Causality before presentation
Events can arrive late or out of order. The intended record model preserves execution identity, parent relationships, action type, and timestamps so a tool result can be associated with the call that caused it. A display order alone should not be mistaken for causal proof.
A recovery should not erase a failure
When an agent retries successfully, the original error still matters. The planned timeline keeps both attempts and the transition between them. This helps a reviewer distinguish an uninterrupted success from a result achieved after recovery.
Open the sample explorer to expand a failed lookup and its retry. The timestamps and outcomes are deliberately fictional.
Human intervention belongs in the record
Approvals, declines, clarification requests, and manual retries change the execution. Our proposed model records those transitions explicitly, including wait time where it is observable. It does not infer a human approval from a later successful outcome.
Capture has a boundary
Activity outside instrumentation, unavailable model details, and missing measurements need visible coverage notes. A timeline must not fill in those blanks with invented steps.