What a real claim chart looks like in 2026

Patent Radar produces claim charts in exactly the format that an IP litigator would expect from a paid associate. The format predates the LLM era by decades. The technology is new, the shape is not.

Two-column row, optional third

Left cell: the claim limitation, verbatim, with its limitation ID in square brackets. Right cell: one to three stacked evidence cards, each with a quoted snippet, source URL with crawl date, and a read-strength pill. Optional analyst-comment third column for when counsel wants to override the agent.

Long limitations are auto-split into phrases so the evidence maps to the actual language that matters in litigation, not to the broadest paraphrase that fits a slide.

Read strength, never red

Confidence on a Patent Radar chart is a three-tier scale: green for high, amber for medium, slate for low. There is no red tier. Red would read as 'infringement confirmed' which is a legal call, not an algorithm call. Our agent will never assert a legal conclusion.

Provenance per row

Every assertion the model makes carries: model identifier, prompt version, source URL, crawl date, and a SHA-256 hash of the evidence pack. Standardising provenance in the export was Patent Radar's first design call, before we had a Stripe webhook or a sign-in screen.

The Kennedys piece on Rule 702 and AI-derived evidence is the work to read on why. Without an audit trail, AI-generated evidence is difficult to admit. With one, it is closer to a regular work-product memo.

Exportable in the format your counsel uses

Word for the partner who annotates by hand. PDF for the court-ready version with the watermark 'Draft prepared by Patent Radar for legal review only'. CSV for the analyst who builds their own comparison sheet. All three from a single click on the chart.