Summary

Implement quality scoring Layers 2 and 3. Layer 1 (CI signals) is already implemented. Consolidates #230 (LLM judge) and #231 (human override).

Design Spec Reference

• SS8.3 Performance Tracking -- D2

Layer 2: LLM judge (formerly #230)

• Small-model LLM judge from a different model family than the agent being scored
• Evaluates task output against acceptance criteria
• Integration with QualityScoringStrategy protocol
• Cost target: ~1 EUR/day
• Specific model to be evaluated at implementation time

Layer 3: Human override via API (formerly #231)

• API endpoint for human quality score override
• Highest weight in the scoring composite
• Integration with QualityScoringStrategy protocol and PerformanceTracker
• Dashboard UI for submitting overrides

Dependencies

• API authentication (already implemented)
• feat: implement quality scoring Layers 2+3 -- LLM judge and human override #230 original deps (five-pillar evaluation framework -- feat: five-pillar evaluation framework for HR performance tracking #1017, CLOSED)