SovGuardAI returns a structured analysis of one submitted document. It is not a court filing, not expert testimony, and not a finding about any person. The home-page sample is a composite: fictional court, fictional caption, built so the layout is visible without using a real docket. View the rendered sample.
Correspondence score (0–100)
How closely the language and structure match patterns already labeled in research and in U.S. court decisions. Tiers: None, Low, Moderate, High. A score is an indicator, not a measurement, and the same document can move within a range across runs. Methodology.
Tactics detected
Named tactics from a researcher-maintained taxonomy (strawman, UCC scripts, HJR-192, false liens, fee schedules, and others). Each tactic is supposed to be anchored to language in the document. See tactics.
Flagged phrases
Short excerpts as they appear in the filing — “accepted for value,” ALL-CAPS dual names, UCC 1-308 stamps — with a page/paragraph location when it can be verified. See phrases.
Court-rejection language
Citations come from a fixed library, not from the model inventing reporter cites at analysis time. You still have to verify them. The library includes published circuit language such as United States v. Benabe, 654 F.3d 753, 767 (7th Cir. 2011), and United States v. Schneider, 910 F.2d 1569, 1570 (7th Cir. 1990).
60-second brief
A short narrative of what the document is trying to do and how similar theories have been treated. Useful for triage. Not a substitute for reading the filing.
Clerks: first-pass triage. Do not attach a report to the docket as a finding.