Shipping cadence beats velocity

Velocity is a measure of output over a fixed window. Cadence is a measure of rhythm. They sound like the same idea until you look at two teams with identical throughput and completely different reputations.

Predictability is a feature

A team that ships something every three days is easier to plan around than a team that ships nine times in a fortnight and then nothing for two weeks, even when the totals match. Everyone downstream, from support to sales to the person writing release notes, builds their own routine on top of yours.

Cadence degrades earlier

In our data, the gaps between merges widen well before total output falls. By the time velocity drops, whatever caused it has usually been in place for a month. Watching the rhythm gives you a chance to ask what changed while the answer is still fresh.

How to improve it without gaming it

Shrink the unit of work until it fits comfortably inside your rhythm, rather than stretching the rhythm to fit the work. The point is not to commit more often; it is to keep the distance between ready and merged short enough that nothing has time to go stale.

What cadence looks like in the data

We measure the gap between merged changes and look at the spread, not the average. Two teams can both merge ten times a month: one every three days, the other eight times in the last week. The averages match. Only one of them can be planned around.

The spread is also the part that responds to changes in how a team works. Averages move slowly and hide reversals. Gaps widen immediately.

Velocity tells you how fast a team moved last week. Cadence tells you whether anyone can rely on it next week.

Why cadence breaks first

In our data the gaps start widening two to three weeks before total output falls. The work is still happening; it is just arriving in bigger, later pieces because something is blocking the small ones — usually review, occasionally a flaky pipeline, sometimes a decision nobody has made.

By the time velocity drops, that cause has been in place for the better part of a month. Fixing it then costs more than it would have, and the team has usually already started explaining the slowdown in terms of scope rather than friction.

How to improve it without gaming it

Shrink the unit of work until it fits comfortably inside the rhythm you want, rather than stretching the rhythm to fit the work you have. A change that takes four days to review is not a review problem. It is a size problem wearing a review problem’s coat.

Cap work in progress before you cap anything else. Cadence is mostly a queueing property, and the fastest way to widen the gaps between merges is to have four things half-finished instead of one thing finished.

Do not chase the number by merging trivia. We count meaningful changes, not commits, and a week of formatting will not move your score. If it does, tell us — that is a scoring bug and we would like to fix it.

When irregular is fine

Research weeks, migrations and holidays all produce ragged cadence for good reasons, and a score that punishes them is a score that is wrong on purpose. Look at the trend across a quarter rather than the number on a Tuesday.

The signal worth acting on is a widening gap that nobody can explain. That one is almost never about capacity.