The idea
When observations repeatedly determine the next experiment, it becomes easier to discover patterns that exist only in the data already seen. An explicit prior hypothesis constrains what counts as evidence.
Worked example · illustrative
Did the new study routine help?
Situation. You try a new study routine and afterward search your results for improvement. One quiz score is higher, so you consider declaring the routine a success.
The tempting conclusion. The higher score shows that the new routine works.
What changes the view. You also changed study time, chose an easier topic, and looked at several outcomes before selecting the improved one. The result can suggest a hypothesis, but it is weak confirmation. Before the next trial, specify what you expect to improve, how you will compare similar material, and what result would count against the idea.
What this supports. Use a fresh comparison to test the expectation instead of repeatedly redefining success around the results. This cannot remove every source of variation, but it makes the evidence easier to interpret. Exploration can generate the hypothesis; it should not be presented as if it had already tested it.
Useful when
Watch for
Not all discovery should be hypothesis-first. Exploration can legitimately generate hypotheses; just distinguish discovery from confirmation.
Where this came from
Quantitative research, Finance, Machine learning