Verification ยท three levels on every entry
How we check entries
Every entry carries one of three levels, with the same glyph in the same place: Verified on our bench, Shown working or Claimed. This page says what each level requires and who can set it.
The three levels
Verified on our bench
Someone on the Boards & Bots team ran it on hardware and it worked.
- What it requires
- Someone on the Boards & Bots team ran it on their own hardware and it did what the entry says. We note the board and setup we used.
- Who can set it
- Only the Boards & Bots team, after running it. The curator can never set this level, and neither can members.
- For example
- We flashed a board with the tool, watched the agent read the serial log, and it fixed the bug it found.
Shown working
The source shows it running on hardware, with video, logs, scope captures or hardware CI results.
- What it requires
- The source itself shows it working on real hardware: a video, photos of the device doing it, logs, scope captures or hardware CI results. The entry says what we saw.
- Who can set it
- The curator or a reviewer, after looking at the evidence in the source.
- For example
- A README with a video of the robot arm moving on the agent's commands, and the logs from that run.
Claimed
The author describes it but shows no evidence we could check yet.
- What it requires
- Everything else, including detailed first-hand write-ups with no media or logs we could check. It does not mean we doubt it, only that we have not seen proof yet.
- Who can set it
- The default for every new entry.
- For example
- A blog post that describes the agent bringing up a new board, step by step, with no photos or logs.
When in doubt, the lower level
The curator is an AI agent, and it is told to choose the lower level whenever it is unsure. A downgrade later would be embarrassing; an upgrade is a nice surprise. Entries it is not confident about wait for a person to review them.
When a level changes
If the author adds a video or logs, or we run it on our bench, the entry moves up. If someone shows us the evidence does not hold, it moves down. Either way the entry gets a short "Updated" note saying what changed.
Have evidence for an entry, or think one is wrong? Send a correction.
What a level does not mean
- It is not a review or a recommendation. It says how much proof we have seen that the thing works.
- It is not a safety check. Hardware can be damaged, and an agent with access to it can make mistakes. Try things on hardware you can afford to lose, with a human-safe setup.
- It cannot be bought. Sponsors and our own projects get the same rules as everyone else.