Encyclopedia Evalica / Evaluation / Alignment
Alignment
/uh'leyen.muhnt/The process of calibrating an LLM judge so its scores match human judgment. Alignment typically involves iterating on rubrics and prompts until agreement is high. (noun)
“After our alignment sessions, the LLM judge scores matched human ratings much more closely.”
Customer example
Navan treated alignment as an iterative process, tuning its eval prompt like an ML classifier until the evaluator reached >0.9 macro F1 and using reasoning to debug failures. Read more
Related Evaluation terms
- Absolute scoring •
- Agent •
- AI eval •
- Annotation schema •
- Baseline •
- Baseline experiment •
- Benchmark •
- Calibration •
- CI/CD integration •
- Coherence •
- Confidence interval •
- Eval harness •
- Eval leakage •
- Experiment •
- Factuality •
- Failure mode •
- Faithfulness •
- Feedback signal •
- Groundedness •
- Hallucination •
- Inter-annotator agreement (IAA) •
- LLM-as-a-judge •
- Loop •
- Model comparison •
- Multimodal •
- Non-determinism •
- Offline evaluation •
- Pairwise evaluation •
- Pass@k •
- Playground •
- Quality gate •
- RAG (retrieval-augmented generation) •
- RAG evaluation •
- Reference-based scoring •
- Reference-free scoring •
- Regression testing •
- Release criteria •
- Remote evaluation •
- Rubric •
- Safety •
- Score distribution •
- Scorer •
- Semantic failure •
- Signal-to-noise ratio •
- Task (eval task) •
- Toxicity score
From the docs
Braintrust is the observability platform for agents in production. By actively applying intelligence to agent traces and automatically surfacing critical patterns, Braintrust helps teams at Notion, Stripe, Box, OpenAI, and Cloudflare ship quality agents at scale.
Start building