Attack Campaign & Corpus Freeze

Objective

Execute the preregistered attack campaign and freeze the benign/adversarial corpus with trustworthy provenance.

Why this matters

Model experiments require a stable corpus whose labels and collection conditions can be defended independently of model performance.

Design considerations

• Repeatability, provenance, scenario coverage, contamination controls, and immutable dataset manifests.

Key decisions

• Execute only after entity, telemetry, baseline, and harness gates are satisfied.