The +3.96e-06 improvement candidate is withdrawn: at verifier revision b3b7784 the leader's headline is accepted on 8 of 8 shards with the gate alive, while this hunt's candidate was refused on 4 of 8 and a descent at the candidate's own weights bottoms at 0.0078960, 1.5e-5 under the leader's floor, so the claimed pair-weight, pressure, and window headroom are withdrawn with it.
hunts/amtopa_ceiling (Record 68 of 98 in chronological sequence)
Guiding Question
Holding the AMTOPA exact-pressure construction fixed (its 17-term cosine window family, its span-capacity-2 pair-weight polytope, its fixed-total position pressure, its scalar-Gram block assembly), what is the supremum of the proportion that family can produce, is 0.6734164909714992949 at that supremum, and which constraint is binding when it stops improving?
Method & Verification
AMTOPA's own table builder and verifier were run at pinned commit 7253fdc and at the candidate.json revision b3b7784, then this hunt's candidate was refused on 4 of 8 shards and a stronger cut oracle re-solved the pair-weight LP.
Lineage & Relationships
Primary Sources (at pin 8fa46e134)
Editorial Notes
No HANDBACK at the pin. Question filled from the HuntSpec, with the two hyphen-clause asides rewritten as parentheses. RESULTS.md section 1 still prints the +3.96e-06 sentence as written, then marks it WITHDRAWN 2026-09-06; the headline follows the correction, not the uncorrected H1 title. The separate tip-replay finding (INCONCLUSIVE at HEAD because the convexity gate fires zero times) is retained in RESULTS.md and the case log and is not treated as a second disposition. Every constant inherits the unreviewed analytic bridge named in the mission.
Date Provenance
commit 6eb3408d6b5bf08291c703576eeee61d5eb09028, hunts/amtopa_ceiling/RESULTS.md, author 2026-08-24T17:27:10-05:00