Every Gantt tool prints dates it can't defend. This one refuses to: milestones carry a P50–P90 range and render with uncertainty whiskers — and the snapshot compare turns "what moved since the last board pack" into one slide.
Start typing — or load an example.
A milestone date is a forecast, and forecasts have distributions — but the moment a date lands in a Gantt cell it reads as a commitment, and the argument shifts from "how uncertain are we" to "why did you miss it". Writing every date as a P50–P90 pair (the same 90%-interval habit as the estimator tools) keeps the uncertainty in the artefact itself: the diamond is the middle of your belief, the whisker is its honest width, and a date with no whisker is visibly making a claim nobody should sign.
The snapshot compare is the tool's real job. Snapshot before each board pack; next cycle, the comparison renders slips, pull-ins, widened ranges, and additions as one exportable slide — the "what moved" conversation with the drama already drawn. Those saved snapshots quietly become reference-class data: your promised-vs-actual history, waiting for the day you want an outside-view forecast (Flyvbjerg's cure for the planning fallacy).
Sources: Douglas Hubbard on calibrated 90% intervals (How to Measure Anything); Bent Flyvbjerg on reference-class forecasting; Reinertsen on why variability is information, not failure.