Lab notes

What we measured, and what it does not show.

Each note is one piece of work written up: the question, how it was run, the numbers, and the reasons not to read too much into them. None of it is peer reviewed and every note says so at the top.

  1. 01Which small model should run your Mac launcherFifteen small models, the same 55 jobs on a simulated Mac, scored on what the machine ended up doing. The cheapest scored highest. Refusing an impossible request is what separated them.2026-09-19