llm-eval-playbook

Builds LLM evaluations end to end: representative test sets, task-matched metrics (BLEU, ROUGE, BERTScore, precision/recall, MRR/NDCG), groundedness and toxicity checks, calibrated LLM-as-judge and human annotation, statistical A/B comparison, regression gates and honest benchmark reporting. Use when the user asks to "evaluate my model", "compare two prompts", "set up an eval suite", "run an A/B test on model outputs", or "catch quality regressions before release". Not for training or fine-tuning models, or for calling paid providers and downloading weights without the user's approval.

Version

v1.0.1

Size

13 KB

Updated

2026-10-06

Your client verifies the signature before reading. This explicit manual read is available on every plan. Free automatically loads five chosen skills; paid plans automatically load relevant skills across the library.