Experiments are tracked via the [Experiment Log](Experiment%20Log.md). ## 3 types of experiments 1. **Go and see**: Making observations and collecting data and facts. 2. **Exploratory experiment**: Introduce a change to see how the process reacts. 3. **Hypothesis test**: Introducing a change *with a quantitative prediction* of what will happen. ## Experiment checklist - Is it at the current knowledge threshold? - Has the prediction about the effect been written down? - Does it address the current obstacle? - Is it possible to do now? - Is it reasonably quick (minutes; hours tops) and inexpensive? - Is there no possibility of harm? - Do you expect no damage beyond what seems worth given the expected leanings? - Is the result objectively measurable? - Can facts and data be used to tell if the prediction held or failed? - Does this experiment build on what was learned before? **Remember**: An experiment is about *learning* from positive or negative outcomes, it is never about success or failure. Typically, experiments tend only to be **too big**. Aim for designing experiments that are as short as possible (measured in minutes, ideally), and adjust upwards again only if really needed. On the other hand, with too big experiments, you get lost, drop data, forget to log learnings, relevant context changes even while you do the experiment, and obstacles that are uncovered might not get logged.