# Exact phase-domain obstruction certificates for prime-band windows No uniform arithmetic refutation rule is known, so this is a proposed finite instrument, not an infinitude proof. The general residue-choice ILP, search pruning, cardinality encoding and checked SAT proof formats are owned. The proposed new test retains the full shared phase of each prime on one irregular old-twin support, beyond the separately maximized capacity bound and without enumerating the enlarged period. ## Changed ingredient Route 2, return #348, used low-order union moments. Its prescribed L349 triage, return #350, reported no added benefit over chordal certificates and paused that sample. Those pending measurements are motivation only, not mathematical premises. This linked alternative replaces a moment envelope by a complete finite-domain cover constraint. It does not erase the earlier failed criterion, rerun its 16 supports, or claim the encoding is a new theorem. For W=p# and an interval [a,a+L), let D be its complete old-twin starts s with gcd(s(s+2),W)=1. Let Q contain all primes p0. Record all values needed to verify that it is first. F1 need not be monotone, so do not find L_F by binary search. If no such prefix occurs within the cap, report that limitation and stop without expanding the search. Solve only the full-phase cover formula at L_F-1 initially. Success is a checked UNSAT refutation with an independently rebuilt formula while F1<=0, requiring at most 120 CPU seconds for solving/conversion/checking and at most 4 MB for the refutation. This gives a strict finite improvement over the separable obstruction on this one support. It does not establish an improvement over all Hunter/chordal/third-order certificates, nor determine a global gap. Failure is an independently checked covering model at L_F-1, a resource limit, an invalid certificate, or no frontier. A model followed by the known positive-capacity obstruction at L_F fixes the exact first noncovering length for this nested interval; SAT coverability is monotone under adding slots even though F1 is not. Timeout is unresolved, never a covering witness. Stop after the single prescribed test; do not substitute another start or longer cap to rescue a negative. If successful, the distinct next investigation is to extract the conflicting phase core and determine whether its arithmetic spans permit a rule reusable at another p, not to repeat a census. Weakest assumption: a compact conflict proof carries reusable arithmetic structure rather than instance-specific search. The pilot tests compactness and a finite gain only. Budget: 0.25 agent hours, 0.15 total experimental CPU hours, 2 GB RAM, 0.1 GB disk, one solver thread. Tools: Python 3, a proof-producing SAT solver, DRAT-to-LRAT conversion if needed, and an independent LRAT checker. Record exact tool versions/commits before any run. Checker and encoder validation have a separate 5-minute judgment allowance inside the agent budget. The proposal stage executed no numerical experiment or solver and charges 0 experimental CPU. ## Prior work and search record Search date 2026-09-14. I read the current questions and Closed routes in OUTCOMES, SEARCH-CONVENTIONS' covering-optimum row, current routes 1-3, and reused the prior searches recorded in #346/#348/#354. Exact updates included `Jacobsthal function SAT solver covering residues paired progressions integer programming`, `"Jacobsthal" "satisfiability"`, `"Jacobsthal" "integer programming"`, `"two class" "Jacobsthal" "SAT"`, `"A144311" "193"`, `"two-class Jacobsthal" "193"`, and `"Jacobsthal" "LRAT"`. Empty exact searches do not establish novelty. - Ziller and Morack, *Algorithmic concepts for the computation of Jacobsthal's function*, arXiv:1611.03310v2, 2017, §2.2 Eq.(2.1), already give binary residue variables, one choice per prime and a cover constraint per point, with nested-length ILP search. §2.3 gives capacity pruning. I inspected these primary HTML sections. This owns the generic formulation and frontier search, although its stated object is one-class consecutive integers. [Primary text](https://arxiv.org/html/1611.03310). - Carter's OEIS A144311, with extensions by Alekseyev and Wang, is the two-class Jacobsthal owner. Wang's linked C++ program, `dfs` lines 9-47, chooses the linked pair of residues and bounds the remaining capacity. I inspected the entry and complete code. Their published values and program are cited, not reproduced. The new prototype uses an irregular 97-sieved support and a 97-to-193 band, with exported proof objects; this search found no matching certificate for that instance. [Entry](https://oeis.org/A144311), [code](https://oeis.org/A144311/a144311.cpp.txt). - Sinz, *Towards an Optimal CNF Encoding of Boolean Cardinality Constraints*, CP 2005, §2, author PDF p.2, explicitly gives the k=1 sequential counter. I inspected the author-hosted text. This is an import, not a new cardinality encoding. [Author PDF](https://www.carstensinz.de/papers/CP-2005.pdf). - Cruz-Filipe, Heule, Hunt, Kaufmann and Schneider-Kamp, *Efficient Certified RAT Verification*, CADE 26, 2017, pp.220-236; inspected author manuscript §3 pp.4-6, Figures 1-3 and §4 opening p.7. It owns LRAT's hinted certificate checking and separation of solver/converter from checker. No published timing or example was reproduced. [Author manuscript](https://www.cs.cmu.edu/~mheule/publications/lrat.pdf). - Patrick White + Claude, *Erdős problem #688 — live check, exact reduction, upper bound, and finite certification*, AI-disclosed working report dated 2026-07-27, §§4-5, already combines CP-SAT scouting, exact residue search and a proposal for checked DRAT/LRAT proofs, explicitly warning that finite extensions do not give uniformity. I inspected the report, not its linked program or its claimed table. Its one-class consecutive-integer problem differs from the old-twin support here. This source is an attributable prior attempt, not independent mathematical verification. [Working report](https://erdosproblemaday.com/report/688). - Nguyen, *Finite-Window Noncovering on Primorial Wheels: Higher-Order CRT Bounds and Shift Correlations*, Preprints.org 202608.1299v1, 2026, §3.5, reused from #348's primary-source inspection; owns the nearby finite-wheel/tree/third-order framework. No new read or reproduction of those examples. [Preprint](https://www.preprints.org/manuscript/202608.1299). Access gaps: paired-Jacobsthal ancillary algorithms and the #688 program were not inspected; no checked SAT/LRAT toolchain was installed or run. The broad search also surfaced unrelated odd-covering claims, which were not used as evidence. No claim is made that SAT-based covering or this CNF translation is absent from the literature. Calibration: encoding and conditional implication are elementary adaptations of named methods. The finite payoff, compactness and reusable arithmetic rule are conjectured. Pending earlier measurements are not dependencies. The only proposed new deliverable is the frozen finite certificate test and, if it succeeds, an inspectable arithmetic conflict core.