Start-Stop-Continue

Method

A retrospective format that sorts everything into three columns — start, stop, continue — so the next cycle has concrete changes.

Why it exists

Open a retrospective without deciding what to talk about and it drifts one of two ways: an hour of "that went well," or an hour of complaints that leaves nobody knowing what will be different next week.

Pagedeck's retrospectives were the second kind. Reviews are slow, requirements keep changing, there are too many meetings — an hour of that. Twelve lines got written down and nothing changed in the following cycle. Counting them, Mia found exactly one line that said who would do what. The cause was that there were no columns to put the sentences in.

How to run it

Draw the three columns first and write only inside them. The columns are the questions, so nobody has to work out what kind of thing to contribute.

  1. Draw three columns: start, stop, continue. The first two are for changing something; the third is for protecting what's working.
  2. Have everyone write silently. Five to seven minutes, at least one item per column per person. Starting with discussion tilts the session toward whoever spoke first.
  3. Post them up and group the duplicates. "Reviews are slow" and "review requests pile up" are one item.
  4. Vote and keep the top three. Trying to address twelve lines changes nothing, and three is roughly what a two-week cycle can absorb alongside its actual work.
  5. Give each surviving item an owner and a date inside the next cycle. Stop items need owners too — stopping the Wednesday meeting requires somebody to cancel the invitation.
  6. Spend the first five minutes of the next retrospective checking the previous items. Without those five minutes, the format decays back into a list of things that were pleasant to say.

The column that most often comes back empty is stop. Offering to do more is easy and declining to do something needs agreement. An empty stop column reads as a signal that the team is only ever adding work, which is also why each cycle feels slightly fuller than the last.

This format doesn't dig into causes. "Respond to reviews faster" comes out of it; why reviews are slow doesn't, so when the same item appears three times, that topic gets its own session. There's also no column for feelings or learning, which makes Liked-Learned-Lacked-Longed for (4L) a better fit after a hard cycle, and Keep-Problem-Try (KPT) a better fit when problems and experiments should be listed separately.

How it sounds at work

"The stop column is empty again. Something has to come out."

Only additions are being proposed.

"Take three items and roll the rest over."

Capping how much one session tries to change.

"Let's start by checking last time's items."

Confirming what was decided before adding anything new.

The first retrospective Mia ran in this format produced three items: respond to review requests within a day, stop the Wednesday meeting, keep the daily fifteen minutes as is. Checking two cycles later, the first two were holding — and the time freed by dropping the meeting turned out to be where the review responses came from.

Where to go next

Last updated 2026-09-16