The Erdos-Turan factor 2 per prime cannot be improved to (2-delta) by joint or reordered prime processing: the architecture reduces entirely to c(A), the largest subset whose pairwise sums are powers of 2, which satisfies c(A) in {1,2} and is defined order-blind, so the selection lemma with phi(k)=(2-delta)^k is logically equivalent to the target bound rather than a refinement of the 1934 bookkeeping.
hunts/support_517b887f (Record 74 of 98 in chronological sequence)
Guiding Question
Can the Erdos-Turan 1934 factor of 2 per prime be replaced by (2-delta)^k by processing primes jointly, retaining valuation information, or changing the order?
Method & Verification
The 1934 proof was reconstructed into a base case c(A) <= 2 plus one binary residue choice per odd prime, then two sketch defects were witnessed by exact integer sets re-verified by trial division, with a 104183-set sweep kept as bounded-box evidence only.
Lineage & Relationships
Primary Sources (at pin 8fa46e134)
Editorial Notes
No HuntSpec.question field; question filled from the bounded question in MISSION.md. Disposition maps settled, negatively to completed. The two defect witnesses are about the literature sketch, not about the theorem, which stands. The exhaustive sweep is not a proof of the repaired lemma.
Date Provenance
commit b68e1ae603d32c93dced652c76b38cb6424d9ee8, hunts/support_517b887f/RESULTS.md, author 2026-08-24T20:56:27-05:00