Nobody needs to be told their idea is too broad. They arrive having noticed, and what they get back is the advice they already gave themselves: narrow it down, pick a niche, find your ICP. The decidable questions are different — what a narrowing must CONTAIN before it counts as one, and whether a narrowing exists at all for this founder.
So the analysis returns either a wedge or a refusal, and the refusal is the reason it is worth anything. The instruction is blunt: if every remaining slice still needs enterprise sales, still needs a conversation, or still runs through a platform the founder said no to, say so and do not invent a fake wedge.
That clause exists because inventing one is easy and rewarding. A plausible narrowing can be generated for any idea in a sentence, it reads like insight, and the founder cannot tell it apart from a real one until months have gone.
Note where 2 of the 3 refusal triggers are read from: the founder profile, not the idea. An idea that narrows beautifully for somebody who will get on calls has no wedge at all for somebody who will not — which is why a generic how-to-niche-down post cannot answer this, and why the profile is treated as binding rather than as a preference. A narrowing you refuse to execute is not a narrowing.
The contract asks for 7 things. 4 of them are mandatory, and here is how that list was produced rather than transcribed: each field was blanked in turn and fed to the invariant that guards real output. These are the ones it refuses to pass. If the rule changes, the list changes with it.
The other 3 are asked for and NOT policed, and we publish that too, because a page claiming an enforcement guarantee should say exactly where the guarantee ends.
The mandatory one people quietly skip is the explicit what-this-is-not. It is mandatory because a narrowing you have not stated as a CUT is not a narrowing, it is an emphasis — everything you kept in reserve is still in the product, still in the roadmap, and still in the conversation with the first customer who asks for it.
Second mandatory field worth its own paragraph: a NAMED buyer. Not small businesses, not teams — specific enough that you could go and find twenty of them this afternoon. The most common fake narrowing is a broad idea with a narrower adjective in front of it, and the named-buyer requirement is exactly what that fails.
The refusal branch is thinner than the wedge branch on purpose: it requires a reason and nothing else, checked the same way by building a refusal with every other field empty and watching the invariant let it through. A one-sentence no is a legitimate output, and knowing that before you receive one is better than assuming something broke.
What it deliberately will not do: re-score your idea. The projection is neutral — the decision comes back unknown and the score is a placeholder 50 — because a narrowing recommendation is not a verdict, and letting it produce one would mean scoring an idea nobody has evaluated yet. Narrow first, then judge the thing you intend to build.
And the suggested next step is constrained to async, scanned against 24 banned conversation phrases before you see it. Not a style preference: a validation step that requires a conversation is a step most people never take.
The derivation, both field lists, and the before-and-after: https://whittleos.com/guides/my-startup-idea-is-too-broad
Top comments (0)