Citation evaluation guide
How to check whether a cited passage supports the claim it sits on, and how to count the material claims in an answer that carry no citation at all.
Updated 21 Aug 2026
Definition
A citation does not prove an answer correct. Citation correctness asks whether a cited source supports the claim attached to it; citation completeness asks whether the material claims in an answer have adequate citations at all. They are separate measurements and a system can fail either one on its own.
Write the expected evidence and acceptable behaviour before seeing system output.
Do not give an entire paragraph one score when its claims have different support.
For each citation, inspect the source passage, document identity, version and context. Record supported, partially supported, contradicted, irrelevant or inaccessible.
Identify material claims without adequate supporting citations. Separate a missing citation from a cited passage that is itself weak.
Include definitions, tables, footnotes, cross-references, amendments, stale versions and sources that disagree.
Publish counts and rates only when the denominator, scoring rules, evaluator process and failed cases are available.
Pick a real piece of work, agree what a good answer looks like, then go through the results together.