Start With a Hypothesis
What did I expect to observe before I saw the result?
Situation
Repeated experimentation is producing increasingly good-looking results.
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.
Useful when
Backtestingrepeated experimentsfeature discoveryexploratory analysis
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