Task times from measurements
You have timed a step on the floor and want the model to vary as the real times do, long tail included, without choosing a distribution.
Built in A field or setting does it directly.
How
On the machine's task (Hardware → the machine → Timing & flow rules, or Factory data → Tasks) set Time distribution to empirical and list the Measured times (s): 72, 78, 80, 81, 83, 84, 85, 86, 88, 71, 94, 118. In the profile: "dist": "empirical", "dist_samples": [72, 78, ...]. Each job draws one of them at random, each as often. Set the task's time to their mean: the draws are scaled so their mean is the task's time, so a time that comes from a formula (longer for a larger product) keeps the measured spread around it. Variation must be on: Project settings → Simulation model → Use authored cycle-time variation (settings.randomMode: true).
With a few measurements and no tail, a triangular time (shortest, most likely, longest) or a normal or lognormal one with a variation (CV) is the alternative; they are on the same menu.
Check
Valve assembly timed 12 times, 71 to 118 s, mean 85 s; valves arrive one every 95 s. Over 40 h the bench assembles all 1,514 valves, each in one of the measured times. The long assemblies make the next valves wait: the mean lead time is 2.63 min, against 2.58 min with a fixed 85 s.