Original prefix-alternatives-v0 development comparison. Read record/REPORT.md, ACCOUNT.md, MANUAL_REVIEW.md and the disclosed filename-only monitoring departure. All identities were previously exposed; stochastic blocks are not independent identities. The original source, freeze, policies, caps, rules and environment are in prospective-source.zip. The archive preserves original saved scalar rows, aggregate readout and once-only account; it does not rerun scientific analysis or change selection. Missing outcomes and usage remain unknown. Private histories, task source, grader payloads and credentials are not included. Reproduce all saved tables and the figure from this archive with matplotlib installed:
python -B scripts/render_prefix_alternatives.py --study record --out reproduced
Compare POINTS.csv, CONTRASTS.csv, DECISIONS.csv, CHECKS.csv and DISPLAY.md against record/display-original/. Figure metadata may differ. Full experimental reproduction requires the unchanged external benchmark, baseline, declared images and metered Luna credential described in the source archive. The reconstructed Mode50 image is not claimed historically identical to the peer image. Inspect the report and archive for scope and content before public upload.