Write a description the panel can grade
The panel can only judge what you tell it. A one-liner gets a shallow verdict; a clear paragraph or two gets a sharp one. Cover three things:
- 1. The problem — who has it and how badly.
- 2. The solution — what you'd build and why it's different.
- 3. The money — who pays, how much, how often.
Then pick the right lens in the Evaluate as dropdown — judging a content idea as a business venture will give you a misleadingly harsh verdict.
Reading the verdict
You get a final score, a GO / PIVOT / NO-GO recommendation, a synthesis summary, and five individual verdicts. Don't skip the individual ones — the most useful insight is often a single agent's objection, and the disagreement between models is where the real risk hides. The Risk agent is paid to be a pessimist, so read it as a stress test, not a final word.
A PIVOT is not a rejection. It usually means there's a real idea buried in the wrong packaging — the best angle box tells you where to dig.
Sharing & privacy
Every verdict has a unique URL you can share. If you marked it public it also appears in the idea catalog; if you marked it private it stays off the feed and is reachable only by that link.
Troubleshooting
“Limit: 20 submissions per hour”
That's a per-IP rate limit to keep bots out. Wait until the hour rolls over and try again.
“Evaluation failed”
Occasionally too many models hit a transient outage at once. Each agent has a fallback chain, so this is rare — just hit Try again and the verdict almost always completes.
An agent returned no content
The synthesis still runs on the remaining agents, so your verdict is valid. Re-running usually fills in the missing voice.