Patent Claims Turn a Technical Description into an Exclusion Test

Embeddings.

Semantic search.

A transformer model in a black turtleneck explaining that two devices share a latent representation of hydraulic aspiration.

Somebody will hand you a rainbow similarity score between an industrial pump and a countertop espresso machine, because both contain a valve and have disappointed a person in a lab coat.

Then the claim arrives.

It isn’t interested in your similarity score.

It has requirements.

It would like them all accounted for, in order, and it has the patient administrative disposition of a man sent to collect receipts from a marina.

A patent application is built to tell a technical story.

Problem, components, steps, alternatives, drawings, implementation detail, proposed claims.

The PCT’s minimum filing bundle makes the division visible in five objects.

A request, a description, the claims, drawings where they’re required, and an abstract.

The description has room for the story.

The claim removes selected parts of that story from the buffet, puts them in a combination, and later asks whether the combination is present.

Same document.

Different machine.

The specification has a basement

Your technical record can sit anywhere on a four-level range.

Bare concept.

One working embodiment.

A core embodiment plus variants.

Or a structured platform disclosure with alternatives, interfaces, definitions and data.

The higher settings give you more material for later claims and narrower fallbacks.

They also make a published application a more elaborate document to leave on a train.

“More words” is not the setting.

Your working unit is the claim-relevant feature.

What it is, how it connects, how it operates, its alternatives, its limits, and evidence that it can be used.

A sensor system may need its signal source, sampling location, calibration, threshold logic, output action, error behaviour and variants.

That’s a technical account.

It can hold alternative ways to do a thing, historical dead ends, a drawing crowded with arrows, and a delicately euphemistic paragraph about the prototype that caught fire near the loading dock.

Claim architecture has three dimensions.

Number of independent claims, total claim count, dependency structure.

In US utility practice, the ordinary fee threshold covers up to 3 independent claims and 20 total claims.

Claims above either threshold bring excess-claim fees.

That’s an administrative threshold.

Not a law of creative writing, and not a reason to make your last claim carry a patio heater on its back.

The calendar has its own bad habits.

You get four non-substitutable clocks.

Public disclosure.

Priority.

Publication.

Prosecution response.

A priority-bearing PCT route commonly puts 12 months, about 18 months, and around 30 months in different parts of your file.

Time to file the PCT after a first filing.

International publication from priority.

National-phase preparation.

None of that changes what your claims require.

It does explain why a drafter wants a disclosure matrix before the claims start wandering off to offices in several languages.

A claim chooses the test

Imagine the published application for the Mabel N. Vickers Municipal Pigeon-Alert Apparatus.

Its description speaks warmly of a sensor, a controller, a transmitter, mounting brackets, calibration variants, an emergency chirp, and a battery enclosure shaped like a very small municipal building.

There may be several embodiments.

There may be a drawing that gives the pigeon an expression of administrative concern.

Your independent claim can select a much smaller arrangement:

a sensor; a controller receiving the sensor output; and a transmitter sending a derived alert.

Three limitations.

That’s the whole proposition.

It isn’t met because some later device has a sensor and can send messages.

“Receiving” and “derived” are relationships.

If their meaning is disputed, construction is the question before evidence is the question.

Then the dependent claim does the thing that makes otherwise reasonable people read only the last sentence.

It says, in effect: claim 1, further including a calibrated threshold and the absurdly sturdy bracket.

That is not a claim for a calibrated threshold.

It inherits the sensor, the controller-receiving relationship, and the transmitter-derived-alert relationship.

Then it adds its own limitation.

At three claim levels, your bottom claim carries both preceding packages plus its addition.

The clause at the end is not a menu item. It is a backpack.

You can also have two or more independent claim paths on the same technical subject.

They aren’t interchangeable because they live in the same patent and attend the same Christmas party.

Each supplies a separate limitations list.

Each is presumed valid independently under US law.

A conclusion about claim 1 doesn’t mechanically settle claim 2.

Even when claim 2 has a very convincing drawing of the same pigeon.

Transitional wording decides how additions get treated.

“Comprising” is open.

An added feature doesn’t by itself remove an item that has every recited limitation.

“Consisting of” normally excludes unrecited elements for the relevant recitation.

“Consisting essentially of” asks the more tiresome question of whether an addition materially affects the stated basic and novel characteristics.

The Gillette example in USPTO guidance involved a safety razor with a guard, a cap, and first, second and third blades.

“Comprising” still reached a razor with more than 3 blades.

Ordinals are not always a fence.

Read it like a reluctant accountant

Your routine starts with an application.

Not a product brochure, and not an image search conducted after midnight.

Download one published application.

Identify one independent claim and one dependent claim.

Copy the independent claim out into separate limitations.

Expand the dependent claim by carrying in every parent limitation before you add its new language.

Then locate the description or figure passage that supplies technical support for each limitation.

Use a drafting-support worksheet first.

claim form  | complete limitation | application support | source location
independent | sensor              | sensor description  | paragraph / figure
independent | controller receives sensor output | control sequence | paragraph / figure
independent | transmitter sends derived alert | alert sequence | paragraph / figure
dependent   | all parent limitations + calibrated threshold | threshold variant | paragraph / figure

Your source location is not decoration.

A disclosure matrix exists to show feature-by-feature support in the submitted specification and figures.

Where the application offers alternatives, record the alternative actually linked to the limitation.

Not “see generally the whole document,” which is the patent equivalent of leaving a forwarding address at a landfill.

This support pass doesn’t turn your description into the scope test.

It prevents the opposite failure.

Treating a claim as a magical sentence that needs no technical material behind it.

Claims can be adjusted later.

Missing factual teaching is much harder to supply without changing the subject.

The description is the warehouse. The claim is the pick list.

Confuse them and you get either a warehouse-sized scope statement or a pick list with no inventory.

A later comparison is a different sheet

For an asserted claim and an accused activity, your comparison needs three working columns.

Full limitation.

Alleged counterpart in the activity.

Record support for the match.

The drafting-support worksheet tells you where the application describes the limitation.

The comparison chart asks whether the later activity has it.

They may share a row label.

They don’t share a job.

Literal comparison asks whether every required limitation is present as construed.

In the United States, the statutory baseline in 35 U.S.C. 271 concerns unauthorized making, using, offering to sell, selling, or importing a patented invention during the patent term.

Your chart is not a finding under that provision.

It’s a way of making missing and asserted limitations equally visible.

Which is a service to everybody except the person who hoped “it is basically the same thing” would carry the afternoon.

Construction comes before any scope conclusion when a term is disputed.

Markman assigns construction to the court in US federal litigation.

Under Phillips, your three intrinsic sources are the claim text, the specification, and the prosecution history.

USPTO examination uses broadest reasonable interpretation consistent with the specification, for pending claims.

A PTAB AIA trial uses the court-style Phillips approach, under the 2018 rule, for petitions filed on or after 13 November 2018.

The forum is not a footnote.

It changes the tool on the bench.

Then the second route arrives, standing in your doorway with a clipboard.

A literal difference asks whether a required limitation is absent under the relevant construction.

A substitution difference asks whether the replacement could be alleged to be equivalent.

Warner-Jenkinson requires that inquiry one element at a time, and doesn’t permit it to erase an element.

Festo adds that a narrowing amendment can invoke prosecution-history estoppel without necessarily eliminating every equivalent to the amended element.

“Not literal” is not a conclusion about equivalence.

It’s the end of one column, not the end of your file.

The chart’s limits

Call it the limitation chart.

It doesn’t decide infringement, or validity, or whether the pigeon apparatus reaches a product with a suspiciously similar chirp.

It separates technical narration, claim language, inherited limitations, support passages, construction questions, literal absence, and equivalence risk.

By the time you’ve opened the pages and filled the rows, the grand semantic resemblance has usually become a stack of smaller questions.

All of them with names, and places to sit.