Furkejuvvon:
| Váldodahkki: | |
|---|---|
| Materiálatiipa: | Recurso digital |
| Giella: | |
| Almmustuhtton: |
Zenodo
2026
|
| Fáttát: | |
| Liŋkkat: | https://doi.org/10.5281/zenodo.18842736 |
| Fáddágilkorat: |
Lasit fáddágilkoriid
Eai fáddágilkorat, Lasit vuosttaš fáddágilkora!
|
Sisdoallologahallan:
- <p>Hard problems fail in two predictable ways: we overfit to tractable corners, or we accept persuasive narratives that cannot be replayed. Our work proposes a receipt-first, replay-first program for P vs NP in which the unit of progress is a subfield: a regime-bounded slice of the problem with locked scope, admissible transforms, invariants, falsifiers, and a solver proofpack. Progress becomes promotable only when accompanied by verifiable witnesses and append-only receipts that support clean-room replay. We formalize a universal budgeted hypothesis (UCCB(k)) that, if satisfied under a locked witness menu, yields a polynomial-time decision route for SAT. We then show why expander-coupled families stress this hypothesis by collapsing separator- and treewidth-style witnesses and by forcing frontier growth that breaks locally verified elimination-style compilation traces. Finally, we map the frontier cleanly: proof-carrying decisions (W6′) re-express the core question as universal polynomial certificate generation, while DRIFT-style proofpacks can provide honest intermediate deliverables (e.g., SAT ∈ BPP) without overclaiming P = NP. The result is a Clay-compatible research loop: it produces publishable negative tracks and transparent conditional frontiers, and it prevents illusory “near-proofs” via explicit stop rules.</p>