One real run

What the output actually looks like

A complete run against a public specification, captured on 2026-08-18 and printed here unedited. Ten claims drafted, and the tool's own critique flagged all ten: you start from the problems, not a blank page.

This is one run, on one model, against one small public specification. Re-running the same command produces different claims, because the drafter is a language model.

Two of the ten drafted claims, and what its critique said

Shown in full, exactly as produced. The critique runs against the same specification and reports what it cannot find support for.

Drafted claim 2

2. A manual override mechanism for a temperature-actuated valve, comprising: a rotatable knob threadably engaged with a valve body; and an override plunger operatively connected to the knob and configured to mechanically hold the valve element in either an open position or a closed position independent of a state of the shape-memory alloy spring.

UNSUPPORTED: 'manual override mechanism' is not described in the parent specification - the spec mentions a manual override mechanism but does not describe its components or structure

ANTECEDENT BASIS: 'the valve body' lacks antecedent basis - introduce it or change to 'a valve body'

Drafted claim 3

3. A temperature-actuated valve assembly for industrial process control applications, comprising: a valve body defining an inlet port, an outlet port, and a central flow passage; a valve seat disposed within the central flow passage; a valve element movably positioned to selectively engage the valve seat; a shape-memory alloy spring coupled to the valve element and configured with a transition temperature in the range of 40 to 80 degrees Celsius; a bias spring urging the valve element toward an open position when the shape-memory alloy spring is below the transition temperature; and a thermal mass made of copper or aluminum positioned adjacent the central flow passage and in thermal communication with the shape-memory alloy spring.

RESTATEMENT: This claim merely restates the parent claims 1-3 without pursuing different scope

MOSAIC: The combination of elements from parent claims 1, 2, and 3 is not disclosed as a single invention - the spec does not present a valve assembly with all these features together in one embodiment

Ten drafted, ten flagged, and that is the point

Every claim this run produced was flagged by the same tool that drafted it. The second claim above recites “the valve body” without introducing it; the third restates the parent rather than pursuing different scope. Ordinary drafting defects, found before you spend time on them, which is the work this tool does.

The input

The first claim of the parent, as filed, from the committed walkthrough fixture:

1. A temperature-actuated valve assembly comprising: a valve body defining an inlet port, an outlet port, and a central flow passage; a valve seat disposed within the central flow passage; a valve element movably positioned to selectively engage the valve seat; a shape-memory alloy spring coupled to the valve element, the shape-memory alloy spring having a transition temperature, wherein the shape-memory alloy spring has a first length below the transition temperature and a second length less than the first length above the transition temperature; and a bias spring urging the valve element toward an open position when the shape-memory alloy spring is below the transition temperature.

One command supplies both the specification and the parent claims, so it needs no files of your own:

continuation-drafter draft --demo --revise-loops 1

What produced this

Model: qwen3-coder:30b, running locally. Nothing left the machine.
Input: fixtures/micro (the walkthrough the published lessons use)
Captured: 2026-08-18
Claims drafted: 10
Claims flagged by the critique pass: 10

This model sits in the middle of the measured range; the benchmark shows how the others compare. To run it yourself, see choosing a model for your machine or the download.

The same kind of run, in the app

What it looks like on a real application

The desktop app on a granted public application, its specification and allowed claims from the USPTO's data, and its Patent Center file history attached. Every state below is one the app produced; nothing was arranged by hand.

First, the directions, before anything is drafted

Each direction is marked as found in the specification or as a drafting strategy. You remove the ones not worth drafting, and only the ones you keep are drafted.

Continuation Drafter (now Patent Practitioner Tools), its first step on 18/541,216, run locally with qwen3.5:9b: the 8 directions it proposed for a continuation before drafting anything, each marked as found in the specification or as a strategy, for the practitioner to keep, edit or remove. — Unedited screenshot of Continuation Drafter (now Patent Practitioner Tools) v0.5.0, captured 2026-09-30: application 18/541,216, public record, run locally with qwen3.5:9b.

Then the drafted claims, with what the tool flagged in each

The whole page after drafting: the claims with the defects the tool's own critique pass raised under each, and beside them the cost of the matter so far, the claim list, the directions, and the papers it read from the file history. Scroll inside the picture to see all of it.

Continuation Drafter (now Patent Practitioner Tools), the whole Draft tab after a run on 18/541,216, run locally with qwen3.5:9b: the 8 directions it drafted toward, then 8 drafted claims, 6 of them with the defects the tool's own critique pass flagged beneath them. Beside them: next steps, what the matter has cost, the claim list, the continuation targets, and the prosecution record read from 58 papers of the Patent Center file history. — Unedited screenshot of Continuation Drafter (now Patent Practitioner Tools) v0.5.0, captured 2026-09-30: application 18/541,216, public record, run locally with qwen3.5:9b.

A second real run

What it reads from a file history

The record reader over the Non-Final Rejection and the Notice of Allowance of application 18/541,216, a public, granted Patent Center file, captured on 2026-09-30 on qwen3.5:9b running locally and printed here unedited.

Every quotation is checked against the paper's text. Verified means the words are in the paper as quoted; not verified means they are not (usually a quotation shortened with an ellipsis, or a misread scan), so you know exactly which ones to check. It reports what the paper says, never whether a rejection was right or an amendment needed.

The rejections, as the reader reported them

Each rejection in the Office Action of 18/541,216, as reported by the record reader
Claims	Ground	References	Quotation
19-20	35 U.S.C. 101	none	Claims 19-20 are rejected under 35 USC 101 because the claimed invention is directed to non-statutory subject matter... The specification is silent to limit that such medium exclude signals. Therefore covers transitory propagating signal per se (non-statutory subject matter). See In re Nuijten... (not verified)
1-12 and 19-20	nonstatutory double patenting	U.S. Patent No. 11,847,178	Claims 1-12 and 19-20 are rejected on the ground of nonstatutory double patenting as being unpatentable over the claims 1-12 and 19-20 of prior U.S. Patent No. 11,847,178... they are not patentably distinct from each other because they are directed to substantially the same invention. (not verified)
13-18	statutory double patenting	U.S. Patent No. 11,847,178	Claims 13-18 are rejected under 35 U.S.C. 101 as claiming the same invention as that of claims 13-18 of prior U.S. Patent No. 11,847,178. (verified)

The examiner's reasons for allowance

After further searches and consideration the Examiner concluded that the claimed invention has been recited in such a manner that the independent claim is not thought by any prior art of record. Independent claims 1, 13 and 19 are considered allowable since when reading the claim in light of the specification, none of the prior art of record either alone or in combination fairly disclose or suggest the combination of limitations specified in the independent claims

Verified against the paper. The limitations the reasons credit, quoted by the reader:

generating, for each concept clip of a set of concept clips, an embedding, wherein the set of concept clips corresponds to a concept searched by a user; generating a concept embedding based on the embeddings of the set of concept clips; determining a semantic relationship between the concept embedding and one or more document embeddings that each correspond to a document clip of a set of documents; and based on the determined semantic relationship, providing, for display at a client device, an indication of the semantic relationship between the concept and one or more document clips of the one or more document embeddings.

Verified.

generate, for eachconcept clip of a set of concept clips that correspond to a concept searched by a user using a machine learning model, an embedding; generate a concept embedding based on the embeddings of the set of concept clips; determine a semantic relationship between the concept embedding and one or more document embeddings that each correspond to a document clip of a set of documents; and generate a graphical user interface based on the determined semantic relationship depicting the semantic relationship between the concept and one or more document clips of the one or more document embeddings.

Verified.

generating, for each concept clip of a set of concept clips, an embedding, wherein the set of concept clips corresponds to a concept searched by a user; generating a concept embedding based on the embeddings of the set of concept clips; determining a semantic relationship betweenthe concept embedding and one or more document embeddings that each correspond to a document clip of a set of documents; and based on the determined semantic relationship, providing, for display at a client device, an indication of the semantic relationship between the concept and one or more document clips of the one or more document embeddings.

Not verified: compare with the paper.

How well the reader does across more papers was measured, on this file and one other.