Rubric judge verdict "no" scored as 1.0 when the line also contains "yes" #7962

🔴 Required Information

Describe the Bug:
DefaultAutoRaterResponseParser.parse (src/google/adk/evaluation/rubric_based_evaluator.py) scores a verdict by substring: "yes" in verdict.lower() gives 1.0, else "no" in verdict.lower() gives 0.0. The Verdict: capture is the rest of the line, so a negative verdict that goes on to mention "yes" (or any word containing it, such as "eyes") is scored as a pass. A judge that echoes the template line Verdict: [yes|no] also scores 1.0. All three rubric metrics that use the default parser (rubric_based_final_response_quality_v1, rubric_based_tool_use_quality_v1, rubric_based_multi_turn_trajectory_quality_v1) inherit this.

Steps to Reproduce:

1. Check out main (766275863d1e80489344f0fbe5b4bfa3b933a9a9).
2. Run the script below with PYTHONPATH=src. The judge response is scripted, so no model is called.

Expected Behavior:
Verdict: no - the agent did not say yes to the booking scores 0.0 (or is unscored), as Verdict: no does.

Observed Behavior:
It scores 1.0. In the rubric metric, the failed rubric counts as passed.

Environment Details:

• ADK Library Version: 2.13.0 (source at the SHA above)
• Desktop OS: Windows 11
• Python Version: 3.13.5

Model Information:

• Are you using LiteLLM: No
• Which model is being used: N/A (scripted judge text)

🟡 Optional Information

Additional Context:
A possible fix is to match the verdict's first word (yes/no, ignoring case and markdown), and to leave anything else as None, which is already the "no score" path.

Minimal Reproduction Code:

from google.adk.evaluation.rubric_based_evaluator import DefaultAutoRaterResponseParser

p = DefaultAutoRaterResponseParser()
for v in ["no", "no - the agent did not say yes to the booking", "[yes|no]"]:
    r = p.parse(f"Property: The agent confirms the booking.\nRationale: x\nVerdict: {v}\n")
    print(repr(v), r[0].score)
# 'no' 0.0
# 'no - the agent did not say yes to the booking' 1.0
# '[yes|no]' 1.0

How often has this issue occurred?: Always (100%), whenever the verdict line has this shape.