The promises and the honest questions
What does Discover Agents promise about evidence?
Why it works this way
Where we cannot back a figure, the card says "our estimate".
We would rather label a guess than let it pass as a fact. The expensive mistake is building the wrong thing, and a confident-sounding invention is the quickest way there. So a figure we cannot back says so, and a silence is not turned into a verdict.
What we gave up: A board where every card looks complete and confident. A demo with fewer confident numbers reads as less impressive, and that is the point.
What it is
Six promises about how Discover Agents treats evidence. The idea under all of them is that this product is built not to flatter your idea. It is why a card sometimes says "our estimate", "No one quoted yet" or "not checked" instead of looking confident. Each promise says what the product does when it cannot back something, not that nothing ever goes wrong.
How to use it
Six things to look for on your board.
- A claim is not a measurement. Where we cannot back a figure, the card says "our estimate". A problem's reading says what it rests on: "Our read, from N quotes on this card", or "No one quoted yet".
- When nobody wrote about a problem, the card says so. A problem nobody wrote about says "not checked" and gets no meter. It is not given a low score.
- Your words stay yours, and quotes are checked. Your account of a round is kept exactly as you said it, and a finding is recorded only when it comes from what you said. Nothing becomes your bar until you commit it yourself. When a quote from your conversations is saved to your board, it is checked against the transcript you added. If the words are not there, it is left out and Iris, your AI guide, says so.
- Once you have evidence, the bar is locked. A bar set after you have looked only describes what you saw. The next round gets a new bar.
- When a step is due, its card appears. A round waiting for its result, or a road to choose once a result is in: the card shows up in your conversation without you asking.
- We say when we have not looked. On the Blueprint, a scan that found nothing says so. A scan that has not run, or that failed, says that instead. Neither is dressed up as a quiet market.
To check where a figure came from, open "Sources we read" on a board. It is a list of pages we read for the board.
What it will never do
- Turn silence into a verdict. A problem nobody wrote about is not given a low score.
- Call a rating proof that anyone will pay. It reflects what people wrote, not what they did.
- Say a result proves your idea works. A pass means the next tests are worth running.
- Present a simulated answer as evidence.
If something looks off
- A figure carries no label and no source. Tell us through Tell Omer. We treat it as a defect.
- "No one quoted yet" on a problem you know is real. It means no quote is on the card, not that the problem is small. Ask someone who has it.
- An older board says "our estimate" more often than a new one. That is on purpose. Lines written before the labels existed are marked rather than trusted.
- A scan failed. The page says so and offers a retry. A failed scan is not an empty market.