Use this format when
- The team has stable practices worth tuning.
- Start Stop Continue feels too blunt.
- People disagree about whether to remove a practice or reduce it.
The Starfish retrospective separates complete changes from adjustments in intensity. Teams can protect what works, ask for more or less of a practice, and still make clear start and stop decisions.
Use the prompts to keep cards specific. The examples show the level of detail that makes discussion and voting useful.
What works at roughly the current level?
What still has value but happens too often or too heavily?
What useful practice needs greater frequency or attention?
What should be removed completely?
What new experiment should the team introduce?
A complete path from framing the room to assigning one concrete follow-up.
Name the sprint or project being reviewed, restate the purpose of each column, and remind everyone to describe observable events.
Give everyone uninterrupted time to add one specific observation per card across every column.
Read the cards column by column and let authors add context without debating solutions yet.
Combine cards that describe the same pattern, then vote for the themes that would most improve the next iteration.
Explore causes, consequences, and tradeoffs behind the highest-voted themes while making space for dissenting evidence.
Turn the strongest insight into a small action with an owner, deadline, and signal the team can review next time.
Collect cards asynchronously, then spend the meeting only on voting and discussion.
Apply all five prompts to planning, development, review, and release flow.
Limit observations to communication, pairing, handoffs, and team support.
Before the retro
They are Keep Doing, Less Of, More Of, Stop Doing, and Start Doing.
Starfish adds Keep Doing, Less Of, and More Of, making it easier to adjust the intensity of useful practices.
Use it with an established team that has enough shared experience to make nuanced process adjustments.
Group related cards, vote across the whole board, and select only one or two experiments with owners and review dates.
Open the ready-made columns, invite the team, collect observations, vote, and leave with an owned improvement.