What makes a startup hypothesis actually testable (and why most aren't)

Ask most founders what their hypothesis is and you'll get something like "people will love this" or "there's a market for this." Both sound like hypotheses. Neither is one - and the difference matters far more than it sounds like it should, because a hypothesis you can't fail is a hypothesis that can't tell you anything.

A hypothesis has to be falsifiable, or it isn't a hypothesis

"People will love this product" can absorb almost any evidence you throw at it. Ten people say something noncommittal - well, maybe they're not the right people. Two people are enthusiastic - see, people love it. There's no outcome that would count as failure, which means there's no version of talking to people that could actually change your mind. That's the tell for an unfalsifiable claim: notice whether you can describe, in advance, a specific result that would prove it wrong. If you can't, you don't have a hypothesis yet, you have a hope.

The three-tier way to think about hypothesis quality

It helps to think about hypothesis quality on a simple scale - bad, good, and great - rather than treating "testable" as a binary you either hit or miss.

• Bad is a vague or unfalsifiable claim - it can't be meaningfully tested or disproven, no matter how you phrase the test.
• Good is a specific, testable claim that names a customer and a behavior - for example, "solo engineering managers at 20-50 person startups will pay for a tool that summarizes their 1:1 notes." That's already a real improvement: it names who, and it names what they'd actually do.
• Great is everything Good requires, plus a measurable threshold that defines pass or fail in advance, and it targets the single highest-leverage unknown for the idea - the riskiest assumption, not a comfortable, peripheral one. "At least 6 of 10 solo EMs we interview will say they currently pay for, or would pay for, a 1:1-notes summarizer, when asked directly about their current workflow" is testable and it's aimed at the thing most likely to sink the idea if it's wrong.

Notice that a hypothesis can be specific and still only be Good, not Great - specificity alone doesn't guarantee you're testing the right thing. A hypothesis that's precise but about a minor detail (say, a specific onboarding screen's copy) is still only Good. Great is reserved for hypotheses that are both specific and aimed at the riskiest unknown - the assumption that, if wrong, invalidates the most of what you'd otherwise build on top of it.

Why the threshold has to be set before you gather evidence

The measurable threshold in a Great hypothesis isn't decoration - it's the mechanism that actually protects you from your own motivated reasoning. Without a number committed to in advance, it is remarkably easy, after the fact, to read almost any set of interview results as validating. Three enthusiastic conversations out of ten can feel like a green light if you badly want it to be one. Deciding "6 of 10" (or whatever threshold fits your idea) before you talk to anyone converts a fuzzy, after-the-fact judgment call into a real test you either pass or fail.

Finding the riskiest assumption, not the easiest one to test

There's a natural gravitational pull toward testing whatever is easiest to test, rather than whatever matters most. It's genuinely easier to validate that people like a certain color scheme than to validate that they'll actually pay money for the underlying idea - but the color scheme was never going to sink the company. A useful question to ask when you're picking what to test first: if this assumption turned out to be false, how much of everything else would I have to throw away? The assumption with the highest answer to that question is the one worth testing first, even when it's the most uncomfortable one to put in front of real people.

Turning a vague hunch into a Great hypothesis, step by step

Start with the hunch in plain language - whatever you'd say to a friend at a coffee shop. Then work through three questions in order: Who, specifically, is this about - not a demographic label, but a persona specific enough that it couldn't describe a different idea without changing a word? What specific behavior would prove this true - not a feeling they'd report, but something they'd actually do (pay, switch tools, change a habit)? And what number, decided right now, would count as enough evidence to move forward? Answering all three in one sentence is usually enough to turn "people will love this" into something you could actually fail - and a hypothesis you can fail is the only kind that can teach you anything.

The payoff

A testable hypothesis doesn't guarantee your idea is right. It guarantees that when you go find out, you'll actually find out - instead of collecting a pile of ambiguous, comfortable conversations that could be spun into a yes no matter what people actually said. That's the entire point of writing one down precisely before you start talking to anyone. Nucleus pushes on exactly this - it won't let a hunch pass as a hypothesis, and it helps sharpen a vague claim into one with a real customer, a real behavior, and a real threshold before you spend a single conversation testing it.