Forecasting delivery honestly
Velocity is useful for one team's own planning and actively harmful as a target. Throughput forecasting sidesteps the problem entirely.
Velocity is not productivity
Velocity is a property of one team's estimating scale. Two teams delivering identical work can differ threefold purely in how they size things, which makes cross-team comparison meaningless.
It is worse than meaningless as a target. The moment velocity becomes a number to increase, it increases — through estimation rather than delivery. The metric stops measuring anything and the team loses its forecasting tool in the same move.
Plan with the range
A commitment built on average velocity is missed about half the time. That is not a failure of the team; it is the definition of an average.
If recent sprints ran between 18 and 26, a six-sprint plan lands somewhere between roughly 110 and 155. Quoting the midpoint as a date throws away everything the spread told you, and the spread is the part that was informative.
Forecasting without estimates
A throughput forecast needs no estimates at all. Count how many items you finished per week, resample that history at random, and you have a distribution of outcomes.
It does not matter that items differ in size, because the history already reflects however big they were — a week of large items shows up as a low count, which is exactly the information you want. Counting what you finished is usually more predictive than estimating what you are about to start.
Resample rather than fit a curve
Your own history contains the holiday week, the incident, the sprint that went sideways. Fitting a smooth distribution to that data removes precisely those events — and they are the reason forecasts slip. Drawing from the real history keeps them in.
The two questions read from opposite ends
This trips people up and it is expensive when it does.
For how long will N items take, the cautious answer is the HIGH percentile: 85% of runs finished within that many weeks.
For how many items by a date, the cautious answer is the LOW one: you delivered at least that many in 85% of runs. Quoting the high number here is optimistic, not safe, and it looks identical to a careful forecast until it is missed.
When to throw the history away
All of this assumes the future resembles the recent past. A reorganisation, a shift to unfamiliar work, or half the team leaving breaks that, and nothing in the arithmetic will warn you.
Three months of current history beats two years spanning three different teams. When the team changes materially, start the series again rather than averaging across the discontinuity.