Workflow Status

The fastest way to check progress is to ask. Put the question to whoever owns the work and an answer comes straight back. At three people that is the cheapest arrangement there is, and setting up anything else would be waste.

At Pagedeck it fell apart around ten. Mia collected progress from ten people each morning, which cost an hour by itself before any of it was written down. The larger problem was that most answers were some version of "almost done."

That phrase meant a different state for each person. Someone who had finished the code but not opened a review, someone whose review had passed and was waiting on a deploy, and someone whose work had come back from QA and was being reworked all used the same sentence. Mia wrote the answers down and still could not count how many items would ship that week, which is the same as not having asked.

Making the same words mean the same thing requires not letting people choose the words. Defining the columns up front and having work point at one of them is what this value is.

The team fixes the list of columns first, then every item carries one. Five things get decided.

1. What divides the columns. The basis is where the work changes hands, not what kind of work it is. To do, in progress, in review, done is the most common starting set, and teams with a separate QA step add one more, rarely to any great benefit. Column names describe a state rather than an activity: in progress is a state, developing is an activity.
2. Whether each column's meaning is written down. The last one especially. Done meaning deployed and done meaning merged differ by team, and both are defensible, so that column gets tied to the definition of done, the shared standard for calling work finished, in the same sentence.
3. How many columns to allow. More columns show where things get stuck more clearly. But each added column adds a move somebody has to make, and once moving becomes a chore people put it off and the board drifts from reality. Staying under about six is the practical limit for a board people actually maintain.
4. Whether waiting gets its own column. In review and awaiting review are different states. The first means somebody is looking; the second means nobody is. Merged into one, work nobody has opened for three days still reads as in progress. Since waiting is where the time actually goes, separating it is often enough on its own to make the bottleneck visible without any further measurement.
5. Who moves it. The person who owns the work. When a manager moves it on their behalf, the value stops being a record of fact and becomes a trace of management, and the owner stops treating the board as their own tool, which is the beginning of a board nobody trusts.

Automation reduces the manual burden. Opening a pull request moves the card to in review; merging moves it to done. The fewer moves a person has to remember, the closer this value stays to reality, and accuracy here is entirely a function of effort.

The first place is the kanban board, which arranges work into columns by state to show the flow. This value is the column, so defining the list and designing the board are effectively the same task; change one and the other changes.

The second is the daily scrum, the short daily sync on progress. Instead of each person narrating what they did, the team looks at the board together and discusses only the cards that have not moved. A card sitting in the same column as yesterday is the definition of a blocked card, and that is a shorter conversation than ten narrations.

The third is measuring flow. Cycle time, the time from starting work to finishing it, gets calculated from the record of when this value moved to which column. Seeing which column cards pile up in exposes where capacity is short, and it is also where you decide to apply a work-in-progress limit, a cap on how many items may run at once.

The fourth is reading it alongside other attributes. Status alone does not say who is blocked, so it gets read with the assignee who took the work on. Six cards in awaiting review all pointing at one person is not a staffing problem; it is a review-distribution problem, and the board said so without anyone raising it.

The fifth is reporting. Progress on an epic, a large body of work split into several items, falls out of the statuses of the cards beneath it. Which means that when the status disagrees with reality, every number built on top of it disagrees too.

"That card is still in progress. Can you move the status?"

The board and reality have come apart.

"Let's add an awaiting-review column. Right now the waiting time is invisible."

Splitting a column to expose a bottleneck.

"Let's settle whether done means deployed or merged."

Aligning the meaning of the last column.

At Pagedeck, Mia keeps five columns: to do, in progress, awaiting review, in review, done. A QA column existed for a while, but people kept forgetting to move cards through it, so QA went into the definition of done and the column came out. Fewer columns, same standard.

Theo set up automatic moves. Cards travel on their own when a pull request opens and merges, so the only manual move left is the first one at the start. Complaints about the board not matching reality dropped noticeably after that.

Believing more columns mean better management is common. Finer stages look like a more accurate picture. But a column's accuracy holds only as far as people keep moving cards. At nine columns, some work needs moving twice a day, and then people batch the moves every few days or forget entirely. A detailed board that is wrong is worse than a coarse board that is right, because the first one manufactures numbers that look authoritative.

The second is treating it as a value for managers. It gets used for progress reporting, so it reads that way. But the people who get the most out of it are the ones moving the cards: how many items are in their hands right now, and where what they handed off has stalled, become visible without asking, which is the return on the effort of moving a card. A board used only for reporting makes moving cards somebody else's job, it goes stale quickly, and a stale board muddies judgment more than no board does.