Original constraint-reconstruction-v0 development comparison. Read record/REPORT.md, ACCOUNT.md and MANUAL_REVIEW.md. All ten identities have prior exposure. The four policies, physical request caps4/8/150, two stochastic blocks, original240attempt barrier, selection rules and source/image reconstruction are preserved in prospective-source.zip. No new selection or capability claim is supplied by packaging. Unknown quality and usage remain unknown. Source hashes and solver receipts do not establish semantic correctness. The fixed review contains twelve declared actor-visible deliveries, with exact artifact hashes. Private histories, benchmark source, grader payloads and credentials are excluded. Review helper source is included for methodological transparency; it needs the original private histories to run. Reproduce all five tables from this archive alone with Python's standard library:
python -B -m scripts.display_constraint_reconstruction --study record --out reproduced --no-figure

Compare reproduced/TABLES.md, POINTS.csv, DECISIONS.md, COMPARISONS.csv and CHECKS.csv with record/display/. This displays all132saved contrasts and all17branches/39reference checks without resampling, rerunning tasks or making a new selection. Omit --no-figure to regenerate figures using matplotlib3.10.8; figure metadata may differ. Full experimental reproduction requires the external unchanged benchmark/baseline, declared pinned images and metered Luna credentials described in the prospective source archive. Mode50 is credited to Gemini's public implementation; historical image identity is unestablished. Post-freeze literature reading and presentation/review support are identified separately from the frozen method.