Sammy Orangkhadivi

Postmortem Before the Fact

Backtesting AI-Assisted Failure Forecasting for Strategic Decision-Making

Language models make premortems cheap to repeat, but a plausible failure narrative is not a validated forecast. This paper proposes Failure-First Forecasting, a protocol that freezes a plan, generates causal failure pathways, remediates them, reforecasts the revised plan, and scores the whole process against what actually happened, with temporal isolation treated as a first-order requirement.