bet-experimentlisted

# Reduce the uncertainty The goal is not to eliminate uncertainty. It is to spend a little to remove the uncertainty that is actually deciding the question, and to leave the rest alone. ## First, size the response - **Trivial** — the trial *is* the experiment. Say "just try it and watch for X." Designing a study for a reversible ten-minute change is the waste. - **Moderate** — name the deciding uncertainty and the cheapest probe, with criteria. - **Consequential** — full design plus counterargument and falsification, and write it to the bet file. ## Which uncertainty actually decides this Most unknowns do not matter. The one that does is the one where a different answer changes what you do. Test each candidate: *if this resolved the other way, would the decision change?* If not, it is background anxiety, not a deciding uncertainty. Do not design a test for it, however easy it would be to run — the easy-to-test unknown is a trap, because measuring it feels like progress. Where several genuinely decide the question, take the one with the worst combination of high impact and low cost to resolve. ## The cheapest probe that would move you Ask what the *smallest* thing is that would change your mind, not what would settle the question completely. Certainty is rarely available and rarely worth its price. The catalogue is domain-neutral: a prototype, a benchmark against real data, a spike timeboxed to a day, five customer conversations, a concierge version done by han