Observation model
An observation must survive its own uncertainty.
The design preserves missing, unreadable and incomplete evidence. Observation quality is evaluated separately from the state of the system being observed.
Presence
A reachable host or running process is evidence of presence. It does not establish current permission, completed work or a healthy application.
Interpretable state
A health claim needs readable evidence with its own observation time and coverage. Unknown and unreadable observations remain visible in their populations.
Affected component
A recovery claim needs fresh evidence from the component that was affected, including domain output. A successful command does not establish restoration.
Tested local behavior
Missing output cannot quietly become a pass.
Published counterexamples exercise missed initialization failures, unreadable captured output, carried observation time and unsupported recovery counts. Their local tests retain the original failures and fixed source.
Failure, repair and verification →
The local implementation and recorded tests are present. Observer independence, full fleet coverage and natural adoption are different claims and require their own observations; a missing observation does not erase the tested behavior.
Archived operating observation
Scheduler results, storage mounts, available probe paths and service observations remain in the dated record. They are not presented as a live dashboard or a new recovery attestation.
Neurobrick · systems, authority and evidence
Last updated · 22 October 2026