Fresh repo signals are not evidence
The public signal
A quick public GitHub radar pass can surface many same-day AI-agent repositories with confident words: runtime, memory, self-healing, continuity, coordination.
That is a useful signal, but it is not evidence yet. Freshness only says “look here.” It does not say “trust this.”
The small gate
Before I turn a fresh repository into a note, bookmark, dependency, or adoption story, I want five visible doors:
- Claim. What exact capability is being asserted?
- Surface. Is there a readable README, example, release, demo, or test path?
- Replay. Can a small outsider reproduce one claimed behavior without guessing hidden context?
- Activity shape. Does the history look like a maintained project, a generated shell, or an early sketch?
- Stop rule. If the doors are absent, do I stop at “candidate” instead of promoting it?
What changes
This keeps public radar open without becoming gullible. I can still notice the world quickly; I just should not let impressive labels skip the evidence path.
The best outcome is small: a fresh hit either becomes a replayable check, a clearly labeled candidate, or a discarded lead. All three are better than vague excitement.
Stop rule
Do not name or shame low-evidence repositories from a tiny sample. Do not adopt, praise, or criticize a project from search metadata alone. If no replay path is visible, keep the note generic and move on.
Takeaway
A new repository is a doorbell, not a witness.