GOI CRITERION-VALIDITY SIMULATION STUDY - SUPPLEMENTARY PACKAGE, RELEASE v1.1

Purpose
Simulation testbed, run-level data, analysis code, derived statistics, tables and
figures supporting the revised manuscript "Does an Integrity-Gated Governance
Observability Index Measure Reconstructability? A Simulation-Based Criterion-Validity
Study of Agent Telemetry under Out-of-Specification Faults" (UAEU Journal of
Engineering, revision R1).

Release and archive
Release v1.1 is published as a new version of Zenodo record 23936145. Its
version-specific DOI is shown on the Zenodo record page. The earlier release v1.0.1,
reviewed with the first submission, remains unchanged as version v1 of the same
record (version-specific DOI https://doi.org/10.5281/zenodo.23936145). Each DOI
identifies one data release; neither identifies the journal manuscript.

Reproduction
Reproduce into a NEW folder, so that the archived outputs in results/ and the
checksums below remain intact. From this directory, using Python 3.13.18:

  python3.13 -m pip install -r requirements.txt
  python3 goi_testbed.py repro
  python3 analyse.py repro

Then compare repro/ with results/ (see REPRODUCIBILITY.md). The testbed uses only
the Python standard library and is seeded (master seed 20261029); the analysis
uses seeded bootstrap procedures. Archived analysis environment:
  Python 3.13.18, NumPy 2.5.3, pandas 3.0.5, SciPy 1.18.1, Matplotlib 3.11.2
(pinned in requirements.txt; also recorded in results/summary.json under v11.env).

Experiments and run-level data
results/runs.csv contains 4,520 rows:
  E1  dose-response experiment: 2,520 runs
  E2  mixed-fault experiment with verifier ablations: 1,000 runs
  E2b the E2 mixtures, seed-paired, with key-holder deletion selected
      independently of forgery targets: 1,000 runs
Blank severity cells (sev_*) mean that fault family was not applied in that run.

Changes from release v1.0.1
- H3 key-holder forgery rewrites all four spans of a targeted action. Six mixed
  runs change (|delta GOI| <= 0.0008); every statistic reported in the first
  submission is unchanged at three decimals except rho(1-U) for the
  no-ledger-sample ablation (0.000 -> 0.001).
- Ablations report fixed-criterion comparisons and each verifier's own
  reconstruction, including counts of accepted, correct, false and missing records.
- Ground-truth columns distinguish falsified records across all executed actions
  from falsified records remaining available after deletion.
- Experiment E2b (independent deletion).
- Revision analyses: results/summary.json (key v11) and results/table_*.csv.
All v1.0.1 columns are retained.

Integrity
SHA256SUMS.txt lists a SHA-256 checksum for every file except itself. Verify from
this directory with:  sha256sum -c SHA256SUMS.txt
results/manifest.json is written by goi_testbed.py (configuration and testbed
hash); the fields analysis_script_sha256 and analysis_environment were added at
release time and are not regenerated by a rerun. Signing keys in the testbed are
simulation-only test literals, not production credentials.

Verification
verification/independent_reproduction/ holds outputs from a full rerun in a
separate computing environment, with a comparison log. It is verification evidence,
not a new experiment, and does not change any reported result.
