An agent trajectory dataset needs more than final answers

Automated excerpt

That means keeping the task, observations, tool calls, tool results, decisions, corrections, and authoritative outcome. Observations, permitted reasoning, tool names, arguments, results, errors, and final outputs should use a versioned schema. I remove duplicates, leaked tests, private values, malformed tool calls, and trajectories without an authoritative outcome.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.