Evaluation & Benchmarking
Offline and online evaluation, harnesses and evidence generation.
Research question
What evaluation evidence is sufficient to approve a model or agent for production, and who should generate it?
Scope
- Offline evals, online experiments, human review
- Eval harnesses and LLM-as-judge
- Evidence as an approval artifact
Key findings
Summarize findings from linked notes. Each finding should cite a note or source.
Related notes
Link research notes tagged with this area.
Comparisons & patterns
Link vendor comparisons, benchmarks and architecture patterns relevant to this area.