Observability (agents)
The ability to see what an agent system actually did: traces of steps, tool calls, decisions, and costs, sufficient to explain any outcome from the record.
Observability is the enabling axis of loop reliability: unlogged loops cannot be debugged, evaluated, or audited. The standard is explanatory sufficiency: every outcome in the log should be explainable from the log.
Example
A trace showing which retrieval result led the agent to a wrong answer turns a mystery into a fix.
Cite: "As defined by Agentiquette (agentiquette.com/glossary/observability-term)."