"Gen-4: a ledger that grades the battery industry — and a kill switch for itself"

The Innovation Lab's fourth generation was the second of the two specialist topics the operator tabled: solid-state battery interface degradation — what's theoretically mitigable, what's measured versus claimed, and what would a falsifiable improvement experiment even look like?

Specialist mode changes everything: scouts scope their domains to the topic, deliverables are dated write-ups as often as code, and "interesting" was defined up front as decision-relevant findings. This generation is the purest expression yet of what Mode B does.

What the scouts found

Three topic-scoped discovery cards: the interface science (interphase formation, dendrites, space-charge layers, chemo-mechanics, stack pressure), the characterization methods (impedance spectroscopy, dV/dQ diagnostics, accelerated-test protocols — and the manufacturer measured-vs-claimed disclosure gap), and the decisions that actually hurt SSB developers and integrators: cycle-life prediction, warranty risk, fast-charge limits, second-life economics.

The one that survived: the 2027 LEDGER

The critic scored four concepts hard; one survived outright. The 2027 LEDGER is a pre-registered measured-vs-claimed expectations ledger for the five SSB production claimants — QuantumScape, Toyota, Samsung SDI, Solid Power, CATL. Every public performance claim (energy density, cycle life, charge rate, production dates) is paired with:

1. a pre-registered evidence bar — which measurement, which protocol, published by when, counts as support; and 2. a dated falsifiable prediction — "cycle-life data public by date Y, or the claim is downgraded."

Unmet bars mean recorded downgrades in a public change-log. It's a scientific instrument pointed at a whole industry's press releases.

Two binding conditions made it an instrument rather than commentary:

- Inter-rater validation: two independent graders, blind to each other, graded the same 16 claims against the verdict taxonomy (MEASURED / PARTIALLY-MEASURED / CLAIMED-ONLY / CONTRADICTED). Raw agreement: 14/16 (87.5%); disagreements adjudicated to consensus verdicts. The agreement rate is published with the method. - A pre-registered kill switch: if the ledger earns zero independent citations by 2026-11-11, it is killed as an instrument and survives as internal documentation. The lab pre-committed to executing its own deliverable's death if nobody cares. The check is open as of this writing.

The v1 result: claimants cluster heavily on CLAIMED-ONLY, with the few MEASURED verdicts concentrated at the single company that publishes its own worst-case protocol comparison. That asymmetry is the finding.

The two that died at their gates

Mode B's discipline: check whether the data exists before building.

- CALSCREEN needed ≥3 independent dual-protocol (calendar + cycle) SSB cell systems in the literature to build a cross-comparable screening method. Exhaustive search found exactly 2 — one of which is the very paper the concept was built on. Killed as "the RSC paper relabeled," no build attempted. - P-MAP wanted to build a falsifiable pressure-transfer-function model from published cycling datasets. All five candidate datasets lack a recorded stack-pressure channel during cycling — verified directly against the UMD data mastersheet (19% not recorded, 38% not applicable, the rest nominal constants only). Battery Archive hosts zero SSB datasets. The model is untestable on public data — and that absence, documented with receipts, is itself a finding the field should care about.

Both kills got dated write-ups. Cost of the whole generation: a couple of dollars.

The pattern across four generations

Gen 1 shipped a product. Gen 2 shipped a product and validated the mutation loop. Gen 3 produced five research write-ups and a novel finding. Gen 4 built a research instrument with an expiration date and killed two concepts at their gates. The lab's negative results are as documented as its wins — and the cost curve is still heading down, not up.

All five Gen-4 write-ups are in the lab's public documentation: the ledger itself, the interface-degradation science survey, the characterization-and-disclosure scan, the developer-pains brief, and the disproof-scrape write-up. The ledger's adoption check runs to November 11th.

— The Innovation Lab team