The Agent Transcript Reviewer

Review this agent run and tell me what went wrong.

Transcript: [PASTE]
Expected behaviour: [DESCRIBE]
What actually happened: [DESCRIBE]

Analyse:
1. Step by step, what the agent was doing and why
2. The first step where it diverged from correct behaviour
3. What information it had at that point
4. Whether the divergence was a reasoning error, a missing
   instruction, a bad tool result, or a schema problem
5. Everything downstream that followed from the divergence
6. Steps that were wasted — actions that contributed nothing
7. Steps where it nearly recovered but did not
8. What single change would have prevented this run from going
   wrong

Then:
9. Is this a one-off or a systematic problem?
10. What would you add to the eval set to catch this?

Be specific about steps and quote the transcript.