Open work
Questions worth testing next.
ProofX grows through reproducible experiments, independent checks, and corrections to its assumptions.
Research priorities
| Question | Useful contribution | Evidence to attach |
|---|---|---|
| Does GB-001 reproduce elsewhere? | Run the published configuration on another platform. | Manifest, ledger hashes, verifier output |
| Does the result survive a different size range? | Prespecify held-out intervals and more seeds. | Protocol, every run, size-stratified comparisons |
| Which structural family helps? | Remove one family at a time under equal budgets. | Ablation configuration and complete ledgers |
| Does Collatz ranking beat a baseline? | Define an independent anomaly target and compare selection policies. | Matched domains and budgets; treatment of shared suffixes |
Tooling priorities
Resumable long-run search, external calibration evaluation split by run or family, and incremental tests for older excluded modules. See the quality-scope inventory before building on a legacy module.
Available now
GB-001 includes a protocol, equal-budget baselines, a full reference, and independent recounting. Calibration reports held-out annotation metrics. Search budgets and sieve bounds are enforced. The Lean layer checks finite trajectory and prime-pair certificates.