Section 101 Eligibility Starts With the Disclosure Record

Eligibility is decided on facts the inventor had to surface

The two-step framework from Mayo Collaborative Services v. Prometheus Laboratories, Inc., 566 U.S. 66 (2012), and Alice Corp. v. CLS Bank International, 573 U.S. 208 (2014), is now twelve years old. The doctrine is not what most software patents lose on. The record is.

In Berkheimer v. HP Inc., 881 F.3d 1360 (Fed. Cir. 2018), the Federal Circuit held that whether a claim element or combination is “well-understood, routine, and conventional” to a person of ordinary skill is a question of fact. That holding moved Step 2 of the Alice/Mayo analysis out of the realm of pure law and into the realm of what the prosecution and litigation record actually contains.

A specification that describes specific technical improvements creates a factual dispute. A specification that does not, does not. Aatrix Software, Inc. v. Green Shades Software, Inc., 882 F.3d 1121 (Fed. Cir. 2018), extended the same logic to motions to dismiss: plausible factual allegations drawn from the specification can defeat a Rule 12(b)(6) eligibility challenge before discovery.

Drafting cannot manufacture facts the inventor never disclosed. Litigation cannot manufacture them either. The window where the technical record is built is the disclosure window.

What survives Step 1 outright

The cleanest § 101 wins do not happen in Step 2. They happen in Step 1, where the Federal Circuit has consistently held that software claims directed to specific improvements in computer functionality are not “directed to” an abstract idea at all.

In Enfish, LLC v. Microsoft Corp., 822 F.3d 1327 (Fed. Cir. 2016), the court rejected the view that all software claims are inherently abstract. A self-referential database structure that improved how computers operate was eligible at Step 1. McRO, Inc. v. Bandai Namco Games America Inc., 837 F.3d 1299 (Fed. Cir. 2016), reached the same result for specific rules embodying a concrete technical mechanism.

The pattern across both cases: a specific, articulable technical improvement, described in specific technical terms, supported by specific examples. The court can identify what is improved and how the claimed system operates differently from prior approaches.

That description does not arrive at the patent attorney’s desk by default. It is a function of how the inventor’s disclosure was structured.

What the 2024 USPTO update reinforced

The USPTO’s 2024 Guidance Update on Patent Subject Matter Eligibility, Including on Artificial Intelligence, 89 Fed. Reg. 58128 (July 17, 2024), did not modify the Alice/Mayo framework. It reinforced what makes claims survive Step 2A Prong Two, the “practical application” inquiry the USPTO formalized in its 2019 Revised Patent Subject Matter Eligibility Guidance.

The Update added three AI-focused examples (Examples 47, 48, and 49). Across all three, the analytical move is the same: claims that recite a specific technical improvement to computer functionality survive; claims that recite AI-flavored functional language without identifying the technical mechanism do not. Anomaly detection by neural networks is eligible when the claim ties detection to a specific architectural improvement. Speech separation is eligible when the claim ties separation to specific real-time application constraints. Personalized treatment is eligible when the claim recites a particular treatment, not a generic “AI recommendation.”

The lesson is consistent with Enfish and McRO. The factual support for the technical-improvement story has to be in the specification, and therefore has to have been in the disclosure that produced the specification.

The trap counsel should avoid

A persistent error in § 101 argument is to treat Step 1 as if it were a novelty inquiry. “Our claim is novel, therefore it is not abstract” is not the analysis the courts apply.

In Two-Way Media Ltd. v. Comcast Cable Communications, LLC, 874 F.3d 1329 (Fed. Cir. 2017), the Federal Circuit reiterated that eligibility and novelty are separate inquiries. SAP America, Inc. v. InvestPic, LLC, 898 F.3d 1161 (Fed. Cir. 2018), pushed further: no matter how groundbreaking the advance, an abstract idea remains ineligible.

The corollary for disclosure intake is operational. What rebuts § 101 is not “we are novel.” What rebuts § 101 is “this is a specific, concrete technical improvement, and here is the factual record of how it operates differently from conventional approaches.” That is a Berkheimer and Enfish record. It is not the record §§ 102 and 103 ask for.

The disclosure has to be built for the right fight. An IDF that describes the invention as “novel” without describing the technical mechanism creates a § 102 record while leaving the § 101 record empty.

What the disclosure record needs to capture

For software and AI inventions where § 101 will be a live question, the disclosure should preserve five categories of fact.

• The technical problem in the prior art. Specific, not “users wanted a better experience.” A constraint that conventional systems failed to address (latency, bandwidth, memory, accuracy, robustness, security), described in concrete terms.
• The specific improvement to computer functionality. What the system does differently. Not “uses AI to improve accuracy.” The architectural choice, the ordering of operations, the data structure, the constraint resolution that produced the improvement.
• Alternatives considered and rejected. Engineers usually evaluate two or three approaches before settling on one. Capturing those alternatives, and why they were rejected, directly rebuts a future “well-understood, routine, conventional” finding under Berkheimer.
• Concrete metrics where available. Latency reductions, accuracy gains, throughput improvements. Numbers anchor the technical-improvement narrative in measurable reality.
• Architectural specifics. Non-generic data structures, particular orderings, specific operating regimes, particular machines. Enfish’s self-referential table is the canonical example. The specificity is the eligibility lever.

The standard IDF asks for “a description of the invention.” That prompt does not produce this record. It produces a paragraph an examiner can read in two minutes and an opposing expert can dismantle in five.

The cost of waiting until drafting

There is a recurring failure mode in software patent prosecution: the application is drafted from a thin disclosure, an examiner issues a § 101 rejection, and counsel scrambles to backfill technical detail in an interview or a § 132 declaration. That backfill comes from interviews with the inventor, interviews that are now happening months or years after the work, against the constraint that nothing new can be added without violating the new-matter rule.

The cleanest wins in § 101 prosecution come from applications drafted from disclosures that already contain the Berkheimer facts. The cleanest losses come from disclosures that did not.

Examiner-by-examiner outcome variance compounds the risk. In art units that draw heavy § 101 scrutiny, a thin specification is a coin flip. A specification that anticipated the inquiry can defend itself.

Disclosure-stage workflow that produces eligibility-ready records

A disclosure intake that supports eligibility under the current framework looks different from one that supports only novelty.

It forces the inventor to articulate the prior-art problem before solution work begins, where possible. It captures the technical mechanism at sentence-level specificity, not at the level of “the system uses machine learning.” It preserves the alternatives the team considered and the engineering reasons one approach was chosen over the others. It keeps the artifacts that contain the technical record (design docs, architecture reviews, performance analyses) linked to the disclosure rather than detached from it.

This is not a doctrinal innovation. It is the same evidentiary discipline a strong specification has always required. The change since Berkheimer is that the specification is now read as a fact-finding instrument, not just a teaching one. The disclosure feeding that specification has to be built for that role.

What to read next

The doctrinal anchors for this fight are stable. Alice and Mayo set the framework. Berkheimer and Aatrix made Step 2 fact-bound. Enfish and McRO identified what wins at Step 1. The 2019 USPTO guidance and the 2024 Update operationalized the framework for examiners.

For counsel revisiting disclosure intake against this framework, the structural companion is our note on attorney-led invention discovery, the intake mode that produces Berkheimer-ready specifications by reading the engineering record directly.

ALID surfaces invention candidates from engineering artifacts and packages each one with the technical specifics (problem, mechanism, alternatives, source citations) that a § 101-defensible specification needs. To see how the discovery and disclosure flow operates, read how ALID works, or request access to run a first discovery against your own engineering data.