Wiki

Core workflow, key terms, how to read an eval result, and a first demo path through the tool.

How to read and generate a 13-section evaluation report — from verdict to appendix.

15 core principles and 7 anti-principles that govern how evaluations are designed and interpreted.

10 reference dimensions, 5 starter rubrics, weight normalization rules, and versioning strategy.

How claims are labeled as supported, partially supported, unsupported, or contradicted.

Faithfulness to retrieved context — how to score source utilization and detect citation drift.

How to detect measurable drops between runs on the same dataset, rubric, and retrieved context.

Queue ordering, override mechanics, two-reviewer policy, and calibration loop protocol.

How the four scoring methods work in this tool — LLM judge, claim pipeline (groundedness), deterministic checks, and human review — and how to configure and test them.

8 failure modes, judge selection rules, calibration loop, and when not to use LLM judges.