RUBRIC
LLM Evaluation Rubric Template
A machine-readable rubric defining deterministic checks, statistical signals and model-graded criteria for LLM feature evaluation runs.
- Intended user
- Teams building their first reproducible LLM eval harness
- Required tools
- promptfoo, DeepEval or any eval runner reading JSON
- Input requirements
- Sample outputs plus domain knowledge to calibrate score bands.
- Output example
- JSON rubric consumable by eval tooling after adaptation.
- Compatibility
- Adaptable to most evaluation frameworks
- Download format
- JSON (.json)
Safety notes
Model-graded scores are heuristics: validate against human-labeled samples before trusting trends.