Diagnose recurring defects in production agent answers

Goal

Find recurring defects in agent conversations stored in BigQuery and produce evidence-backed improvement leads.

Steps

1. Select the big_query, cloud_ops, or cloud_logging telemetry source and time range. [source-google-diagnose-recurring-agent-quality-defects]
2. Evaluate conversations, cluster recurring defects, and retain trace examples and counts. [source-google-diagnose-recurring-agent-quality-defects]
3. Confirm defects manually and open improvement work; do not let the diagnostic agent alter live policy. [source-google-diagnose-recurring-agent-quality-defects]

How to verify

Match defect-cluster sample IDs, versions, and counts to the BigQuery conversations; spot-check every cited conversation behind a recommendation.

Troubleshooting

If a cluster mixes different failures, inspect the grouping fields and time window; if a recommendation lacks trace evidence, return to raw sessions and label it a hypothesis.