SignalFlow supports two fundamentally different ways to aggregate across time: moving (sliding) windows and calendar windows.
A moving window like .sum(over='1d') slides continuously—at 14:30, it covers yesterday 14:30 to today 14:30. By contrast, a calendar window resets cleanly on fixed boundary ticks (hour, day, week, month) aligned to a chosen timezone (such as America/New_York or UTC) configured in Chart/Detector options or the SignalFlow Execute API (timezone is not passed to .sum()). Calendar windows answer business questions like "total orders since midnight today" or "monthly SLA error budget consumed so far".
For smoothing and trend velocity, SignalFlow also offers Exponentially Weighted Moving Averages (.ewma(over='5m') or .ewma(rho=0.2))—which weights recent datapoints more heavily than older ones—along with .delta() and .rateofchange() to measure how rapidly a gauge or counter is climbing or falling.
# Moving 24h window (rolling)
rolling_24h = data('orders.total', rollup='sum').sum(over='1d')
# Calendar day window (resets at midnight; cycle_start takes an hourly offset like '00h' or '08h')
# Note: timezone (e.g., 'UTC') is a Chart/Detector/API execution setting, not a .sum() parameter
cal_today = data('orders.total', rollup='sum').sum(cycle='day', cycle_start='00h')