Each of 12 simulated regions reports a daily case count xt. Instead of alarming on raw noisy counts, the detector tracks an exponentially weighted moving average (EWMA) of each region's counts — the same statistic behind CDC-style early-aberration reporting systems (EARS):
z_t = λ·x_t + (1−λ)·z_(t−1), 0 < λ ≤ 1
UCL = μ₀ + k·σ₀·sqrt( λ / (2−λ) )
μ₀ and σ₀ are the mean and standard deviation estimated during a 14-day noise-only baseline window, before any outbreak is allowed to start. Because zt averages recent history, it reacts to a sustained rise in cases while shrugging off single noisy days — smaller λ smooths harder but reacts slower; larger k reduces false alarms at the cost of a longer detection lag.
- λ — how much weight the newest day's count gets in the running average.
- k·σ — how many baseline standard deviations above the mean the smoothed statistic must climb before an alarm fires.
- Outbreak growth rate — after the baseline window, the outbreak region's true mean grows as μ₀·(1+r)days since onset; 0 keeps every region at baseline noise.
- Each tile's bar height is that day's case count and its color is the EWMA z-score band — it turns red the instant that region's zt crosses its own control limit.
- Drag the region grid to pan, scroll (or pinch) to zoom — useful once the outbreak tile turns red and you want a closer look.
Real-world relevance: this is essentially the algorithm behind syndromic-surveillance early-warning systems used by health departments to catch outbreaks in emergency-room visit or pharmacy-sales data days before a formal case report confirms them.