The vocabulary of patentability, in plain English

Patentability requirements

Prior art

The body of public knowledge that predates your claimed invention: granted patents, published applications, journal papers, product manuals, conference talks, or any other disclosure available to the public before the relevant date. Novelty and non-obviousness are both judged against prior art, so the quality of a patentability opinion is bounded by the quality of the search behind it. In Patendra: prior art is retrieved from four sources (USPTO, Espacenet, Google Patents BigQuery, and a local vector store), and only retrieved, ID-verified references may support a rejection.

Anticipation (§102)

A claim is anticipated when a single prior-art reference discloses every limitation of that claim, arranged as in the claim. Anticipation is the legal failure of novelty under 35 U.S.C. §102: if one document already teaches the whole invention, there is nothing new to patent. It takes only one such reference, but it must cover every element. In Patendra: the simulated examination asserts a §102 rejection only when one reference maps onto every limitation, and the reference is cited by its verified ID.

Obviousness (§103)

Even a technically novel claim fails under 35 U.S.C. §103 if the differences between it and the prior art would have been obvious to a person of ordinary skill in the field. Examiners typically show this by combining two or more references and articulating a motivation to combine them. Obviousness is where most real-world rejections live, because it doesn't require any single reference to teach everything. In Patendra: §103 analysis builds explicit reference combinations with an articulated motivation, mirroring how examiners actually write these rejections.

Enablement

The §112(a) requirement that the specification teach a person of ordinary skill how to make and use the full scope of the claimed invention without undue experimentation. A claim can be novel, non-obvious, and still fail because the document doesn't actually teach how to practice it, a common fate for claims broader than what was described. In Patendra: the drafted specification is written against an enablement checklist, and enablement is one of the four deterministic scoring signals.

Written description

A distinct §112(a) requirement: the specification must show that the inventor actually possessed the claimed invention at filing, not merely that a skilled reader could reconstruct it. Enablement asks "does it teach how?"; written description asks "did they really have it?". Claims amended during prosecution to cover things the original filing never described fail here. In Patendra: written-description checks run against the specification as part of the simulated §112 examination.

Definiteness (§112(b))

Claims must "particularly point out and distinctly claim" the invention: a reader of ordinary skill must be able to tell, with reasonable certainty, what infringes and what doesn't. Vague relative terms, missing antecedent basis, and internally inconsistent elements all breed indefiniteness. In Patendra: definiteness problems are caught twice, by free structural validation in Tier 0 of claim refinement and by §112(b) checks in the simulated examination.

Claims and the specification

Claim (independent and dependent)

The numbered sentences at the end of a patent that define the legal boundary of the invention. Everything else in the document is explanation. An independent claim stands alone; a dependent claim references an earlier claim and narrows it ("The system of claim 1, wherein…"), inheriting every limitation of its parent. Dependent claims act as fallback positions if the broad claim falls. In Patendra: claim scope is treated as a search problem (broaden what survives examination, narrow what gets rejected), with per-claim scores in the output.

Claim limitation (element)

A single requirement within a claim: one component, step, or property that an accused product must have (for infringement) or a reference must disclose (for anticipation). Every added limitation narrows the claim: easier to get allowed, easier to design around. In Patendra: §102 analysis is an element-by-element mapping of each limitation against the closest retrieved references, so you can see exactly which limitation saved the claim.

Antecedent basis

The drafting rule that any claim element introduced with "the" or "said" must refer back to an element previously introduced with "a" or "an". Writing "the sensor" when no sensor was ever introduced leaves the reference ambiguous and makes the claim indefinite under §112(b). It is the most mechanical and the most common drafting defect. In Patendra: antecedent-basis checking is deterministic and free, run in Tier 0 before any tokens are spent refining a structurally broken claim.

Specification

The written body of the patent application: the field, background, summary, detailed description, and figure descriptions that support the claims. The specification is what must satisfy enablement and written description, and the claims can only reach as far as it teaches. In Patendra: the workflow drafts a full specification grounded in the claims that survived examination, with the background written as a prosecution argument. See patent drafting.

Embodiment

A concrete implementation of the invention described in the specification: one specific way the claimed idea can be built or performed. Describing multiple embodiments supports broader claims by showing the invention isn't limited to one form, and gives fallback support if claims must be narrowed later. In Patendra: drafted specifications include multiple embodiments precisely to widen §112 support for the optimized claim set.

Examination and prosecution

Examiner

The patent-office official who searches prior art, applies the statutes to your claims, and decides whether the application can be allowed. Examiners work within a technology area, follow office procedure, and communicate through office actions. Prosecution is, in essence, a written negotiation with this person. In Patendra: the examination phase simulates this role (search, then per-statute analysis) so you see the likely objections before a real examiner does.

Office action

An official written communication from the examiner, typically listing rejections and objections with the reasoning and cited references for each. Most applications receive at least one; the applicant responds by amending claims, arguing, or both. In Patendra: the simulated examination memo is structured like an office action, with per-statute findings and ID-verified citations, so the format is familiar to counsel from day one.

Rejection vs. objection

A rejection attacks the substance of a claim: it is unpatentable under §102, §103, or §112. An objection attacks form: informal drawings, a misdescriptive title, spec formalities. Rejections are the serious ones; they must be overcome by argument or amendment, while objections are usually fixed clerically. In Patendra: the workflow concentrates on rejections, since those are what determine whether the invention is patentable at all.

File wrapper (prosecution history)

The complete public record of an application's examination: every office action, response, amendment, and argument. Courts read it when interpreting claims, and statements made in it can narrow what you can later assert, so what you argue during prosecution matters long after grant. In Patendra: every run preserves its full evidence trail (hypotheses, search logs, examination memos, score tables) as a pre-filing analog of the file wrapper, stored locally.

Search, filing routes, and strategy

Novelty search (patentability search)

A prior-art search performed before filing to estimate whether an invention is new and non-obvious. It is cheaper to learn the answer now than after paying for drafting and filing. A novelty search asks "can this be patented?", which is a different question from whether you can sell it (see freedom-to-operate). In Patendra: this is the core retrieval phase: hybrid search across four sources, fused and reranked, feeding the simulated examination.

Freedom-to-operate (FTO)

An analysis of whether making or selling a product would infringe someone else's in-force claims in a given country. FTO and patentability are frequently confused: a product can be unpatentable yet free to sell, or brilliantly patentable yet infringing. FTO turns on live claims, expiry dates, and jurisdictions, making it a distinct legal exercise. In Patendra: Patendra assesses patentability, not FTO. Do not treat any Patendra output as clearance to launch a product; FTO requires counsel.

CPC classification

The Cooperative Patent Classification, a hierarchical scheme run jointly by the USPTO and EPO that sorts patent documents into fine-grained technology classes (for example, G06N for machine-learning arrangements). Searching by class finds art that shares technology even when it shares no keywords. In Patendra: classification signals complement keyword and semantic retrieval in building diverse search strategies across the patent-data backends.

Provisional application

A lower-cost U.S. filing that establishes a priority date without claims or examination. It lasts 12 months, within which a full non-provisional application must be filed to keep the date. A thin provisional is a trap: it only protects what it actually describes in enabling detail. In Patendra: a run's drafted specification and evidence package gives a provisional real substance, reviewed by an attorney before filing, as always.

PCT (Patent Cooperation Treaty)

An international treaty that lets one application preserve filing rights in over 150 member countries, deferring the expensive per-country decisions by up to 30 to 31 months from priority. It doesn't grant a "world patent" (eventually you must enter national phases), but it buys time and an international search report. In Patendra: prior art doesn't respect borders, which is why retrieval includes Espacenet's worldwide coverage rather than U.S. sources alone.

Retrieval technology

Vector (semantic) search

Search that represents text as numeric vectors, called embeddings, so that documents about the same concept land near each other regardless of wording. It finds a "heat-dissipating lattice" when you searched for a "cooling mesh", which keyword search cannot. Its weakness is the mirror image: exact terms of art can matter, so it works best paired with keyword search. In Patendra: a local ChromaDB vector store provides semantic retrieval on your machine, built with a sentence-transformer model downloaded on first run.

Reciprocal Rank Fusion (RRF)

A method for merging ranked result lists from multiple search engines: each document scores the sum of 1/(k + rank) across every list it appears in, so items that rank decently in several independent systems beat items that top just one. It needs no score normalization between engines, which makes it robust when the engines are as different as a keyword API and a vector store. In Patendra: RRF fuses results from USPTO, Espacenet, BigQuery, and the local vector store into one candidate list for examination.

Cross-encoder reranking

A second, more expensive relevance pass: instead of comparing precomputed embeddings, a cross-encoder reads the query and each candidate document together and scores the pair directly. It's too slow to search millions of documents but excellent at reordering a shortlist, catching relevance that independent embeddings miss. In Patendra: fused candidates can be cross-encoder reranked so the simulated examiner sees the strongest references first.