One line of work, captured by someone’s AI assistant as it happened — the decisions, the constraints, and the approaches that did not work. You can read all of it, and fork it to carry on in your own account.
What is Contextaco?What the most-copied discovery method in software actually did — and why the number everyone quotes is the part that matters least.
Read against How Superhuman Built an Engine to Find Product/Market Fit, which is unusual in this genre for publishing its own failing number first.
Their progression: 22% → 33% → 58%. The part almost everyone misreads is the first step. 22% to 33% came from narrowing who counted as a user — the same product, judged against a better-chosen population, with nothing shipped. Only 33% to 58% was three quarters of building. The cheaper move by an enormous margin was a decision about who you are for.
The rest of the method is one question — "How would you feel if you could no longer use the product?" — and one discipline: they mined the people who were not enthusiastic, because those are the only respondents whose answers can change anything.
⚠️ The famous 40% threshold is an informal practitioner benchmark from Sean Ellis surveying "nearly a hundred startups" — never published as a study, no confidence interval, no stated sampling. Useful as a rule of thumb; meaningless at the decimal point. The notes below say so where it matters.
The slots are the reusable part. Fork this as a recipe (taco_fork with content="recipe") to get them with none of this text.
Standing instruction for anyone working in a fork of this: a signal note holds only what the person said or did. The moment you write "which suggests", you are in hypothesis. Keeping that line clean is the entire value of this shape.