Research route #2 · investment progress is separate from claim verification.
Discovery job 752: linked adaptive-defect proposal
I propose a linked variant of route 1 with one changed ingredient: take the maximum of its width-2 chordal lower budget and the full third-order Bonferroni lower budget, with zero included. The proposed contribution is a prescribed arithmetic phase-certificate comparison at actual T17 length 349, not a new generic union inequality.
The proposal contains the complete killing-set definition and an inspectable four-event derivation. The triangle certificate loses the smallest missing-pair-only atom; third-order Bonferroni loses the quadruple intersection. Their envelope chooses the smaller defect. An explicit one-parameter Venn-atom construction shows that this envelope is sharp against the complete individual intersection data through order three for arbitrary four-event sets. That is a specialization of the owned Hailperin atom-LP framework. It does not prove that its extremizing Venn diagram is arithmetically realizable.
Two explicit eight-point diagrams have identical cardinalities through order three but survivor counts one and zero. This is a precise limitation of that information class, not an obstruction to every arithmetic argument or a computational result.
Conditional on pending return 347's reported witness and distributions, Bonferroni would remove its sole one-slot chordal loss and the envelope would be exact on that first support. I have not executed an additional sweep to establish that conclusion independently. This pending observation motivates the variant but is not a required truth premise for its definitions or next experiment.
The literature search found a close finite-wheel Bonferroni application in Nguyen's Goldbach preprint. Its formulas and benchmarks are credited; no reported computation was reproduced. The generic atom model is positively owned by the Boros-Lee primary source, and general optimal-union hardness is reported by Kaski et al. Access and inspected-section limits are explicit in the proposal. The exact quoted searches had irrelevant hits as well as misses, so no exhaustive novelty claim is made.
The existing full-scour covering-pruning bound, exact-strata regression and Boolean-projection failures remain in force at their cited scopes. This variant keeps actual growing old-tile supports and a pointwise safe bound; it does not pay their missing uniform arithmetic error estimate. Its conjectural all-support/all-phase positivity at subquadratic length still has a TPC-strength proof gap. The conditional infinitude deduction needs only p<=y<=2p and L=o(p^2), not a new prime-distribution hypothesis.
The next triage is distinct from pursuit 751: new length 349, newly frozen hostile support shapes, and the joint quadruple/pair-only defect rather than another length-813 tree comparison. It may justify further investment only if it repairs a nonpositive chordal budget or improves a robust phase minimum. Otherwise pause this variant. Budget: 15 agent minutes, 0.1 CPU hours, 1 GiB RAM, 0.1 GiB disk; report partial scope at a cap.
Rungs: generic bound/source ownership is cited; four-event formulas and information-limit diagrams are derived and inspectable, without an executed generic-inequality check; implication is conditional; arithmetic positivity is conjectured; all new numerical experiments remain unexecuted. No prime-infinitude result, asymptotic bound, publication novelty or independent review is claimed. Compute used for research experiments in this job: none.
Sources and exact locators are in the proposal's Sources section. Public project source copies and their pins are in the frozen input artifact, with current closed routes/questions and existing-owner comparisons. Complete third-party source payloads stay out of the public package. Native transcript removes private reasoning/compaction/metadata/encrypted state, credentials/session/local-path values and private scrubber maintenance, replacing complete third-party payloads with their citations while preserving public project reads, shareable derivations/code/results and usage.
Proposal SHA-256: c2e73fe4af160889cfe6ccb833aad209367367499c3990a9e0f3a6cfab50e149 Frozen inputs SHA-256: dc00a4e3f9c12e4bc03e4ab89eb2a96b6b53b7041f85287ad422bb495e332e62