- Runnable FusionSim benchmark and source/model registry
- RunSpec/ResultSpec contracts, fixtures, tolerances, hashes, and release container
- Source transcription notes and discrepancy log
- Engineering support for reproduction and archive packaging
Bounded university collaboration
Choose one work package, not an entire platform
Each package begins with an intended use, frozen artifacts, explicit partner effort, a validation or verification domain, decision rights, and acceptance criteria that permit an unfavorable result.
No institution, investigator, reviewer, dataset, approval, endorsement, or collaboration is claimed here. These are ready-to-negotiate protocols; external participation remains an open gate.
Recommended first ask
WP1 — Nuclear Data and Reactivity Reproduction
Research question: Can an independent team reproduce selected p–¹¹B reactivity results from versioned data and equations, inside explicit domains and tolerances?
- Independent source interpretation and implementation
- Review of intended use, source domain, transformations, and uncertainty
- Blind comparison before discrepancy resolution
- Signed result/review record scoped to the exact immutable artifact
| Weeks | Activity | Deliverable |
|---|---|---|
| 1–2 | Freeze intended use, source set, protocol, and tolerance | Signed reproduction plan |
| 3–4 | Independent source transcription and pedigree review | Reviewed source records |
| 5–6 | Checkpoint and curve reproduction | Independent result bundle |
| 7–8 | Domain, sensitivity, and uncertainty review | UQ report |
| 9–10 | Resolve differences without erasing them | Issue and decision log |
| 11–12 | Freeze and archive | Benchmark report and immutable package |
Acceptance: exact source versions and transformations reproduce; independent results are compared before tuning; all discrepancies remain visible; the validation/reproduction domain is explicit; conclusions do not extrapolate beyond it.
Alternative first ask
WP2 — Alpha Diagnostics and CR-39 Validation
Research question: How accurately and repeatably can one frozen AlphaTrack model count declared real CR-39 track classes under one controlled acquisition protocol?
- Licensed, partner-controlled real CR-39 data
- Plate-level split with no plate or batch leakage
- Independent expert annotation and adjudication
- Statistician-reviewed power/precision and abstention plan
- Local-first processing, custody, retention, and deletion terms
- Weeks 1–3: intended use, acquisition protocol, classes, shift risks
- Weeks 4–6: annotation study, precision planning, preregistration, freeze
- Weeks 7–9: blinded acquisition, annotation, and sealed evaluation
- Weeks 10–11: unblind, primary and failure/subgroup analysis
- Week 12: archive a positive, negative, or inconclusive report
Acceptance: no plate/batch leakage; prespecified primary metric and interval; human and model disagreement remains auditable; all deviations and abstentions are reported; the result is released whether favorable or unfavorable; any claim is limited to the tested detector and acquisition domain.
Supporting option
WP3 — Reduced Plasma and Orbit Model Verification
Research question: Does one tightly defined HBF orbit model converge and agree with analytic cases and an independent solver for the same mathematical problem?
- One analytic field model and coordinate convention
- One integrator configuration and error metric
- Manufactured/analytic cases and refinement sequence
- Finite-window and right-censoring semantics
- Analytic fixtures pass
- Declared refinement behavior is demonstrated
- Partner implementation is independent
- Blind comparison precedes reconciliation
- Disagreement and applicable domain are published
Governance before data
Terms that must be agreed
Scope, staffing, data and code licenses, publication timing, authorship, foreground/background IP, confidentiality limits, safety responsibility, conflict disclosure, use of names/logos, retention and deletion, security incidents, and the right to publish negative results.