Work is coordinated in issue 19 and issue 20. The full k ≥ 15 theorem is checked: in every finite k-uniform generating family with exactly three generators through each point and pairwise intersections of size at most one, every point belongs to at least 657/1000 of the distinct unions, including the empty union. The full high-girth bound is also checked for every k ≥ 3: every point has frequency at least 518208/1000000 when the incidence graph has no cycle shorter than sixteen. The cyclic span-four theorem remains incomplete. Li remains deferred until the other results are fully verified.
Complete large-generator bound
The accepted theorem is Frankl.Large.large_generator_frequency. For generators G : ι → Finset α, it assumes finite ι, k ≥ 15, card (G e) = k, exactly three indexed generators through every point, and card (G e ∩ G f) ≤ 1 for distinct indices. It concludes
657 * card (family (univ.image G))
≤ 1000 * card ((family (univ.image G)).filter (a ∈ ·))
for every ground point a. The family contains each distinct union once, including the empty union. It does not weight generator representations.
The proof derives the event bound and lopsided dependence by Harris slicing, counts the actual local selections, establishes the query independence, and transfers the canonical selections through a proved bijection. The exact 16-module closure passed strict source checks, all 231 theorem declarations' transitive axiom audit, and independent pinned-kernel replay at 663f72751f7cfb13e533050985072e59204cbd26. The source hashes and job link are recorded in the receipt. All modules in that closure are registered; the full merged-package run is a separate gate.
Earlier accepted components
The six initial modules passed their exact hosted gates at 57d699c770471fc898850079bde384a269da5d4f. They prove distinct-union semantics, exact local boundary fibers, the weighted local-certificate implication, the incidence-graph bijection, finite lopsided conditioning, and the numerical estimates. Initial receipt.
Complete Frankl high-girth frequency theorem
FranklHighGirth.high_girth_generator_frequency proves the full 518208/1000000 bound for every point of a finite k-uniform generating family, for every k ≥ 3, with exactly three indexed generators through each point and no incidence cycle shorter than sixteen. The counted family contains each distinct union once, including the empty union. Finite ground is derived from the stated regularity assumptions. The small-degree cases use all twelve checked K03–K14 cone certificates and actual graph-ball configuration counts; the large-degree case derives linearity from the girth assumption and applies the accepted 657/1000 bound.
The graph proof identifies the radius-seven incidence ball with its recursive tree, transports actual boundary-conditioned independent-set configurations, and averages the finite fiber bounds. There is no unproved recurrence, counting identity, finite-certificate assertion, or numerical bound in the final theorem's hypotheses. The girth and extended-egirth formulations and the frequency bound after removing the empty union are also checked.
At source commit f421f4434ffc8117a57a14b99c19818e76dcfd8e, all 64 project modules passed fresh strict compilation. The explicit import-all audit covered 1,363 owned declarations, including 909 theorems, and all 35 required endpoints. Exact replay with the pinned, unchanged stock checker passed in 72.750 seconds, with peak RSS 12,834,268 KiB. Every recorded source, object facet, compiler, checker, and gate hash was unchanged after checking.
Original receipt · Hosted run. The original receipt SHA-256 is e872e48c8199c0c713d6a9cbc50e3d9ef42d8e32ed62e418ddcd282e104f65cf. All modules in this theorem's closure are registered in the shared package. Its full merged-package build and replay are a separate gate. The cyclic span-four theorem remains incomplete, so the Frankl result entry remains partial. Li remains last.
Cyclic span-four proof in progress
The candidate sources prove local-certificate soundness, singleton/two-set and two-neighbor shortcuts, finite support/core enumeration, literal cyclic window semantics, and rank certificates for closed walks. The finite window weights and graph reductions still need their complete Lean checks and final assembly.
A separate finite computation checked all 38,044 eligible wrapped-window cases for cycle lengths at most ten with the original weights. This is a candidate route to removing the written proof's use of the published eleven-element theorem; it is not yet an accepted Lean proof. The three-common-point-triples theorem is being proved internally with exact finite union-closed extension certificates. No published theorem is silently treated as an axiom.