0749 — Benchmark runner lives at the .bench root
Edit this pagePosted: 2026-09-11 · Report this thread
Superseded by 0755 for command naming; the root layout and evidence preservation continue unchanged.
Context
The reliability lab is the shared running system for benchmark scenarios. Its extra directory puts that system beside older independent experiments.
Decision
- Promote the lab packages, entry point, scripts, Compose stack, and results
from
.bench/reliability/to.bench/. - Keep
just reliability-labandjust reliability-reportas the commands. - Preserve recorded evidence bytes and frozen source layouts; update current evidence links. Historical reproduction commands target the frozen source.
- The evidence retention policy in 0748
continues at
.bench/results/published/. - Classifying or migrating the other benchmark programs is separate work.
Consequences
New scenarios extend the existing root runner. Imports, Docker build paths, and native source capture follow the shallower directory. Existing workloads and verdicts are unchanged; old result paths in historical records describe where those artifacts lived when recorded.