The workbench is built around the Rejection Graph — a single structure linking each rejection to the claims, elements, cited art, specification support, and response options it touches. Parse it once; read it from any angle; trace any line back to the record.
Cards, tables, charts, and drafts are all renderings of the same graph. Edit a node — re-segment a claim, correct a misread reference label, accept or reject a mapping — and everything downstream re-derives. There is no second copy of the truth to drift out of sync.
Rejections branch on statutory ground. The workbench runs the right analysis for each and presents findings, never conclusions.
Detects the rejection, summarizes the examiner's eligibility theory, and classifies the asserted judicial exception. Surfaces claim features that may tie to a practical application or technical improvement — as a framework for your argument, never an eligibility verdict.
Separates primary from secondary references, summarizes the motivation-to-combine rationale and flags it when conclusory, and looks for teaching-away and non-analogous-art angles. Identifies dependent-claim limitations that may carry the day.
Flags §112(a) written-description and enablement concerns, §112(b) antecedent-basis and indefiniteness issues, and §112(f) functional terms that may need corresponding structure in the specification.
Each limitation is classified against the cited art — expressly disclosed, arguably disclosed, inferred by the examiner, missing or weak, or unclear — with an explanation and a quoted passage.
Then the quote is checked against the source text. Found verbatim, it's anchored to an exact offset. Paraphrased, it's marked partial. Absent, it's marked unverified and held out of the coverage analysis — so a confident-sounding sentence can't stand in for evidence that isn't there.
Generate and edit amendment candidates with tracked changes against the current language — each one checked against the original disclosure before you rely on it.
Maps each amendment to where the original spec, claims, or drawings support it — and flags when the document that would supply support was never provided.
Flags language that may introduce new matter under §132, plus antecedent-basis breaks and inconsistent terminology the amendment would create.
Surfaces how much claim scope a narrowing amendment surrenders and the prosecution-history estoppel it risks — so allowance is never bought blindly.
Per rejection, response paths are laid out as a comparison: argument strength, scope cost, support strength, estoppel risk, and drafting burden. Argument-only traversals are kept distinct from amendment-based paths. You select a path and add notes; the unselected options stay visible for comparison and flow nowhere you didn't choose.
The draft is organized by rejection, built only from the strategies you selected, with sources carried per section and low-confidence passages marked. Export to .docx with USPTO strikethrough/underline amendment markup, highlighted [ATTORNEY: …] placeholders, and a page-broken “Attorney Work Product — remove before filing” appendix of open issues and sources. The system never files.
Every field the parser produced is editable. Fix a misread reference label and examiner-shorthand matching improves; re-segment a claim and its stale mappings drop and rebuild. Corrections are marked, so you can see what was changed.
Every action — a parse, an edit, an accepted suggestion, a generated draft — is recorded to the matter's history. The record enriches as you work it, and your judgment, once entered, governs.
A small, rigorous Office-action workflow for attorneys who want leverage without losing control.