Evidence over vibes
What we believe about finding things worth building, and the constraints we have put on ourselves to stay useful.
Ideas are cheap. Evidence is not.
Anyone can generate a list of startup ideas. A language model will produce a hundred before you finish reading this sentence, each one fluent, confident and completely untethered from whether a single human being wants it.
The scarce thing was never the idea. It is knowing whether the demand under it is real — and that requires evidence somebody else already created, without being asked, before anyone tried to sell them anything.
Fluency is not proof.
A model describes a market of millions with exactly the same assurance whether the underlying data is forty independent complaints or one ambiguous forum post. Nothing in the prose tells you which.
So we show the data. Every opportunity lists the complaints that produced it, with links to the originals. If our summary and the source disagree, believe the source — and tell us.
We will say when the evidence is thin.
The easiest product in this category to build is one that always sounds confident. It is also the least useful, because a tool that is never uncertain gives you no way to tell its strong findings from its weak ones.
When a cluster is small or comes from one platform, the evidence score drops and the write-up says so. We would rather show you a weak signal honestly than dress it up.
Show the working.
The score is a weighted composite of seven dimensions, and every dimension, every weight and the arithmetic are published. You should be able to disagree with our ranking on principle rather than because you cannot see inside it.
One dimension — evidence strength — is deliberately kept out of the model's hands entirely. It is computed from how many independent sources reported the problem, so nothing can inflate it.
This is a starting point, not an answer.
No database tells you what to build. This narrows a field of infinite possibilities down to a few dozen well-evidenced ones. You still have to pick, and you still have to go and talk to the people who have the problem.
Anyone claiming to validate an idea for you is selling something. Validation is a conversation with a customer. We can only tell you which conversations are worth having.
What we will not do
- Hide weak evidence to make an opportunity look stronger
- Invent a competitor because an empty list looks bad
- Gate the methodology behind a paywall
- Sell what you search for as a signal to anyone else
- Publish a market-size number we cannot derive from the inputs