The report summarizes measurements of the same exact offset sequences. The experiment guide describes the candidates, validation contracts, construction boundaries, external source, and regression.

manifest.json: measured source, binary, source-file hashes, external-source hashes, generated boundary-patch hashes, compiler, host, and timing schedule.
sequences.json: all 192 actual exported directories, six replay recipes, three boundary recipes, and exact sequence hashes/extents.
raw.tar.gz: every process CSV and stderr, including excluded warmups, failed/ineligible candidates, and export metadata. No timing observations were dropped. raw-sha256.json checks each original uncompressed member.
summary.json.gz: per-sequence/candidate/access statistics, process medians, immutable-array bytes, allocated capacity and object bytes, construction times.
checks.json: complete-group, hash, and cross-candidate checksum validation.
verification.json and qualification.txt: source qualification and the expected upstream Half boundary failure.
host.json: supplemental read-only host query, needed because sandboxed sysctl was denied during initial collection.
sha256.json: hashes of every other retained evidence file.

To regenerate the tables, copy this directory to scratch storage, unpack raw.tar.gz there, and run optional/select_compare/analyze.py on that copy. The analyzer verifies all original raw hashes, then regenerates summary.json, checks.json, and report.md. Regenerating the integer sequences themselves uses collect.py and the frozen source revision. Generated .seq files remain local; their exact hashes and recipes are retained here.

The post-measurement analyzer adds presentation tables only. The measured candidate/export code and executable remain those recorded in the manifest.