Paste raw cycle times. Get percentiles, histogram, and a tail-shape diagnosis. Forecast at p85, not the median — the mean lies when work has a long tail.
Unit
SprintFlint records cycle time per ticket, surfaces p85 in dashboards, and flags when the tail is widening before forecasts go wrong.
Start FreeCycle time starts when work begins (in-progress). Lead time starts when the request is logged (in the backlog). Cycle time is the team's signal; lead time is the customer's signal. This tool works for either — paste whichever you measure.
Cycle times almost always have a long tail. The mean and median look optimistic; p85 reflects what 85% of work actually finishes within. Forecasting at p85 means you hit your dates; forecasting at p50 means half of forecasts are late.
30+ for stable percentiles. 10-29 is directionally useful but noisy. Under 10 and you're guessing — track for another sprint or two.
Under 2x is tight. 2-4x is normal for most teams. 4-8x means a few stories stall badly — investigate WIP limits and blockers. 8x+ means systemic problems with story sizing or unblocking work.
No. Outliers are the signal. The point of percentiles is to weight everyday flow without throwing away the cases that actually break forecasts. If one ticket sat for 90 days because of a stuck dependency, that's the data.
Most teams use calendar days for simplicity. Working days are more accurate but harder to compute consistently. Pick one, stay consistent, and document it. The percentile shape matters more than the absolute scale.
Sprint forecaster (Monte Carlo at p85), Velocity calculator, Sprint health check, Burndown chart generator.