[Advent of Agents] Day 22: ADK Evaluation: Trajectory Tests and Rubric-Based Scoring

[abraha...@google.com]

Hi everyone,

Day 22 of Google's Advent of Agents — Season 2 is now live! 🌸

Stop relying on 'vibes-based' manual testing and start catching agent regressions automatically with deterministic trajectory checks and rubric-based scoring. Learn how to integrate ADK Evalsets into your CI/CD pipeline to ensure your agent follows the correct reasoning path every time.

Today's Highlights:

• Enforce strict or flexible tool-call sequences using trajectory tests to validate agent reasoning.
• Leverage judge models for rubric-based scoring to grade response quality without needing static reference answers.
• Automate evaluations within GitHub Actions to catch performance drops and logic regressions before they reach production.

👉 View Day 22 Drop Here

---

Missed a day? View the full archive here.

Happy building!
The Advent of Agents Team

© 2026 Google Cloud · Season 2 · Spring Edition