# Chaos Lab results bundle v0.4

This directory contains aggregate result files for the Future Shock paper "Chaos Lab: When the Station Survived and the Benchmark Failed."

Included:

- `public_run_manifest.tsv` / `public_run_manifest.json` — run manifest with run IDs, models, seeds, score fields, and outcome classes.
- `clean_live_v2_per_run.csv` / `.json` — per-run metrics for the 30 clean-live v2 runs.
- `clean_live_v2_model_summary.csv` / `.json` — model-condition aggregate summaries.
- `v2_sensitivity_summary.csv` / `.json` — preserved sensitivity checks over the v2 scoring formula.
- `fallback_controls_summary.csv` / `.json` — opportunistic no-model/fallback-only observations, separated from clean-live model runs.
- `quote_provenance_public.csv` — verified quote provenance with `source_type` values that distinguish `model_output`, `scripted_event`, and `scripted_broadcast`.
- `scoring_formula_v2.md` — public excerpt of the scoring formula and threshold definitions.
- `CHECKSUMS.sha256` — hashes for the files in this bundle.

Not included in this release:

- raw model call logs
- event traces
- full prompts or system/developer instructions
- preflight/provider routing logs
- stdout/stderr logs
- source manifests used during internal analysis

Reproducibility posture:

Exact live-run reproduction is not guaranteed because model snapshots, routing, and nondeterministic generation can differ. These artifacts support audit and recomputation of preserved reported metrics. They do not certify real-world crisis-governance competence and should not be used for model procurement, emergency management, or vendor ranking.