1. The hypothesis

§The question: what do we believe, and what would we see if we were wrong?

§The running list from the fifth whitepaper was a list of assumptions with numbers. An experiment takes one of them and turns it into something that can be tested, which means turning a belief into a prediction with two possible outcomes.

§ 1.1From assumption to hypothesis#

§An assumption is a belief the team holds: "recognized guests visit more often." A hypothesis is that belief stated so that reality can contradict it: "if we show returning guests their tier on arrival, then cross-brand visit frequency in the next quarter will be at least ten percent higher than the baseline cohort, because recognition rewards return."

§The three parts are load-bearing. If we do X names the change. Then Y will happen, measured this way names the observable prediction. Because Z names the mechanism, which matters because a hypothesis that comes true for the wrong reason has taught you nothing you can reuse.

§ 1.2Falsifiability#

§Karl Popper's line between science and everything else is the line this chapter draws: a statement is only meaningful as a test if there is an observation that would refute it. A hypothesis with no false outcome is not a hypothesis. It is a plan to confirm what you already believe, dressed as an experiment.

§The tell is a prediction with no number and no threshold. "Engagement will improve" cannot be false, because any movement can be called improvement and any lack of movement can be called too early to tell. "Day-one profile completion rises from 40% to 60%" can be false: it can come in at 42%, and 42% is a finding.

§ 1.3The mechanism matters#

§The "because" is the part teams drop, and it is what makes learning compound. A hypothesis that predicts an outcome and names no mechanism, when confirmed, tells you only that this one thing worked. A hypothesis that names the mechanism, when confirmed, tells you something about your users that you can apply to the next ten experiments. When the two-step onboarding works because the drop-off was at steps three and four, you have learned that this product's users abandon on effort, which shapes every future flow.

§When the mechanism is wrong but the number is right, you have learned that too, and it is often the more valuable finding. The tier feature might raise visits not because recognition rewards return but because the tier email reminded people the group existed, in which case the email, not the tier, is the product.

§ 1.4What you leave with#

§One hypothesis, in the three-part form: the change, the observable prediction with a number, and the mechanism. Written before anything is built. Its false outcome named explicitly, so that when the result arrives, the team already agreed what it would mean. The next chapter is how to choose which hypothesis to test first.