William Millilaw

Multi-pipeline solver R&D; the halo-SAT rigidity test

Ran a dense two-week campaign in 2026 that put the modern arsenal against the puzzle on a single machine: snake placement, adaptive local search, SAT and MaxSAT, a constraint solver, a generative transformer, parallel tempering, partition crossover. Most of it he wrote up as refutations, negative results recorded so others need not repeat them. Two of his findings sharpen the community's picture of the ceiling: a replica freeze test showing the roots of the best boards return to the same cells when perturbed, and a halo SAT-residual test showing the record boards are strict local optima. This wiki reproduces the second on public boards.