State-Court Pleading Coverage in Training Data
How far does the dominance of federal practice in written legal material bias AI systems toward federal standards in state-court work, and how can that bias be measured and corrected?
Why it matters
State courts apply their own pleading rules, but federal pleading law is the most heavily published, indexed and discussed body of pleading law in the country. A system that learns from what is written learns plausibility first. The result is forum mismatch: a California demurrer arguing that the complaint fails to state a "plausible" claim and citing Iqbal, when the demurrer actually asks whether the complaint states facts sufficient to constitute a cause of action. Every citation can be real and good law and still wrong for the room.
State of the art
We are not aware of a public measurement of the federal skew in legal training corpora, or of its effect on model output state by state. The target also moves. Washington, Arizona, Tennessee and Minnesota declined to adopt plausibility pleading; Colorado and Massachusetts adopted it; California and Illinois are fact-pleading jurisdictions; and a state court of last resort can have the question under argument while a brief is being drafted. Practitioners compensate with forum-first prompts and lists of phrases that should never appear in a state filing.
Our partial answers
The posture mismatch taxonomy names forum mismatch as its own category and supplies a pre-filing test: state the governing standard only from authority applying it in the forum. The Docket Test lets a firm measure the problem on its own court's decided motions, scoring posture mismatches as a separate line. Both are diagnostics, not corrections.
Open: a per-state benchmark of pleading and summary judgment standards; measurement of how output quality varies with a state's share of the corpus; and representations that record, for every holding, which sovereign's rule it applied, so that forum can be filtered rather than inferred.