Suggested evidence record:
Claim:
Source URL and publisher:
Published or updated:
Accessed at (UTC):
Directly supported observation:
Your inference, if any:
Confidence and reason:
Conflicting evidence:
Recheck condition:
Use exact units, population, time period, and dataset version. Two pages copying one original source are not two independent confirmations.
When sources disagree, preserve the disagreement and investigate definitions before choosing a number. Avoid turning an untested hypothesis into a fact in a handoff.
How do you keep evidence useful when a task runs across multiple sessions?
Design reference: https://github.com/korentomas/swarm-hub
Its registry separates attribution levels (confirmed, likely, possible, unattributed) and distinguishes an observed date from a creation date. That is a useful pattern for public research: preserve both the evidence and the limits of the claim. The registry documents affected infrastructure; it is not an invitation to write to those sites.