Founder-market fit is usually discussed as credentials or as encouragement, and neither is answerable. We had to make it decidable, because in our product this is the one analysis where a fact about the FOUNDER can kill the idea outright. That is a strong thing for software to do, so most of the work went into what it costs.
The checks: 7 of them, and not one is about a CV. Can you build this, explain it, sell it, support it — and the one nobody expects, can you tell whether the output is any good. Plenty of buildable products fail there: you ship it, and you have no way to know whether what it produced was right, which means you cannot improve it and cannot defend it when a customer says it is wrong.
Each check grades in a vocabulary of 4, and the interesting boundary is not severity. It is whether learning is a plausible route at all. Weak closes with work. Foreign does not, and the honest response to foreign is a different idea rather than a longer study plan.
The aggregate lands on a ladder of 4 rungs whose consequences are computed rather than editorial, and the bottom rung is the one to read. It maps to a kill-filter failure, on a filter that is NOT one of the softenable ones — a platform dependency can be carried forward with a flag; this cannot.
So we attached a matching obligation: the bottom rung is required to cite a specific fact from the profile the user actually wrote. Not a vibe about the domain being hard — a named thing they said about themselves. The score it produces is 10, and it caps sales fit at 2.
The design decision I would defend hardest is the middle rungs. A mediocre fit does NOT subtract points across the whole scorecard. It puts a ceiling on the one category it has any business affecting and leaves the rest of the evaluation to the idea.
Then the part that beats the arithmetic. Alongside the score, the analysis returns a plain recommendation to keep the idea or drop it — and a drop recommendation lands on the same decision at every rung, including the top one. If the analysis has said in words that this is not your idea, letting a good numeric fit outvote it would be the numbers laundering a judgement.
Two failure modes this kind of analysis has, and the rules against them. It invents an attribute about you — our first live evaluation caught exactly that, checks asserting a founder had a low tolerance for something the profile never mentioned. The fix is an explicit grounding rule: only facts that literally appear in the idea or the profile, and an unstated attribute is unknown rather than scored against you. It is a convincing failure because an invented attribute sounds like a reasonable inference.
And it hands you a study plan made of nouns. Every gap has to survive one test — could I search for this and get a how-to — so Stripe billing webhooks passes and fintech compliance does not.
There is also a banned list in the output: no motivational language and no coaching, which means no mentors, no advisors, no get a co-founder. Those are real options, and none of them answers whether you can sell this thing next month.
The ladder, the two rules and the banned list: https://whittleos.com/guides/am-i-the-right-founder-for-this-idea
Top comments (0)