forecasting-ch26-decision-readinesslisted

# Chapter 26: How to Be Ready Without Being Certain ## Scope and intake Use when a probability forecast must guide a concrete action under explicit costs and ongoing evaluation. Do not use for requiring certainty before acting or choosing thresholds solely to maximize classification accuracy. Ask only for unresolved material inputs: What actions are available? What is the cost of a false alarm and a miss? Are costs constant and probabilities relevant/calibrated? What capacity or feasibility constraints apply? ## Input contract and additional evidence CSV event_id,probability,outcome with p in [0,1], outcome 0/1 for CLI scoring; retain unresolved cases in the journal, not this scoring file. Config false_alarm_cost and miss_cost are strictly positive and finite for the supplied adapter. Record forecast timestamps, event definition and action constraints separately; predictions must predate outcomes. Minimal **format illustration**, not sufficient training data: ```csv event_id,probability,outcome E1,0.3,1 E2,0.1,0 ``` ## Executable interface Exact CLI columns: `event_id,probability,outcome`. Supported method controls: `as_of, false_alarm_cost, frequency, horizon, miss_cost, season, seed`. The adapter requires resolved events and strictly positive finite costs. It acts at p>=threshold. The reported best constant policy is a hindsight diagnostic, not a pre-recorded baseline. It reports Brier alongside decision loss; chapter 9 adds a fuller calibration diagnosis. ## App