Blog ·
Why a single delivery date is a guess
A burndown chart ends in one date. Plans, roadmaps and promises to customers then get built on that date. The date feels exact because it is a single day. It is not exact. It is the middle of a range of outcomes, and the chart does not show the range.
What the single date leaves out
Take a team with 93 points of work left. Divide by the average sprint and you get a date. The average hides two things. Sprints differ: some deliver much more than others. And scope moves: new work arrives while the team works. Neither shows in the division.
When the date slips, nobody can say whether the plan was wrong or the date was only a guess. The chart gave no odds, so no outcome can contradict it.
A worked example
The numbers in this article come from the demo project in Sprint Radar. It is synthetic: made-up issues in a made-up project with 36 made-up sprints, generated on 11 Nov 2026. Treat them as an illustration, not as data about any real team.
In the last 6 sprints the demo team delivered between 18 and 31 points per sprint. The mean was 25.5. A single date from the mean alone ignores that spread of 13 points.
A simulation keeps the spread. It replays the team’s recent sprints at random, 10,000 times, and counts the sprints each replay needs to finish the remaining work. The finish dates of those replays form a distribution. Reading it at a few points gives this:
- 50% of the runs finished by 17 Mar 2027.
- 70% finished by 14 Apr 2027.
- 85% finished by 28 Apr 2027.
- 95% finished by 9 Jun 2027.
The 50% date is what a single-date plan usually reports. By the model, about half of the outcomes are later. If you promise that date, you are taking a coin flip. The 85% date costs a few weeks and is missed about 15 times in 100. You choose the odds you can defend to your customer or your boss.
Is the range any good?
A range with odds makes a claim that can be checked. If the 85% dates are honest, about 85 in 100 past forecasts should have held. Sprint Radar checks that for you. It goes back in time, forecasts from what was known then, and compares the result with what happened. For the demo project it reports: “Past forecasts 3 sprints ahead at 85% confidence held 18 of 19 times”.
That line is a measurement of the method on this data, not a guarantee for the next forecast. It also shows its limits. The backtest checks forecasts a few sprints ahead. Longer forecasts are not validated, and the product says so.
When to distrust a forecast
A forecast is only as good as the history behind it. Distrust it when:
- The history is short. With fewer than 8 sprints, Sprint Radar warns that the odds are less reliable than they look.
- Throughput is falling. The model replays the recent past, so a falling trend makes the dates optimistic. The product warns about this.
- The team changed. A new team, a new process or a new tool makes old sprints a poor guide.
- Many remaining issues have no estimate. They count as zero, so the remaining work looks smaller than it is.
What to do with this
Stop reporting one date. Report two: the date you would bet on at even odds, and the date you would defend at high odds. Say which is which. When the dates move between forecasts, report how much and why. A forecast that moves by three days is news about the work. A forecast that moves by three weeks is news about the plan.
None of this makes the future known. It makes the uncertainty visible, so the people who carry the risk can decide how much of it to take.
See a forecast with odds for your own project.