Assistant Evaluation Basics (3-metric core)

Use this to get a fast, defensible quality baseline before optimization.

Metrics

1. Task success rate — % completed without escalation.
2. Time to correct outcome — median time from prompt to usable result.
3. User trust rating — confidence score after each task.

Protocol

• Pick 3 representative user tasks.
• Run 5–8 moderated sessions.
• Capture handoff failures and correction loops.
• Prioritize fixes by impact on success + trust.

Output template

Task | Success | Time | Trust | Failure mode | Next fix