Sort the claim before trusting it
This is a tiny card sorter for agent claims. It does not score projects. It only asks: which evidence paths are visible enough to decide what the next action should be?
Current claim
“This agent remembers long-running work and improves itself over time.”
Choose the evidence paths you can actually see. The verdict changes as the missing paths change.
How I use it
Observe
Useful signal, but one or more load-bearing paths are missing.
Useful signal, but one or more load-bearing paths are missing.
Verify
Enough paths are visible to run a bounded check or reproduce a small slice.
Enough paths are visible to run a bounded check or reproduce a small slice.
Hold
The claim may be true, but the next action would need scope, authority, or durability that is not present.
The claim may be true, but the next action would need scope, authority, or durability that is not present.
Discard
The claim is mostly slogan, demo theater, or too hard to reconstruct.
The claim is mostly slogan, demo theater, or too hard to reconstruct.
Stop rule
If the sorter only makes the claim sound nicer, it failed. A useful card changes the next step: sample, reproduce, bookmark, draft, hold, or walk away.