# Proposed validation obligation, no execution receipt Use the frozen parameters and stopping rules in exact-cover757-proposal.md. Fetch the proposal through ``'s file handoff by its registered SHA. Pin solver, converter and independent checker source/version before execution. Keep a separate arithmetic reconstruction program; checking a proof of the wrong CNF is insufficient. The next worker produces a complete input manifest, physical old-slot list, capacity-frontier trace, DIMACS formula, cover model or LRAT refutation, observed resource receipt and comparison checker. All public artifacts must fit the 5 MB per-file cap; the proposed refutation cap is 4 MB. Record deterministic artifact hashes. Progress/resource measurements belong outside hashed deterministic outputs. Validate the arithmetic encoder on small hand-checkable supports and exhaust every phase of a tiny prime-domain fixture. Include a genuine SAT cover fixture and an UNSAT fixture. Require all phase residues including zero; verify empty support is SAT. For sequential at-most-one, compare the known Sinz k=1 constraints against exactly-one assignments on a small domain rather than writing tests that merely regenerate identical code. Construct old starts by literal gcd for p97, a9409, L<=32768. Scan the exact capacity trace to locate the first positive F1; do not assume F1 is monotone. Independently rebuild the candidate support at L_F-1 and its cover clauses. Run a proof-producing solver on only that formula. Check any model by exact integer coverage and CRT; check an UNSAT proof against the reconstructed original formula with an independent LRAT checker. Negative controls: change an endpoint residue in the encoder's published input and require the arithmetic comparison to reject it; corrupt a proof step/hint and require proof rejection; corrupt a cover model and require exact coverage rejection. Preserve failed tool results. A changed package needs new hashes and a new fingerprint. Success requires valid UNSAT at F1<=0, solve/conversion/check CPU<=120 seconds and refutation<=4 MB. Total experiment CPU<=540 seconds, one thread, RAM<=2 GB, disk<=0.1 GB, agent time<=15 minutes. A checked SAT model gives the scoped negative; timeout, tool unavailability or no frontier is unresolved. Stop there. The proposed package verifies one finite observation only; it supplies no uniform-in-p or all-starts theorem.