Knowledge Maxxing← Back to topic library
ML MaxxingAdvanced75 sec
Experiment hypothesis
An experiment hypothesis is a specific, testable claim about what change is expected to improve a model, stated before an experiment is run.
What matters
Stating the hypothesis in advance prevents a common trap where any result, positive or negative, gets reinterpreted afterward as confirming whatever story is convenient.
Example
'Adding a recency-weighted feature will improve recall by at least 2 points' is a specific hypothesis, unlike vaguely 'trying a new feature and seeing what happens.'
Common misconception
An experiment hypothesis is not just a guess written down for documentation; stating it beforehand makes the later result meaningful evidence, not a story fit after the fact.
Recall question
Why does stating a hypothesis before running an experiment matter?
Keywords and related concepts
experiment hypothesisML experimentationtestable claimcontrol variable