[Bug - Agent Analytics] Time to First Token vs Total Time

[putrasyafiq]

Describe the bug
I've noticed that in the BigQuery Agent Analytics Plugin, the time to first token vs the total time taken under the latency metrics is always exactly the same. This causes some discrepancies when implementing observability.

To Reproduce

1. Install the BigQuery Agent Analytics Plugin.
2. Run an LLM prompt.

Expected behavior
Expected the time to first token (ms) to be lesser than total time taken.

Screenshots

Environment

• OS: Linux
• Python version(python -V): 3.13.7
• ADK version(pip show google-adk): 1.22.1

[klateefa]

@putrasyafiq Thanks for reporting!
This happens because the time to first token is recorded when the request finishes, instead of when the first streamed token is actually emitted. As a result, both metrics use the same end timestamp.
The fix is to capture time_to_first_token_ms at the moment the first token is produced inside the streaming loop and persist it only once.With this change, streaming responses will correctly report time_to_first_token_ms as less than total_latency_ms.This issue is limited to the analytics plugin and does not affect BigQuery ingestion
Thank you.