Planning Poker

A team needs a sense of how big a piece of work is. Everybody privately picks a card, everybody turns them over at the same moment, and the team talks about the gap between the highest and the lowest before agreeing a number.

The cards run in a scale with widening gaps — 0, 1, 2, 3, 5, 8, 13, 20, 40, 100, roughly Fibonacci, is the one most decks hold. That is deliberate. Nobody can tell 21 from 22, and a deck that lets you try wastes the session on precision that does not exist. The facilitator picks which of four scales the room uses.

The simultaneous reveal is not a flourish, it is the mechanism. Ask a team out loud and the first number spoken becomes the number. The most senior voice, or simply the loudest, sets an anchor, and everybody else adjusts towards it without ever deciding to. That is not a failure of nerve — it is how estimation works when people can hear each other. Hiding the cards until they are all in is what stops it, and it is the only part of the ceremony that cannot be dropped.

Which is why this room does not merely hide the other cards until the reveal. Your card is not sent to anybody else's browser at all. Before the turn, all the room is told about you is that you have voted.

The disagreement is the product. A round where everybody says 5 tells you almost nothing you did not already know. A round where one person says 3 and another says 13 means the two of them are looking at different pieces of work, and the ten minutes that follow are the most valuable ten minutes of the session. The number is a by-product.

A round is one story: everybody picks, the cards turn, the room talks.

Everybody joins before anybody votes. The room opens in a lobby while people arrive, which is where the facilitator picks the deck and pastes the backlog; the first story opens when they say so.

The facilitator reads a story out. Everybody else picks a card — the facilitator holds none, on purpose, because the person running the room putting a number down is the loudest voice in it anchoring everyone else. While that is happening the room shows who has voted and nothing else — an unrevealed vote is never sent to anybody else's browser, the facilitator's included. The cards turn over by themselves once the last person is in, or the facilitator can turn them early if somebody has gone for coffee.

On the facilitator's screen — usually the one being shared — that is drawn as a table: a place for everybody in the room, and a card face down at their place as they vote. At the reveal every card turns over where it lies, lowest first, so the room watches the spread open rather than reading a wall of numbers. The two people at the ends are the two the conversation starts with.

Then the room talks about the highest and the lowest, and records the number the team agreed — which need not be either of them, and need not be the average of anything. That conversation is the output. The number is what you write down so the conversation can end.

Why there are four scales

Different teams size on different scales, and the argument about which scale is right is not one a tool should win — so the facilitator picks a deck in the lobby and the tool has no opinion about which. All four are written out on the Play tab, card for card.

Fibonacci

The plain sequence. Gaps that widen on their own, and nothing on the deck that is not a size.

Modified Fibonacci

What almost every physical deck holds, and the one a room opens with. The half is for work too small to argue about; the 100 is for "that is not a story, it is a project". It rounds the large end off because the difference between 34 and 55 is not something anybody can see.

T-shirt

Sizes rather than numbers, which suits a team being asked to estimate for the first time — and teams who found that their numbers were getting added up. The room reports the most common size and the range, and no average at all, because there is no such thing as the mean of a medium.

Powers of two

Every card twice the one before it. A short deck, which makes the choice quick and the disagreements large enough to be worth talking about.

The facilitator can change the deck between stories, and anything the team has already agreed keeps the number it was agreed on. Changing it mid-round is refused — a card of M cannot be read on a Fibonacci deck.

Four people out of six said 5, so the temptation is to write down 5 and move on. That would throw away everything the round just bought.

The two numbers worth asking about are Ada's 3 and Eve's 13. Ada is thinking of the form and the token — a day's work, and she is right about the part she is looking at. Eve has done it before somewhere else and is thinking about the expiring link, the email that has to go out, the rate limiting that stops it becoming a way to spam somebody, and the audit entry the security review will ask for. She is also right.

They are not disagreeing about size. They are describing two different pieces of work, and nobody in the room had noticed until the cards came over. The team agreed 8, and split the audit entry into a story of its own.

Note what the tool did not do: it did not write 6.5 down. An average of six opinions is a number nobody in the room actually holds, and accepting it quietly is the exact thing planning poker exists to prevent. The room suggests the most common card, the median and the nearest card on the deck. Somebody still has to say the number out loud.