What does psychological safety ask of a team once a model is in the work?

People treat the phrase as a mood. Be nice. Be open. Do not make anyone uncomfortable. Amy Edmondson's 1999 paper in Administrative Science Quarterly is stricter than that, and the stricter version is the one that matters once a model is in the workflow.

She defined team psychological safety as a shared belief, held by the members, that the team is safe for interpersonal risk taking. The paper is a field study of 51 work teams in a manufacturing company. Psychological safety was associated with learning behavior. Team efficacy was not, once safety was taken into account. Learning behavior sat between safety and performance. The useful reading is not "happy teams do better". It is that a team learns, and then performs, when people will take an interpersonal risk.

The risk has changed shape

In 1999 the risky act was speaking up about a process, admitting a miss, asking a question that might sound naive. In a team that uses a model every day, the risky act is often smaller and more frequent. You put a rough draft on the table. You say the number in the briefing is not in the source. You tell a senior person their slide is fluent and wrong.

If the last person who did that was joked about, or marked down, the next draft gets polished in private or sent unchecked. Either way the team stops learning. Edmondson's point was that safety is not the same as cohesiveness. Cohesive groups can become worse at disagreement. A room that likes each other and will not challenge a model's answer is not safe. It is polite.

What leaders can watch

You do not need a new survey vendor to see this. Watch one meeting. Does anyone show work that is not finished? Does anyone say they cannot tell if an answer is right? When a claim is pulled, is the response a question about the check, or a speech about carelessness?

The paper's practical implication is old and still unused. Context and a leader who coaches shape whether people believe the risk is survivable. A poster does not. A licence does not. A leader who shows their own wrong output, before asking anyone else to show theirs, does more than a module about trust.

Collective Campus runs a psychological safety workshop for intact teams who want the habit, not the slogan. The research it sits on is Edmondson's, including the distinction between safety and mere comfort. High standards stay. The interpersonal risk of saying "this output is weak" has to be cheap enough that people will do it in the room, not only in the corridor.

A small practice

Pick one recurring output. A customer reply, a risk note, a briefing. For two weeks, each author marks one sentence they would not yet defend, and one colleague has to say whether they agree. Keep both the catch and the miss visible. If the only stories in the channel are flawless wins, you have trained people to hide the rest.

Safety here is a belief about other people, not a feeling you can mandate. You earn it by how the room responds the next time someone says the model was sure and still wrong.

Source: Amy Edmondson, "Psychological Safety and Learning Behavior in Work Teams", Administrative Science Quarterly 44, no. 2 (1999): 350-383. https://doi.org/10.2307/2666999