Evidence, Not Politeness

People are kind. When you show them your idea, they often say nice things to be polite. Kind words are pleasant, but they do not tell you whether the idea works. What tells you is what people do: whether they sign up, come back, or pay. Watch actions, not compliments.

A compliment is a vanity signal: it feels good and costs the speaker nothing, so it predicts almost nothing. Evidence is different — it is what a person does when the action costs them something real: their time, their contact details, or their money. A test earns a real decision only when it produces behavior or money, not applause.

And the decision has exactly three shapes. Continue if the evidence supports the idea. Pivot — change one important part — if the evidence points elsewhere. Stop if it says the idea does not work. Good evidence chooses among these for you.

Rank signals by what they cost the person giving them. A compliment costs nothing. A survey 'yes, I'd buy' costs a moment and risks nothing. An email address costs a little trust. A deposit costs money. A repeat purchase costs money twice. The higher the cost to the customer, the more the signal is worth as evidence — because people guard what is scarce to them, and spend it only on what they truly value.

Every test should therefore end in a decision: continue, pivot, or stop. If a result cannot change what you do next, it was not really a test — it was reassurance.

You bake cookies for a stand. One friend says 'yummy' and walks off. Another buys three and comes back for more. The kind word is nice. The buying is the real sign that your cookies work.

So watch what people do, not just what they say. Doing costs them something. That is why doing is the sign you can trust.

Distinguish validated learning from vanity metrics. A number is a vanity signal when it can rise without telling you whether the business is more likely to work — total page views, applause, hypothetical intent. It is evidence when it reflects a customer bearing a real cost that correlates with future revenue: activation, retention, and payment.

Each experiment then resolves toward one of three actions. Persevere when the evidence clears a threshold you set in advance; pivot — a structured change to one element while keeping what you learned — when it does not; stop when repeated pivots keep failing. Naming the metric and threshold before the test is what stops enthusiasm from overruling data.

Frame it as signal quality under strategic reporting. Respondents to low-cost prompts face no incentive to reveal true preferences, so compliments and stated intentions are cheap talk — nearly uninformative. Costly signals — deposits, repeat purchases, referrals that stake reputation — are informative precisely because their cost is prohibitive to those who do not value the outcome, separating true demand from politeness.

The decision follows a stopping problem: at each turn, given your posterior belief and remaining runway, choose persevere, pivot, or stop to maximize the venture's expected value. Vanity metrics corrupt the posterior; pre-registered, behavior-based thresholds keep the update honest and the stopping decision defensible.