INDEPENDENT REPRODUCTION AND OUTPUT COMPARISON

Purpose
This directory contains the independently regenerated simulation dataset and
analysis outputs produced during the reproducibility audit of release v1.1.
It is not a separate scientific experiment and does not replace the primary
results/ directory.

Simulation
- Master seed: 20261029
- Regenerated run-level rows: 4,520
- Experiments: E1 (2,520), E2 (1,000), E2b (1,000)
- Full rerun completed in a separate computing environment.
- All 85 run-level columns matched the archived results in value. The file is not
  byte-identical: for fault families not applied in a run, this rerun's runs.csv
  records severity as 0.0, where the shipped testbed writes a blank cell (sev_*
  columns only). Treating a blank as "family not applied" (severity 0), every
  value matches; no other differences were found.

Analysis comparison
- e1_summary.csv, table_6a.csv, table_6b.csv, table_7.csv, table_e1.csv and
  table_validity.csv were byte-for-byte identical to the archived outputs.
- summary.json substantive numerical results matched. Environment metadata
  differs because the independent audit runtime used Python 3.13.5, NumPy 2.3.5,
  pandas 2.2.3, SciPy 1.17.0 and Matplotlib 3.10.8, rather than the archived
  environment Python 3.13.18, NumPy 2.5.3, pandas 3.0.5, SciPy 1.18.1 and
  Matplotlib 3.11.2.
- Figure files were regenerated; image bytes may differ due to library versions,
  rendering and metadata. The numerical source tables are the comparison basis.

Additional audit checks
- E2 and E2b runs are seed-paired: 1,000 matched seeds with the same fault-family
  selections and severity settings; the intended difference is H4 deletion
  selection priority (targeted versus independent).
- Clean-run structuring episodes: 1,080 across 360 clean runs; episode transfer
  count: 3,752 (568 episodes with three transfers; 512 with four transfers).
- The independent reproduction confirms the archived scientific outputs. It does
  not justify changing the model, data, reported estimates or conclusions.

Reproduction-runtime versions above describe the independent audit environment,
not the archived environment required to reproduce the primary output byte-for-byte.
