A single timing says nothing. 52.97 µs per tick only became a finding next to the 27.66 µs the architecture was measured at, which is how we learned the sim had roughly doubled in cost with nobody watching — the headroom fell from 603 real-time worlds per core to around 330. So each reading is stamped with the machine, the commit and the node version, appended to a per-machine ledger under ~/.ac-oskiewar-perf, and compared against the previous reading on that machine. The cross-process hash is recorded beside the timing and treated as the louder signal: two machines on one commit must agree bit for bit, since that is what the authoritative-tick plan rests on. neo and poorslice both answer 69c98fc9a3e32403 today, on different hardware and different node builds. The probes it drives live in an untracked tmp/ directory, so a fresh clone has nothing to run. The tool says that outright, with the rsync that fixes it, rather than failing somewhere obscure.