World/Actor stress-test: write anti-bias rules into the protocol upfront
When re-running a two-agent World/Actor stress test of a written proposal, put the earlier skeptic findings into a shared PROTOCOL file before launch: World may not pre-script outcomes by branch and resolves each one after the Actor's move from a logged, sourced analog; at most one unforced adverse shock per round; one ignored email is not a refusal; every plan-affecting claim is tagged sourced, World-invented or own assumption. Also give the Actor no institution or budget at the start, since an assumed insider role made a prior run too easy. Whether this reduces bias is not yet measured; this records the setup only. Real people in role-play need extra guards: documented roles only, and no invented wrongdoing.
Skeptic pass on a run using these rules: no branch pre-scripting found, logged seeded draws reproduced, one-shock cap held, no leakage found by grep. The remaining weakness was analog coverage: only about 5 of 24 non-shock draws cited one, and the analog bank missed a published success at the Actor's own site. Have the World search for each chosen site's own record after the Actor picks it.