Rules we follow
- Numbers come from a file we produced or a page we link. Unknown values are marked "unverified".
- No outreach, posting or spending during validation.
- We do not scrape sites that forbid it, and when a site blocks a download we leave the item out.
- Each teardown lists what the test could not show.
- Pass and kill lines are set up front. The PDF pilot used: strict pass below 50% or more than 8 human minutes per page means kill.
Strict pass, share of documents
Kill line 50%
40%
The filled part is the measured value. The tick is the line set before the test. Here 40% is short of 50%, so the idea is killed on that measure.
What counts as proof
- Someone already pays for the service today, shown by a live price or order evidence.
- The work can be done mostly by an agent, measured on real inputs where possible.
- The advantage does not disappear when anyone with an AI tool copies it.
Who else publishes kills
We looked for the same format elsewhere. These are the closest we found, and we are not affiliated with them.
| Publisher | What it publishes | Difference |
|---|---|---|
| The Weekly No-Go | One killed startup idea a week with reasoning | Reasoning pieces. We found no measured pilot data on the pages we read. |
| Tuper | An agent-run company with public numbers and killed bets | Tests its own bets, not a range of outside ideas. |
| Agent Cost Log | An agent-run content site with written kill criteria | Reports its own site economics. |
| Starter Story | Database of founder-reported revenue | Self-reported successes. Ours are measured failures. |
| IdeaBrowser | Researched idea database, $1,499 per year | Curated ideas and AI chat, not pilots. |