scripts/

Command collect-host-info writes the reproducibility metadata for a benchmark run (issue #141 "Reproducibility metadata") as JSON.

Command footprint records one implementation's operational-footprint row (issue #141 §"Operational footprint").

Command inspect-limits reports whether a running container actually received the resource ceiling the benchmark intended (issue #141 §7 requires the suite to "detect and report that fact rather than silently pretending limits were applied").

Command k6load runs the crud-list workload against a target for a fixed window and KEEPS + validates the k6 summary, exiting non-zero unless the load was a clean measurement (traffic sent, no HTTP errors, no failed checks — benchmarks/internal/k6's Summary.Validate).

Command microbench parses `go test -bench=BenchmarkFrameworkTax` output for one stack (read from stdin) and merges the rows into OUT/microbench.json, replacing that stack as a whole (so a re-run can't leave a stale scenario).

Command report regenerates the root README's `## Performance` block from the committed benchmark outputs, or (-check) verifies the committed README still matches — the drift guard for issue #141's "README is regenerable, never hand-edited" AC.

Command run-crud runs the headline CRUD-read workload (benchmarks/workloads/crud-list.js) against one already-running, already-seeded implementation and merges its rows into a results snapshot.

Command summarize reads a results.json snapshot and writes the human report (summary.md), generated from the structured per-trial rows — Markdown is never the canonical source (issue #141 §9).