Motions are lost on what is missing, not on what is present.

Every instrument that checks a draft grades its contents: whether each citation exists, whether it says what the draft says, whether it is still good law. An omission is not a property of the contents. It is a relation between the draft and a specification of what the draft should have contained, and a checker that receives only the draft cannot compute it.

Four omission classes do the damage. Unretrieved authority, including controlling authority adverse to the position taken. An unaddressed element, the most dangerous class, because no question was ever posed, so no retrieval or verification ever ran. An uncited record fact, which at summary judgment the court may disregard, since it need consider only the cited materials. And negative space: prior decisions rejecting the very inference the draft makes, which a search for supporting authority will not find.

Dispositive motions make completeness computable. The governing law supplies an element list that can be enumerated in advance, is conjunctive, and decomposes into cells: for each element, the standard, the controlling authority, the record support, the adverse authority, and the courts that have rejected the argument. Each cell must end in a citation or in a logged search that found nothing. A blank cell is not an empty result. It is a question nobody asked.

This also changes how legal AI should be evaluated. A decided motion, with its opposition and order, shows what a complete brief needed. Score what the tool missed, not only what it got right, and ask it cell-level questions rather than document-level ones, so that what it skipped becomes visible.