What changes when the output policy changes?
Four explicitly defined policies are applied to all1,530 unchanged first responses. This page compares accounting choices, not medical correctness. The original registered contract and sealed forecasts remain unchanged.
Same selected cell, all policies
“Outside original contract” describes a policy disagreement. It does not mean that an accepted answer is medically false. Fallback is q=0.5 under every policy; valid-only loss has its own denominator.
Exploratory consultation-level sensitivity
68 held consultations. Losses are averaged over18 fixed settings inside each consultation before2,000 paired consultation bootstraps. Intervals are pointwise, post hoc and conditional on this already-examined mock benchmark. No new p-values or best-policy selection. A lower Brier does not establish clinical safety.
Download all144 group summaries · Download every policy forecast · Fixed analysis recipe