forecasting-ch23-epidemicslisted

# Chapter 23: The Epidemiologist's Dilemma ## Scope and intake Use when reporting delays distort recent counts or a compartment model is being used for conditional epidemic scenarios. Do not use for presenting the controlled SIR/SEIR lesson as a medically validated operational forecast. Ask only for unresolved material inputs: Do counts represent event dates, report dates or prevalence? What was known as of the cutoff? Has reporting delay changed? What population and transmission assumptions are justified? ## Input contract and additional evidence A reporting triangle in long form: one row per (event date, report date) with the count reported on that date for that event date. Set `as_of` to the cutoff; later rows are discarded. The delay law is estimated from cohorts at least `mature_age` periods old (at least ten of them) unless `delay_prob` is supplied. Give `population` to fit the SEIR model; without it the mechanistic forecast is skipped and listed under `not_done`. Minimal **format illustration**, not sufficient training data: ```csv event_date,report_date,count 2020-01-01,2020-01-01,16 2020-01-01,2020-01-02,34 2020-01-02,2020-01-02,25 ``` ## Executable interface Exact CLI columns: `event_date,report_date,count`. Supported method controls: `as_of, delay_prob, frequency, horizon, mature_age, max_delay, origins, population, recovery_rate, season, seed, sigma`. The tool builds the triangle at `as_of`, estimates the reporting-delay distribution from mature cohorts