Every define / precompute / compute / update method,
for every approach across every process group, run once against the generated WROASTED fixture
(site 1, one time step). Timing and allocation are single warm-call measurements — indicative,
not statistically robust like a proper benchmark suite.