Upload a trial proposal and every patient gets simulated through screening, randomization, adherence, dropout, interim looks, DSMB review, supply shocks and budget burn. The output is the emergent Expected Trial Failure Probability (ETFP): how likely the trial is to fail to deliver a correct, conclusive answer, whatever the true effect turns out to be.

Model notes (gen 11): every design shares one operations model — regulatory delay, site activation, screening, consent, supply interruptions and budget burn, with money running out stopping the trial (which then analyses what it has or is abandoned). Design modules: parallel superiority, non-inferiority and equivalence (frequentist group-sequential or Bayesian posterior thresholds); cluster-randomized (cluster-level t-test); stepped-wedge (cluster and period fixed effects on cluster-period means); crossover and N-of-1 (within-patient differences, carryover, period effect); factorial (marginal main effects, optional interaction); multi-arm with arm dropping and patient reallocation, and seamless II/III selection; non-randomized comparisons with drawn confounding bias; single-arm exact or Simon tests against a drifting benchmark; basket trials; dose-finding by 3+3, BOIN or CRM with an optional expansion cohort; and diagnostic accuracy with partial-verification bias. Safety stops are judged against drawn true SAE rates. Dropout is MCAR unless MNAR is chosen. Sensitivity and hedge comparisons are paired on the same seeds and carry 95% intervals. Budget is in your own units.