calibrated state of health, value and risk in seconds, with a confidence band on every number and a certificate anyone can verify.
Trained and measured on synthetic fleets grounded in public battery ageing data. Field validation with a design partner is the next milestone, and we publish the expected transfer discount rather than waiting to be asked for it.
80.0% of held-out packs fall inside this band, against an 80% target. Measured on 149 packs the model never saw.
Arrhenius
Physics-based degradation model, not curve fitting
PyBaMM
First-principles electrochemical reference
CQR
Conformalized quantile regression, P10 / P50 / P90
1.58pp
Held-out mean absolute error
The stakes
The battery is the loan. And the loan is priced blind.
A battery is 30 to 50 percent of an electric vehicle's value, and it degrades invisibly. When a lender writes a loan against that vehicle, the battery IS the collateral, and today there is no standard way to know its condition.
So the market prices the uncertainty instead of the asset: EV loans in India run 10 to 30 percent lower on loan-to-value and 1 to 9 percent higher on interest than the identical petrol loan, insurance premiums sit 20 to 30 percent above petrol cover, and resale values collapse against warranty thresholds no buyer can verify.
Every one of those gaps is the same missing number. Ionscore produces it.
The product
One measurement, three ways to act on it.
A number you can defend
Calibrated state of health as a P10-P50-P90 interval, residual value against your own price basis, and a signed reason code for every driver. Built for credit files and claim disputes, where a bare number does not survive.
80.0% measured band coverage, published
Proof anyone can check
A verifiable battery certificate with a public verify page and a one-line badge that embeds in any listing. Unguessable id, revocable the moment it stops being true. The trust artifact a used-EV transaction is missing.
Issue, verify, revoke: live in the product
The whole book, watched
A lender console that values every position from your LTV policy, monitors the portfolio, and raises covenant alerts when a pack breaches its floor, shows abuse, or goes silent. Deterministic, so every alert is auditable months later.
Valuation, portfolio, covenant alerts: live
The case
Why this number, from us, holds up.
The only published calibration in the category
Every competitor publishes accuracy claims sourced to themselves, and none publishes a calibration curve at all. Ionscore reports a measured coverage rate: 80.0 percent of held-out packs fall inside the stated band, against an 80 percent target, on 149 packs the model never saw. A lender can size exposure against that. Nobody can size exposure against a brochure.
Physics underneath, not just patterns
Underneath the learned layer sits a degradation model with a physical form: Arrhenius calendar and cycle fade with fitted activation energies, grounded on real NASA cells and cross-checked against a first-principles electrochemical reference. A telemetry-only model inherits its fleet's biases; a physics-constrained one has to obey the chemistry.
Aligned with the rulebook being written
India's draft Battery Pack Aadhaar guidelines define battery status in health bands and name financiers, insurers and used-vehicle buyers as the stakeholders, while stating that the validation methodology is still to be developed. Ionscore already reports those A, B, C bands, runs in-region, and returns a score rather than a lending decision, which keeps every user of it on the right side of RBI's outsourcing line.
Neutral by construction
Ionscore does not lend, insure, trade certificates, or recycle. That is what lets both sides of one transaction act on the same number: the lender and the borrower, the seller and the buyer, the OEM and the claimant. A scoring layer with a position in the trade is just another counterparty.
The mechanism
Telemetry in. A defensible number out.
- 01
Read what the pack already reports
No new hardware, no teardown. The signals a battery management system already emits: voltage, current, temperature, charge behaviour, cycle and calendar history.
- 02
Infer the state the sensors cannot see
Two layers recover the latent health: a physics degradation model fitted to real cells, and a learned layer that serves it in milliseconds with a calibrated interval instead of a guess.
- 03
Act on it anywhere
The same score arrives as an API response, a console view, or a public certificate. One number, consistent across the loan desk, the listing, and the claim.
Put a number on your battery book.
Tell us about your fleet or portfolio and we will set up a pilot: calibrated state of health, value, and risk on a sample of your packs.