Pre-Filing Completeness Protocol

The framework

Every citation checker grades the contents of a draft. Motions are lost on the complement. An omission is not a property of what a document contains; it is a relation between the document and a specification of what it should have contained. A tool that receives only the document cannot compute that relation, because absence has no representation in its input.

The protocol supplies the specification. Let D be the draft and S the specification it must satisfy: the claims at issue, their elements under the governing law, the burden on each, and the record.

Four omission classes.

O2 sits upstream of everything: an element nobody asks about produces no query, no citation and no error. O4 differs from O1 in search design: an O1 query asks for authority on a rule; an O4 query asks whether a court has already declined to draw a specific inference.

Why dispositive motions make completeness computable. General research has no completeness denominator. Summary judgment does, because the substantive law supplies an element list that is enumerable in advance, conjunctive (the movant must prevail on every element; the nonmovant needs a genuine dispute on only one), and decomposable into a legal standard and a factual showing per element. The record side is sharper still: Rule 56(c)(1)(A) requires citation to particular record materials, and Rule 56(c)(3) lets the court consider only cited materials. A motion to dismiss uses the same matrix with the pleading in place of the record.

The protocol.

• Step 0, fix the frame. Record the forum, the governing law for each claim, the stage and standard, and the trial burden on each claim. Every posture error described in the Posture Mismatch Taxonomy starts with an unrecorded or incorrect frame.
• Step 1, enumerate the elements. List each element of each claim and affirmative defense from controlling authority in the forum. Tag the burden and the available summary judgment route (Celotex or Adickes in federal court; Aguilar in California). Cross-check against pattern jury instructions, which are not authority but flag discrepancies.
• Step 2, build the matrix. M = elements x {Standard, Controlling authority, Record, Adverse, Rejected}. Each cell ends in exactly one state: Filled (citation with pinpoint), Searched, none found (query, source and date), or Not applicable (one-line reason). No cell may be blank.
• Step 3, authority sweep (O1). Retrieve top-down: court of last resort, then the binding intermediate court, then persuasive authority. Run each through a citator, then confirm the cited proposition does not depend on a different standard, record, burden or forum.
• Step 4, record sweep (O3). List every record location bearing on each contested element, with pinpoints. For the nonmovant, the evidence creating the dispute must be cited, not merely present.
• Step 5, adverse sweep (O1, adverse subclass). For each controlling authority, retrieve decisions that limited it, distinguished it or carved exceptions. Model Rule 3.3(a)(2) requires disclosure of directly adverse controlling authority known to the lawyer; this step is how the lawyer comes to know it.
• Step 6, negative-space sweep (O4). Restate each argument as rule R, applied to fact pattern P, yields conclusion C, and search for it joined with rejection language (rejected, unpersuasive, declined to extend). Search the assigned judge's prior rulings where available. Distinguish each rejection or drop the argument.
• Step 7, terminate and log. Stop when no cell is blank.

Self-check metrics. Element coverage (elements addressed divided by elements listed) must be 1. Blank cells must be 0. Unlogged "searched, none found" cells must be 0. None of these measures persuasion; they measure whether the brief was built against a specification.

Design rule for AI tools. Ask cell-level questions, not document-level ones. A system asked to "draft the motion" gives no signal about the questions it skipped; a system asked to fill one cell either fills it or visibly reports that it could not.

How to apply it

The following is an illustrative hypothetical. A defendant plans a summary judgment motion on a negligence claim in California superior court.

Frame. Forum: California state court. Governing law: California. Stage: summary judgment under Code of Civil Procedure section 437c. Route: the plaintiff bears the trial burden, so under Aguilar the defendant may show the plaintiff cannot reasonably obtain needed evidence, but must do so with evidence.

Elements. Duty, breach, causation, damages, each listed from controlling California authority and cross-checked against the pattern instructions. The draft argues breach and causation only.

Matrix, first pass. An excerpt:

The blanks are findings. Damages is an O2 omission: harmless if the motion deliberately does not reach it, but that choice must be stated as a one-line "not applicable" reason, not left empty. Causation's record cell is an O3 gap: the deposition testimony that undercuts the plaintiff's expert exists in the record but is not yet cited, and a court may disregard uncited material.

Sweeps. The adverse sweep on causation retrieves an appellate decision that limited the leading case to a different fact pattern; the draft now distinguishes it rather than meeting it in the opposition. The negative-space sweep restates the breach argument as a single proposition and finds a trial-court ruling in the same county rejecting it on similar facts. The team records the court's stated reason and narrows the argument.

Termination. Element coverage reaches 1, blank cells reach 0, and every null search carries its query and date. The log is what a supervising partner reviews, and what the firm can produce if a court or carrier later asks how the brief was made.

Known limitations and critiques

It does not prove completeness. Over an open corpus, nothing can. The protocol converts completeness from an unfalsifiable feeling into an inspectable artifact: a frame, an element list, a filled matrix and a dated search log.

A null search is bounded by its source. "Searched, none found" means the named source returned nothing for the logged query. Unpublished orders, which decide many motions, may sit outside the searched corpus.

The specification can itself be wrong. An element list drawn from the wrong authority produces a perfectly filled matrix for the wrong claim. Pattern instructions help, but they are not authority.

Box-ticking risk. Any checklist can be satisfied formally. "Not applicable" cells, in particular, require judgment that the metrics cannot audit.

Negative-space search is hard. Rejections are often a few paragraphs inside opinions about something else and phrased generically; recall for such searches is unmeasured.

Persuasion is out of scope. The protocol makes a brief complete relative to a specification, not convincing. Its value is highest where the element structure is explicit, as at summary judgment, and lower in open-ended research questions.