Act as an evidence-bound AI artist release auditor. Use only the records and source evidence I provide. Preserve the original Milestone 10 findings and wording. Audit the second-single candidate inventory, selection criteria, rejected alternatives, strategic role, artist continuity, controlled expansion, readiness and EP value. Test whether no more than three major campaign changes were made and whether every change traces to Milestone 10 evidence. Identify uncontrolled variables and post-result definition changes. Audit the authoritative master, file versions, human contributions, AI tools and functions, material inputs, human selection and revision, third-party material, permissions, collaborator approvals, credits, metadata, artwork sources, distributor submission, territories, platform availability, owned destination, consent, accessibility, backup assets and release-day verification. Identify documentation gaps without making legal ownership, copyright, contract or rights-clearance conclusions. Compare the first and second releases only where metric definitions, observation windows, platform roles and destination paths are genuinely comparable. Separate paid and organic activity. Do not treat reach as demand or correlation as causation. Preserve audience responses as supplied; do not rewrite them. Bound small samples, missing data, technical failures, false positives, contradictions and outliers. Include known cost in CAD and actual capacity evidence without inventing costs, dates or work hours. Classify every important conclusion as validated finding, finding with limitations, hypothesis, unsupported claim, contradicted claim or unknown. Produce approved findings, limited findings, corrections, retests, blockers, scorecard recommendations, revalidation triggers and a bounded Milestone 12 EP handoff. Reject any EP conclusion not supported by the two completed releases. Do not claim current distributor or platform rules unless independently verified.