Story point calculator

Almost every estimation tool shows an average, and it is the one number on the screen that should not be used to settle a round. The reason is simple: an average manufactures a value nobody voted for, and usually one that is not even on the deck. Votes of 3, 3, 3 and 13 average to 5.5. Not one person in that room believes the story is a 5.5. Four people hold two clearly different beliefs, and the average describes neither of them.

The damage is not the false precision, it is what the number lets you skip. A round of planning poker is not a measurement exercise. It is a mechanism for finding the person who knows something the rest of the room does not, and the way that person announces themselves is by playing a card three steps away from everybody else. Averaging closes that conversation before it opens. You get a number, you write it down, and you find out what the 13 knew about the legacy authentication path in the second week of the sprint.

There is a second, quieter failure. Averaging rewards strategic voting. Once a team learns that the number is the mean, someone who thinks the story is an 8 and expects the room to say 3 will play a 13 to drag the result up. That is a rational response to the rule and it makes every vote less honest.

So what is the average for? Two things. It is a sanity check against the card the room picked, and it is roughly what you would sum for a sprint if you were forced to. Look at it, do not resolve with it.

Agreement. The share of voters who chose the most common card. This is the number to act on. At 100 per cent there is nothing to discuss. At 80 per cent or above, with the stragglers one card away, take the majority card — that is the same line the live room draws when it calls a round strong agreement. Below that, the round has told you the story is not understood the same way by everyone in it, which is a more useful finding than any estimate would have been.

Most common card, the mode. The candidate estimate, and the only statistic here that is guaranteed to be a card someone actually played. When a round resolves, it resolves to this.

Median. The middle vote once they are sorted. It is far more robust than the mean because a single extreme card cannot drag it, and on an odd number of voters it is always a real card. If you are going to reach for a summary number, reach for this one rather than the average.

Average. The arithmetic mean of the numbered cards. Question mark and coffee cards are excluded, which is what SprintsPoker does in the live room too. Useful as a check, dangerous as a decision.

Spread. The distance between the lowest and highest card, counted in steps on the deck rather than in points. This is why the deck matters: on Fibonacci, 3 against 5 is one step and normally a rounding argument, while 3 against 13 is three steps and normally two people describing different work.

Both count towards agreement and neither counts towards the average. That asymmetry is deliberate. Four people all playing the question mark is a completely coherent, unanimous statement: the story is not ready. Treating that as zero agreement would be wrong. Averaging it into a number would be worse.

In practice, a single question mark should stop the round. It is almost always the cheapest signal in the session, because the person holding it is rarely the only one who is unsure, just the only one willing to say so. A coffee card should be taken at face value: estimate quality drops off sharply after about forty minutes, and a re-vote from a tired room is not worth the time it takes.

Velocity is the mean of points completed over the last few sprints. It is a planning aid with two hard rules attached: it belongs to one team, and it belongs to one point scale. A team of six does not have twice the velocity of a team of three in any meaningful sense, and comparing the two is how story points turn into a performance metric and then stop being useful estimates, because the team quietly inflates them.

Use four or five sprints. Fewer and one bad fortnight dominates. More and you are averaging over a team that has since changed shape.

The forecast range above is the lowest and highest sprint the team has actually delivered, not a confidence interval. That is on purpose. A range built from real sprints can be defended in a room, because every number in it already happened. When somebody asks when forty points of work will be done, the answer with the best chance of being true is a range, delivered with the reason it is a range.

The commitment figure, points completed as a share of points committed, is a different signal. If it sits well below 100 per cent for several sprints, the problem is usually over-commitment at planning rather than bad estimates. If it sits at exactly 100 per cent every sprint, the numbers are probably being adjusted after the fact.

This page is for working through a round after the fact, or for settling an argument about how the maths should work. During an actual session you want the votes hidden until everyone has committed, which is the part that makes the numbers worth anything.

SprintsPoker computes all of this live: open a room, share the link, and the distribution, agreement, median and average appear at the moment the cards flip. If you want the reasoning behind the numbers themselves, the estimation guide covers what a point should represent in the first place.