The development machine runs a resident inference server at ≈1500% CPU. make bench passes, but numbers measured under that load are polluted and not comparable across runs; the formal baseline is frozen at 2026-08-05 (benchmarks.md, design/benchmarks.md).
This is an environment problem, not a code bug — nothing in the repository is wrong. Do not treat this as a defect report.
Acceptance (when the machine is quiet enough)
- Re-run
make bench on a real-disk path (GOR_BENCH_DIR) under clean load and update benchmarks.md with the fresh numbers and date.
The development machine runs a resident inference server at ≈1500% CPU.
make benchpasses, but numbers measured under that load are polluted and not comparable across runs; the formal baseline is frozen at 2026-08-05 (benchmarks.md, design/benchmarks.md).This is an environment problem, not a code bug — nothing in the repository is wrong. Do not treat this as a defect report.
Acceptance (when the machine is quiet enough)
make benchon a real-disk path (GOR_BENCH_DIR) under clean load and update benchmarks.md with the fresh numbers and date.