perf_tests/ceiling.py run from three machines at the same time against the 20-pod staging service (image 10142026-4446caf) as a profile with no rate limit, with pod_watch.py on the cluster side for staging and for production. Analysis: docs/performance-report.md, section 5a.
merged/: the three machines combined (levels.csv, summary.md) and the pods’ side per step (cluster.md); studio/, mini/, laptop/: each machine’s own levels.watch-staging/watch.csv, watch-prod/watch.csv: the pods every 5 seconds (the large per-pod files are not committed).