Content Hidden Behind Execution: Audit Labels, Codebook, and Analysis Scripts

Anonymized data release for the paper "Content Hidden Behind Execution: Human-Supervised AI Auditing of Runtime-Revealed Sensitive Content on a Youth Programming Platform" (under double-anonymous review). It contains project-level labels for a 500-project runtime-safety audit of public Scratch projects (content categories C1–C9, risk 0–4, evidence channels E0–E6, reveal mechanisms M0–M8, exploration depth D0–D4, confidence), the annotation codebook and sampling protocol, and a Python script (standard library only) that reproduces and verifies every aggregate number reported in the paper. No usernames, project IDs, URLs, titles, media, or free text are included; the corpus deliberately oversamples sensitive content and does not support prevalence estimates.