Meaningful events in time series rarely respect fixed time buckets. A price swing can start between bars, end early, and last seconds or minutes, yet it still needs a clear, explainable window for investigation. The challenge is scale: with 1 second sampling over a regular 6.5 hour trading session (23,400 timestamps), there are about 274 million possible start–end windows for a single instrument. Across millions of instruments, the search space grows to well over a quadrillion windows, before accounting for noise, mixed timescales, and the need to connect a move in the primary signal (price) to unusually concentrated aligned activity (participation/volume). This talk breaks down why “dynamic event windows” are hard in real systems and how to frame scalable, interpretable detection. The ideas generalize to other domains like system monitoring and healthcare signals.