The §112 problems amendments introduce
Paste a claim set and get a structural read of its references. It finds the antecedent basis errors that appear when claims get edited under deadline — the ones an examiner will find for you in the next Office Action.
Five checks
What it catches
Each check is a separate validator with its own rule. They run independently, so a finding always traces back to one specific reason.
Antecedent basis
Checks that every "the" or "said" reference has an antecedent introduced earlier in the claim set.
…wherein the housing surrounds the motor
No "a housing" was ever introduced.
Pronouns
Flags pronouns — it, its, they, their — which have no place in claim language because their referent is never unambiguous.
…a controller that adjusts its output
Whose output? The controller's, or the motor's?
"Such" phrases
Detects "such X" constructions, which read as references but never actually establish or point to an antecedent.
…configured to receive such signal
"Such signal" refers to nothing definite.
Prepositional scope
Catches terms later referenced with "the" that only ever appeared as the object of a preposition — a real antecedent, but not a properly introduced element.
a valve for a conduit … wherein the conduit is rigid
The conduit was never positively recited.
Reintroduction
Finds terms re-introduced with "a" in a dependent claim when the parent claim already defined them — the classic double-inclusion problem.
Claim 1: a sensor … Claim 3: a sensor mounted to the frame
Claim 3 should say "the sensor".
How it works
The same answer every time
It reads the grammar
Claims are parsed for their actual sentence structure, so it knows where each element begins and ends. That matters — claim language nests far deeper than ordinary prose.
It tracks every term
Elements are followed across the whole claim set, down the dependency chain, so anything introduced in an independent claim counts as introduced for every claim beneath it.
Each rule stands alone
The five checks run separately from one another, so a finding always traces back to one specific reason — and adding a sixth doesn't disturb the others.
Antecedent Basis FAQ
Why not just ask an LLM to check antecedent basis?
Because this is a problem with a correct answer, and a language model will give you a plausible one. Antecedent basis is a structural property of the claim set — which noun phrases were introduced, where, and in what scope. Reading the grammar resolves that directly: the same claims produce the same findings every time, there is nothing to hallucinate, and you can see exactly which rule fired. It also returns instantly.
How accurate is it?
It is a linter, not an examiner. It surfaces candidates for a human to judge, and it will produce false positives — claim language is genuinely ambiguous, and some constructions are deliberate. Treat a finding as "look at this line", not as "this claim is defective".
What does it cost to run?
Nothing beyond your own machine. It uses no AI models, so there is nothing to pay for per run.
Does it handle multiple dependent claims?
It walks the dependency chain, so a term introduced in claim 1 counts as an antecedent in any claim that depends from it. That is what makes the reintroduction check possible — it can tell that a dependent claim is re-reciting something its parent already established.
Comes after the OA Agent
Work out what to amend first, then run the amended claim set through here before the response goes out.
About the OA Agent