An arXiv paper argues that current evaluation of automatic crisis reporting is measuring the wrong thing. The authors contend that evaluation should reflect what practitioners need from situation reports, rather than relying on convenient comparison methods. They derive a 15-slot information schema from professional sitreps, using LLM-assisted content analysis that was approved by the original authors.

The paper's title captures its central warning: 'Looking Complete Is Not Being Grounded.' An LLM-generated sitrep may appear thorough and well-structured, but that does not mean it is grounded in the actual facts on the ground. The proposed schema is intended to help evaluators check for grounding in the specific information practitioners rely on.

Because only the abstract is available, this digest cannot compare the approach with other evaluation methods in detail. The paper's contribution, as described, is to shift the focus of evaluation from surface-level completeness to the substantive information needs of crisis response.