Benchmark report
Twenty checked workload measurements
Twenty distinct AAYU core workload patterns were run at N=10,000. The runner checks expected results before recording timings and repeats each case five times.
Correctness checks
Every matrix case has a deterministic expected output. The runner checks it on warm-up and on every timed run. The report lists expected values and timing ranges.
What the time includes
Wall-clock process launch and bytecode loading are included. Compilation is excluded. These short runs are dominated by process startup, so the page makes no instruction-level speed claim.
Reproduce with benchmarks/run_workload_matrix.ps1. The original end-to-end and hand-built microkernel results are also recorded in the repository benchmark artifact, with their scope explained there.