Retrospect to FAIL fast and OFTEN
Some teams steadily come to a state in which their retrospectives result in less and less improvements. They have been continuously improving in small steps and improved quite a bit since their start, but eventually they stopped learning.
The thing is that you can learn from things you do right but you can also learn a lot from doing things wrong. When for example you try something that you expect to work and it actually works you confirm your current assumptions and values.
When on the other hand you do something you think will not work and it does work, you just learned that one of your assumptions, values or thinking patterns are incorrect.
The issue is that in general we prefer to learn from doing something that works and try to avoid doing things that we expect not to work.
For example, in retrospectives we look back to see how we can improve. We analyze what the root cause of a problem could be and then come up with a hypotheses and accompanying sprint backlog actions to try out and see if our hypotheses holds in the next iteration. The hypothesis we make is based on our current assumptions and values and this is good, but only goes so far.
The issue with learning from doing things that work is that we limit our learning. We narrow the set of possible outcomes because we in general define our actions based on our current assumptions and thinking models. We do not challenge our assumptions and values this way.
Chris Argyris discribes his double-loop learning as a model for challenging your assumptions and values to foster deep learning.
Therefore I propose that we should challenge our assumptions and values by setting up experiments that we think will fail. We should consider having retrospective actions that we expect to fail, we should retrospect to FAIL fast and fail OFTEN.