Fig. 1 · How we judge

How we judge an idea.

Before you build, know the problem in the words of the people who have it, whether the one who suffers is the one who pays, who already sells it, where those people are and what it costs to reach them, and the one test that ends in a sale.

1. 1

   Criteria before the verdict

   Before we read anything, we write down what would make us say no to your idea. It's frozen before any evidence comes in, you see it next to the answer, and the scale below is the one we score against.

   The verdict is then scored against the criteria you saw, not against ones written after the evidence came in. Each of the five questions gets a score from 0 to 5, from the share of what we read that supports it. When we read too little to measure one, it gets no score instead of a zero, and every question says what would move it up.

2. 2

   A published rejection rate

   We count every idea we judge and every no, and we publish the share with the method behind it, one click from the homepage.

   We haven't judged an idea yet. The count appears here with its denominator when we have, and the criteria are at version 5.

   Not yet measured

3. 3

   The one test that ends in a sale

   No amount of research validates an idea, and a purchase does. The last step of every plan is one purchase at the stated price, with the message to send and the number that would count as a no.

   We don't run the test for you. We tell you exactly what it is, and we update the answer with what happened.

Fig. 2 · What we read

Where the people with the problem write, read in their words.

For every idea we choose about ten places where the people who have the problem write, from the forums and communities for their line of work, the sites that sell to them, the reviews their customers leave, and the open web. We search each one in the words you gave us, read what comes back, and then search again in the words the people we found use, because people rarely call their problem what a founder calls it.

A post counts only when a person is in it. Posts that sell something, including the ones that tell the problem in order to sell the fix, are set aside, and so are posts that read as written by a machine. The share we score is taken over the posts on topic, not over everything the search returned, so it is about the people and not about our queries.

• States the problemThe post says it, in the person’s own words, about something they did or tried.
• Asks for a fixThe post asks for a way to do the thing, or whether a tool for it exists.
• Uses a product for itThe post talks about something they already use or pay for to deal with it.

For our own idea, read on 2026-10-01: ten places, 104 searches, 480 posts read, 401 on topic, 209 quotes kept, 90 people counted. The quotes are on the first page.

Fig. 3 · The scale

Each of the five questions is scored 0 to 5 from a share of what we read. These are the two ends of each scale.

The five questions themselves, in the order we ask them, are on the first page. Every score comes from a percentage, so reading more posts can't raise it. How much we read only decides whether there's enough to score a question at all; under the number in the last column, the question gets no score and says so. Levels 1 to 4 are the bands between the two ends.

* These percentages aren't final. We're still tuning them against real reads, and the criteria version goes up each time they change.

Fig. 4 · What would make us say no

One sentence, written before we read, that we can count.

For every idea we write one sentence before any post is read: the finding that would make us say no. It is always something we can count from what we read, never something you would have to go and do, and it is frozen before the search starts. The answer counts what we found against it and says whether the line was met, not met, or couldn't be counted from what we read. Only the first two go into the published rate.

For our own idea, the line was: “We stop if under 5% of the posts on topic in builder rooms have a person who states the problem of nobody buying or asks for a fix, or if none of the printed prices for idea-checking tools is at or above $20 a month.” We read 90 of 401 posts on topic stating it, and one printed price at $25 a month, so the line was not met.

The read never sends you to do our work. It won't tell you to go and interview ten people as the answer; where talking to someone is the right next step, it appears as a suggestion in the next steps, with who and where.

Fig. 5 · Where every line came from

Three states, on every claim, in type and not only in colour.

• Sourced · 27

  A count, a quote, a URL and a date. The line rests on something you can open.

• Your assumption

  You typed it, or you changed what we found. The verdict says how much rests on it.

• No source says this

  We looked and found nothing. That's an answer in itself, and it's shown as a footnote with nothing under it.

Every number carries its denominator, so you read 0 of 27 paid and never nobody paid. Every quote carries who said it, where and when. Every source is numbered once in the read and opens from its number.

Fig. 6 · What we don't do

Four things you won't find in a read.

• A score for a question we couldn't read enough about. It gets no score, never a zero.
• A number without its denominator, or a quote without where and when it was written.
• Homework. We don't answer with “go ask your customers”.
• A yes for everyone. The rate of no is published, with its count, once there are enough ideas to mean something.

Your agent gets the same read, as data, on the same limits. Connect.