[assassin808]

Study now opens with a source-linked research overview. Preparing one experiment passes that scope to the existing agent composer; running it still uses the app's explicit simulation approval.

• Pair scholarly Human Program v2 with the existing protocol, without duplicating executable steps. Check coverage, references, the paired version and current preview before marking new paired drafts complete.
• Add Mazzarella's paper-level questions, predictions, findings and limits while retaining its original executable protocol. The private Knobe pilot remains blocked on the missing administered questions and is not included in this PR.
• Preserve accepted edits during checkpoint writes and retain legacy protocol-only projects.

Verified: typecheck, desktop/web builds, 103 targeted tests, and browser walkthroughs. A localhost responder completed the sample's Study 1 alone and both studies separately; scholarly hypotheses were not sent to participants. Hosted-agent material clarification and private refinery round-trip remain separate acceptance steps.