Safety Check for Retrospective

What is a Safety Check?

A Safety Check is a short anonymous vote taken before a retrospective, asking one question: how safe do you feel saying what you actually think in this room? Everyone answers on a scale from 5 down to 1, the board shows the distribution without names, and the facilitator decides how to run the session based on the result.

It exists because retrospectives silently fail in a predictable way. The meeting happens, the board fills with process complaints, the action items are technical and safe, and the real problem — the one everyone discusses privately afterwards — never reaches the board. A Safety Check does not fix that, but it tells you it is happening, which is more than a normal retrospective will.

When to use a Safety Check

• Before a retrospective about a failure — an incident, a missed deadline, a cancelled project — where people may fear blame.
• When managers or stakeholders are attending a session that is normally team-only.
• After a reorg, a layoff or a change of manager, when the cost of speaking up has just moved and nobody has said so.
• With a new or newly merged team, where the group has no shared history of what is safe to say.
• When retrospectives keep producing only safe, technical items and the problems everyone talks about privately never appear on the board.
• As a periodic trend line — run every few months, the shape of the distribution shows whether the team's trust is improving or eroding.

It is not needed for every retrospective in a settled, open team, and running it constantly turns a measurement into a ritual. It is also the wrong instrument if you are not prepared to act on a bad result: asking people whether they feel safe and then ignoring the answer costs more trust than never asking.

The columns explained

The board has five columns, from 5 down to 1. Each person adds exactly one anonymous card to the column matching their answer. Read the levels out loud before voting so everyone is scoring the same scale.

5 — No problem

I will say anything, including things that reflect badly on me or on someone senior in the room.

4 — Mostly comfortable

I will raise almost anything, though I would soften how I phrase one or two topics.

3 — Selective

I will talk about process and work, but I will keep quiet about people, decisions from above, and my own mistakes.

2 — Guarded

I will agree with points other people raise, but I will not raise anything myself.

1 — Silent

I will say nothing meaningful in this session.

Example board

A ten-person team ran a Safety Check before a retrospective about an outage that had reached customers. The engineering manager was attending, which was unusual.

Seven of ten people were at 3 or below. The facilitator read the totals aloud, said the number was too low to run a blameless incident review as planned, and offered the group a choice: continue with the manager present, or collect the incident timeline anonymously and discuss it the next day without them. The team chose the second. The manager left, which was the single most useful thing that happened in the meeting.

Two sprints later the same check produced four 5s and no 1s. The difference was not a workshop on trust; it was that the incident review had genuinely produced no consequences for anyone, and people had watched that happen.

Timing for a team of 5–10

As a rule of thumb: mostly 4s and 5s means run the session as planned. A cluster at 3 means keep the discussion on process and avoid personal topics. Anything below 3 for a third of the group means change the conditions before continuing.

Facilitator prompts

• "One card, one column, no names. I will not ask who voted for what — now or later."
• "Here are the totals. I am not going to interpret them for you. What should we do differently in the next hour?"
• "This is lower than last time. I do not need to know why from anyone specific — what would raise it by one point?"
• "Would people prefer I collect the input anonymously today instead of discussing live?"
• "Is there anyone in this meeting whose presence is changing what we can talk about? You can say yes without saying who."
• "We scored a 2 average last quarter and a 4 now. What changed that we should keep doing?"

Common mistakes

• Running it with names attached. A non-anonymous safety check measures nothing except how confident people are that you will not retaliate — which is the thing you were trying to find out.
• Asking someone to explain a low vote. This is the fastest way to guarantee every future check reads 5. Say up front that you will not ask, then do not.
• Collecting the number and running the session unchanged. If a low score changes nothing, the check is theatre and the team will score it politely from then on.
• Treating the score as a verdict on the manager. Safety is shaped by history, workload, org changes and previous employers. Discuss what would raise it, not who lowered it.
• Only running it when things are bad. Without a baseline from a normal week, one low reading is uninterpretable.
• Skipping the definitions. Without the five levels read aloud, people score their mood rather than their willingness to speak, and the numbers are not comparable across sessions.

How to run a Safety Check in QRetro

1. Create a board from the Safety Check template — the five levels are already set up as columns.
2. Share the link; participants join in the browser without an account.
3. Read all five definitions aloud and confirm the vote is anonymous.
4. Give everyone two minutes to place a single card in one column, with cards hidden until the timer ends.
5. Reveal the board, state the totals, and agree out loud how the session will be run.
6. Open your retrospective board and run it under those conditions. Record the distribution so the next check has something to compare against.