William Xu
This project develops and validates a forecasting and pilot-test-design tool for first-of-a-kind (FOAK) mineral processing technologies, using direct lithium extraction (DLE) as the target domain and uranium in-situ recovery (ISR) and industrial water-treatment ion exchange as donor domains. FOAK process plants systematically miss their feasibility-study forecasts — ramp durations, capital costs, and recoveries deviate in patterned ways — and pilot-testing campaigns are rarely designed to buy down the uncertainties that actually matter to the build/redesign/delay/kill decision. The project builds candidate forecasting architectures on reference-class, mechanistic, and multi-fidelity foundations, and subjects them to a pre-committed, multi-tier validation battery: synthetic stress tests, a strict walk-forward hindcast over the ISR fleet, a cross-domain transfer hindcast to DLE, retrospective pilot-design audits, and — the subject of this registration — prospective forecasts registered before outcomes exist. Expected outcomes: calibrated skill scores for each architecture against pre-registered baselines (including industry feasibility-study forecasts and reference-class adjustment), an audit of whether model-guided pilot design would have changed real gate decisions, and a public, falsifiable record of prospective forecast performance on operating and under-construction lithium and uranium projects through 2029. Results will be published whether favorable or not; pre-committed failure disclosures are part of the protocol. Released under CC BY 4.0. The license covers the authors' original contributions; cited third-party sources (issuer filings, regulatory records, published datasets) are referenced by citation and remain under their own terms.