Complete CLI comparison archive

All 18 workers are retained: three alternating pairs for each of help, validation and probabilistic refinement. The benchmark report defines the scope and normalization. protocol.json binds the input hashes, commits, frozen source manifests, launcher, Python and thread settings. attempts.json retains every duration and exit code; summary.json reports their medians.

sources-*.tar.gz contain only Calibrex Python sources. The complete raw result, stdout/stderr and actual import paths are kept for every worker. The frozen imports identify the executed code; the result's existing Git field identifies the enclosing checkout. The input geometry and generated correspondence files are excluded. fixture-protocol.json retains their construction and source identities. schema-parity.json binds 130 byte-identical schema/ledger exports to both measured source manifests and the retained export checker.

validation/ retains the 155-test correctness run, the executed test sources, the documentation build log and a receipt confirming that all 377 runtime sources still match the measured candidate after newline normalization.

index.json hashes every other file. Files use their original bytes, including empty stderr files. Run the CPU-only audit from the repository:

python -m tools.audit_lazy_runtime_benchmark docs/assets/lazy_runtime_loading