The test
Could you show it to someone who has no stake in the answer, and would they draw the same conclusion?
That's it. Not 'do you believe it'. Not 'does your team agree'. Whether the artefact exists and whether a disinterested person reading it lands where you landed.
Almost everything founders describe as evidence fails on the first clause. There is no artefact. There is a memory of a conversation that went well.
The four things that masquerade as evidence
- —Enthusiasm in a meeting. People are warm to founders. Warmth is not demand.
- —Advisor agreement. Advisors are pattern-matching against their experience, not your customers'.
- —Competitor existence. That someone else raised money proves someone else raised money.
- —Your own past experience of the problem. You are one data point who has since become unusual.
Why the confusion is structural, not personal
Founders are selected for conviction. The job requires holding a belief through a period where no external signal supports it. The same trait that makes someone able to start a company makes them poor at auditing their own certainty.
So the answer isn't to become less convinced. It's to build a process that produces artefacts, so that conviction and evidence stay in separate columns where you can see the ratio between them.
What an artefact looks like
A transcript in which someone described, unprompted, doing the thing you think they do. A screenshot of the spreadsheet they use instead. An invoice for the tool they abandoned. A calendar with the recurring meeting where the problem gets discussed. A signed pilot.
None of these require the customer to predict their own future behaviour, which is the failure mode that ruins most research.
The two-column habit
After every customer conversation, write two lists before you do anything else. Left column: what I observed or was shown. Right column: what I was told would happen, or what I inferred.
Do this for ten conversations and the ratio will tell you whether you have a research programme or a reassurance programme.
If the right column is longer, you don't need more conversations. You need different questions.
Where this leads
The useful outcome of an honest audit is rarely 'stop'. It's usually 'the thing I was most confident about is the thing I have least support for' — which is a very cheap discovery in month two and a very expensive one in month fourteen.