Rebuilding o9_leaf.py's 1-D O9 walk on kernel leaves gives 476 cells at max depth 22, against the recorded 344 at depth 20, and 85 of those 344 cells (24.7%) fail outright; 476 is a measured prediction, not a kernel verdict, because no Lean was built here.
hunts/r_2926e4 (Record 20 of 98 in chronological sequence)
Guiding Question
What is o9_leaf.py's 1-D O9 cell count when its transcendental leaves are computed the way BandCert/Leaves.lean computes them, rather than with Arb?
Method & Verification
o9_leaf.leaves was swapped for o9_leaves_kernel.kernel_leaves with the rest of the walk left byte-identical, and bucketing the 85 failures into emit_lean's 40-cell chunks reproduced the 2026-08-13 Lean build's nine-chunk verdict without using that comparison to tune anything.
Lineage & Relationships
Primary Sources (at pin 8fa46e134)
Editorial Notes
Headline rewritten from the handback to keep the no-Lean caveat in the same sentence. Evidence is measured, not kernel-checked: the hunt forbids claiming a kernel verdict without a Lean build. Issue #23 correction (the 344-cell table had already been put to the kernel on 2026-08-13) is retained in the method, not treated as the headline.
Date Provenance
commit 53c629443402f7eb35a3a3e40437d56ee38844ed, hunts/r_2926e4/RESULTS.md, author 2026-08-14T11:15:35-05:00