Drafting Patent Claims and Specifications That Survive Section 112: A Practical Guide

The email comes in late on a Thursday. The founder's company has a demo with a large bank on Monday, and an investor wants a patent application on file first. Attached are eleven slides. Slide 4 says the product "uses AI to detect fraudulent transactions in real time with fewer false positives." The founder's note reads, "Can we claim anything that does this? We want the broadest possible coverage."

That request has been the beginning of a great many invalid patents. The honest answer is that you can file something by Friday, you might even get it allowed, and it may be worth very little at the moment someone with a budget reads it closely. Section 112 is the part of the Patent Act that decides whether what you wrote buys what you claimed, and it is enforced years later by people who are paid to find the gap.

This guide tells you what to do about that, in order. It is organized around a patent's life: drafting, prosecution, and then litigation from each side. For the doctrine and the close reading of the cases, read the companion article, Enablement, Written Description, and Indefiniteness After Amgen v. Sanofi: What Section 112 Demands.

The short answer

• Start by asking what the invention is and why it works. The most valuable hour in a patent's life is the one with the inventors, asking what failed, what surprised them, and which feature explains the result. That is the raw material for an enabling, describing disclosure.
• Draft a ladder of claims, and support every rung. Each rung, from broadest to narrowest, needs its own support in the text. A narrower dependent claim has saved many patents whose broadest claim fell.
• Teach the far edge of the claim. After Amgen Inc. v. Sanofi, 598 U.S. 594 (2023), examples plus "screen to find more" is not enough for a class defined by results. Disclose what the working members share, and show diversity.
• For software, disclose the how. Give inputs, outputs, and the transformation for every claimed function. Avoid "module" and "handler" as nouns doing structural work.
• Define anything measurable, and name the test. Pick a tolerance for "about." If a number depends on how it is measured, say how.
• In prosecution, answer the factors and watch your words. Every statement in a response may be read back to you, and a word added to beat prior art can become the claim's weakest point.
• Challengers must be concrete. Name the embodiment that is not enabled and the species not described, and raise each ground in contentions. Clear and convincing evidence is the burden, and you will need it from a real expert.
• Patent owners should test their constructions against Section 112 before winning them, and should prepare inventors and experts honestly for the questions that have ended recent cases.

Part 1: Before you draft

Step 1: Run an inventor interview that goes after the "why"

Who: the patent attorney or agent and the people who did the work. When: before any claim is drafted. Trade-off: an extra few hours of expensive time now, against a disclosure that otherwise describes the marketing instead of the invention.

Most invention disclosure forms ask what the invention does. Section 112 cares more about these questions:

1. What did you try that failed, and why do you think it failed?
2. Which change made the result appear? Can you explain the mechanism, even tentatively?
3. How many alternatives did you actually test, and how many worked?
4. What do the working versions have in common that the failures lack?
5. How would you measure success? Would two labs or two engineers get the same number?
6. What do you believe will work that you have not tried? Why do you believe that?
7. What part of this could a good practitioner do from the literature, and what could they not?

Answers 3 and 4 go directly to the factors in In re Wands, 858 F.2d 731 (Fed. Cir. 1988), and they determine whether you can use the opening the Supreme Court left in Amgen: a few examples can be enough if the specification discloses "some general quality . . . running through" the class. Answer 6 tells you what can honestly be prophetic. Keep the raw answers: they will shape what you can say later, and a privilege-protected record of them helps counsel.

For the forms and intake meeting, see How to Prepare an Invention Disclosure for Your Patent Attorney and Preparing an Invention Disclosure: A Practical Checklist.

If the inventors cannot say why it works and have only one or two examples, tell the client now. A narrower claim plus a program of further experiments and later filings may be a better plan than a broad claim on a thin record. A later application takes a later date, but a claim the patent cannot support has no date worth having.

Step 2: Decide what the patent is for

Claims should reach what competitors would plausibly do, and not everything the founder can imagine. Before drafting, write one sentence each on:

• The competitor's easiest design-around. If the claim can be avoided by changing one component, you need a broader rung that covers it, with support.
• The product you need to protect. At least one claim must read on it plainly.
• What you are willing to disclose. A specification that enables is a specification that teaches your competitors after the patent issues, and often 18 months after filing when the application publishes. If a technique is better kept secret, decide that now, and make sure you are not claiming it broadly and leaving it out. For the choice between protections, see Legal Protection of Software: Copyrights, Patents, Trade Secrets, and Contracts.

The disclosure is not optional. Best mode remains a statutory duty under 35 U.S.C. § 112(a), even though, under 35 U.S.C. § 282(b)(3)(A), a failure to disclose it no longer invalidates a claim. Candor obligations under 37 C.F.R. Part 1, including § 1.56, apply regardless. See Maintaining the Duty of Candor to Avoid Inequitable Conduct: A Practical Checklist.

Part 2: Drafting the claims and the specification

Step 3: Build a scope ladder

A scope ladder is a planned set of claims from the broadest you can defend down to the embodiment you have built. Think of it as an insurance schedule: when a court knocks out a rung, the patent should still reach something valuable on the next one down.

Here is a hypothetical. Suppose Marlowe Cell, a battery startup, discovers that adding a small fraction of a particular fluorinated salt to a solid electrolyte suppresses dendrite growth, the needle-like metal formations that short out lithium cells. It tested 30 additives. Nine worked. All nine had a fluorinated anion below a certain ionic radius. Here is a ladder:

Note what rung 2 does. It has moved from what the additive achieves to what it is. That shift is usually the main job of post-Amgen drafting.

The drafting work should go in this order: write the claims first (a working set, at least), then draft the specification to support each rung, then check back. Claims drafted last tend to outrun the text. Under 37 C.F.R. § 1.75(d)(1), the terms in the claims must find clear support in the description. Original claims are part of the disclosure, but they do not save a claim that fails to convey possession: Ariad Pharmaceuticals, Inc. v. Eli Lilly & Co., 598 F.3d 1336 (Fed. Cir. 2010) (en banc), and Brita LP v. International Trade Commission, No. 24-1098 (Fed. Cir. Oct. 15, 2025), both reject that shortcut.

Step 4: Use examples the way the cases do

Working examples report what was done, in the past tense. Prophetic examples describe what the inventors predict, in the present or future tense. Both are lawful. Atlas Powder Co. v. E.I. du Pont de Nemours & Co., 750 F.2d 1569 (Fed. Cir. 1984), and Alcon Research Ltd. v. Barr Laboratories, Inc., 745 F.3d 1180 (Fed. Cir. 2014), confirm that actual working examples are not required. Never write a prophetic example in the past tense, which is a known route to an inequitable conduct claim.

Several drafting practices follow from the recent decisions.

Spread the examples across the claim. In Juno Therapeutics, Inc. v. Kite Pharma, Inc., 10 F.4th 1330 (Fed. Cir. 2021), two examples could not represent a broad class of binding elements, and in AbbVie Deutschland GmbH & Co. v. Janssen Biotech, Inc., 759 F.3d 1285 (Fed. Cir. 2014), examples from one lineage did not represent a structurally diverse genus. Ten examples clustered in one corner may teach less than four spread to the edges.

Test the claim's endpoints. If you recite a range, have data at or near both ends, and say what happens beyond them. An open-ended range is especially dangerous: MagSil Corp. v. Hitachi Global Storage Technologies, Inc., 687 F.3d 1377 (Fed. Cir. 2012), and Medytox, Inc. v. Galderma S.A., 71 F.4th 990 (Fed. Cir. 2023), both turned on claims running well past what the patents showed.

Consider including failures. Reporting what did not work helps show the boundary of the class and gives content to the common quality. It also costs something, since it shows unpredictability and a low hit rate. Baxalta's record showed that only 1.6% of screened antibodies worked. But you owe candor anyway, and a patent that explains why the failures fail is making the argument the Federal Circuit found missing in that case: the Baxalta patent did not explain why the disclosed antibodies worked or why the others did not.

For in vivo claims, supply in vivo bridges. Wyeth LLC v. AstraZeneca Pharmaceuticals LP, No. 24-2325 (Fed. Cir. July 9, 2026), affirmed a JMOL that wiped out a $107.5 million verdict because the specification had in vitro data and "projected" dose ranges with no explanation of how they were derived. Claim limitations calling for a patient dose require some way of getting from the bench to the patient: animal data, a translation method, or dose-finding guidance. The opinion does not require clinical trial data. It recognizes that claiming doses without that data is accepted practice, as in United Therapeutics Corp. v. Liquidia Technologies, Inc., 74 F.4th 1360 (Fed. Cir. 2023). What it condemns is leaving the critical step entirely to the reader.

Sample language, prophetic example:

Prophetic Example 7. A cell is prepared as in Example 2, except that salt C is used at 2.5 weight percent. Because salt C has an anion radius of [X] pm, within the range shown in Table 3 to correlate with dendrite suppression, the cell is expected to reach [Y] cycles under Test Method A. This example has not been performed.

The last sentence is not legally required, and some drafters prefer tense alone to carry the point. It does remove all doubt, and doubt is what you do not want on this subject.

Step 5: Disclose the common quality, if there is one

The Court in Amgen suggested that a roadmap might suffice if "the inventor identifies a quality common to every functional embodiment." In practice, this calls for a section of the specification that states a rule, shows a correlation, and explains why. Something like:

Without being bound by theory, the inventors have found that [structural feature] accounts for [function]. Table 3 reports [function] for 30 candidates that vary in [feature]. Each of the nine candidates having [feature] within [range] showed [result]; none of the 21 candidates outside that range did. A person of ordinary skill can therefore identify candidates likely to work by [simple measurement], without testing each one for [function].

Mere recitation of a sequence identity percentage is not such a rule. In the nonprecedential Pioneer Hi-Bred International, Inc. v. Inari Agriculture, Inc., No. 25-1287 (Fed. Cir. Oct. 2, 2026), a claim to enzymes with 85% identity to a reference plus a motif was held not enabled. The genus was astronomically large, two examples were disclosed, and the patent owner's own later testing showed that only two of eight tested variants had the claimed function. Variants differing at only six positions behaved differently. The structural number did not predict function, and that was the problem.

If there truly is no common quality, you should know that before you claim the class, and you should consider claiming what you made.

Step 6: Name alternatives you can really support, and do not trash the ones you claim

Courts have reacted strongly when a specification teaches one thing and claims another.

• Liebel-Flarsheim Co. v. Medrad, Inc., 481 F.3d 1371 (Fed. Cir. 2007): claims covered injectors without a pressure jacket. Every embodiment had a jacket, and the inventors could not build one without it.
• Sitrick v. Dreamworks, LLC, 516 F.3d 993 (Fed. Cir. 2008): claims covered movies and video games; only games were taught.
• Automotive Technologies International, Inc. v. BMW of North America, Inc., 501 F.3d 1274 (Fed. Cir. 2007): mechanical sensors taught, electronic sensors merely mentioned.
• ICU Medical, Inc. v. Alaris Medical Systems, Inc., 558 F.3d 1368 (Fed. Cir. 2009), and Tronzo v. Biomet, Inc., 156 F.3d 1154 (Fed. Cir. 1998): one form described, a generic claim asserted.
• Brita: the patent called carbon block filters "unique," detailed the "drawbacks" of other media, and then claimed filters of any media reaching a performance number.

A boilerplate paragraph saying "the invention may use any suitable technique" is close to useless here, and sometimes it hurts by showing that nothing more specific was known. Instead, for every alternative you want covered, give a short but real teaching: what changes, what stays the same, a parameter range, and an example or a reasoned prediction. If the background section criticizes prior approaches, be careful not to criticize the very ones you later want the claim to reach. The specification's words on this come back as admissions. Brita quoted them, and treated them as warnings against the alternatives.

Step 7: For software, disclose algorithms for every claimed function

Suppose Tidewater Ledger is the fintech whose founder wrote the email that opened this guide. Its real invention turns out to be specific: it computes a graph embedding of recent merchant-card links, compares it to a cached embedding from a sliding window, and scores drift so that a transaction can be scored in under 20 milliseconds. That is a how. The slide said "uses AI to detect fraud." That is a what.

Disclose inputs, outputs, and transformation. For each claimed function, the specification should say what comes in, what goes out, and how the first becomes the second. That formula follows the Federal Circuit's requirements for corresponding structure. An algorithm may be expressed in prose, a formula, a flow chart, or any understandable form, under Finisar Corp. v. DirecTV Group, Inc., 523 F.3d 1323 (Fed. Cir. 2008), and Typhoon Touch Technologies, Inc. v. Dell, Inc., 659 F.3d 1376 (Fed. Cir. 2011). What does not count is restating the function, as in Noah Systems, Inc. v. Intuit Inc., 675 F.3d 1302 (Fed. Cir. 2012), or naming a black box, as in Blackboard, Inc. v. Desire2Learn, Inc., 574 F.3d 1371 (Fed. Cir. 2009), and Fintiv, Inc. v. PayPal Holdings, Inc., 134 F.4th 1377 (Fed. Cir. 2025), where the text on the "payment handler" simply parroted the claim.

Give each function its own figure. A single flow chart that repeats the claim adds nothing. A separate figure for each claimed function, with pseudo-code where it helps, gives you fallback structure and supports dependent claims. Practice notes on this topic make the same suggestion, including Carl Bilicska and Gregg Murgia's Functional Claiming for Computer-Implemented Inventions (Practical Law).

Sample specification language:

In step 310, the drift scorer receives (i) the current embedding vector E_t for the card-merchant pair and (ii) the cached window embedding E_w, each of dimension 128. The scorer computes the cosine distance d between E_t and E_w. Where d exceeds threshold T, set as described with reference to FIG. 5, the scorer applies the weighting in Equation 2 and outputs a score s between 0 and 1. In one embodiment, step 310 completes in under 20 ms on [hardware], because the cached window embedding avoids recomputing the graph traversal described in FIG. 4.

That paragraph does several things. It gives structure if § 112(f) applies, supports the enablement of the method, and gives a technical-improvement story for 35 U.S.C. § 101. On eligibility, see Prosecuting and Defending a Section 101 Challenge: A Practical Guide.

Write claims that recite operation. The § 112(f) test is whether the words have "a sufficiently definite meaning as the name for structure," Williamson v. Citrix Online, LLC, 792 F.3d 1339 (Fed. Cir. 2015). "Module" has failed repeatedly, including in Rain Computing, Inc. v. Samsung Electronics Co., 989 F.3d 1002 (Fed. Cir. 2021). "Code" can survive, as it did in Dyfan, LLC v. Target Corp., 28 F.4th 1360 (Fed. Cir. 2022), when coupled with language describing what it does and an unrebutted expert. But TrackTime, LLC v. Amazon.com Services LLC, No. 24-1102 (Fed. Cir. July 2, 2026), shows that "executable program code configured to" results in a live fight, whose outcome turns on how specific the claimed function is. The practical test the opinion suggests: does the claim say how, in inputs, outputs, and operations, or only what the user gets?

Decide deliberately whether you want § 112(f). Sometimes it helps. A means-plus-function element is construed to cover the disclosed structure and equivalents, which can provide validity protection in a crowded field, and it does not require the claim to cover every way of reaching the result. If you use it, disclose several algorithms and link each clearly. If you do not want it, avoid the nonce words. Do not rely on the narrow exception of In re Katz Interactive Call Processing Patent Litigation, 639 F.3d 1303 (Fed. Cir. 2011), for anything beyond generic processing, receiving, or storing: EON Corp. IP Holdings LLC v. AT&T Mobility LLC, 785 F.3d 616 (Fed. Cir. 2015), keeps it narrow.

Avoiding § 112(f) does not avoid § 112(a). A result-focused claim that escapes means-plus-function treatment must then be described and enabled across all of its scope. Under LizardTech, Inc. v. Earth Resource Mapping, Inc., 424 F.3d 1336 (Fed. Cir. 2005), one method does not describe all methods of reaching the same goal. So the stronger claims above are not there merely to avoid a label.

For machine learning, disclose the pipeline as well as the model. The Patent Office has moved its focus toward §§ 102, 103, and 112 for such claims, in the Appeals Review Panel's precedential Ex parte Desjardins, Appeal No. 2024-000567 (2025), and the Federal Circuit in Recentive Analytics, Inc. v. Fox Corp., 134 F.4th 1205 (Fed. Cir. 2025), has rejected claims that simply apply generic machine learning in a new field. There is no precedential Federal Circuit answer yet on how Amgen applies to a claim to any model that reaches a metric. The careful approach anticipates the question. Disclose the training data characteristics and how they were obtained, feature engineering, architecture, the loss function, hyperparameter ranges that worked and failed, the validation method and the metric's definition, and whatever explains why the design succeeds. Claim architecture or features at some rung, not just performance.

Step 8: Define what you measure, and audit every term of degree

Nautilus requires claims that inform skilled people of scope with reasonable certainty. Nautilus, Inc. v. Biosig Instruments, Inc., 572 U.S. 898 (2014). The way to meet it is to give them what they need on the page.

Name the test method. "Molecular weight" with three possible measures doomed claims in Teva Pharmaceuticals USA, Inc. v. Sandoz, Inc., 789 F.3d 1335 (Fed. Cir. 2015). So did an unspecified slope measurement in Dow Chemical Co. v. Nova Chemicals Corp., 803 F.3d 620 (Fed. Cir. 2015), and two inconsistent drying tests in HZNP Medicines LLC v. Actavis Laboratories UT, Inc., 940 F.3d 680 (Fed. Cir. 2019). Write the method out, and tie the claim to it.

Define "about," or do not use it. Enviro Tech Chemical Services, Inc. v. Safe Foods Corp., No. 24-2160 (Fed. Cir. May 4, 2026), held "about 7.6 to about 10" pH indefinite where examples tolerated varying deviations, the applicant never said what the word meant, and art sat at 7.0. "About" is not always fatal. Ortho-McNeil Pharmaceutical, Inc. v. Caraco Pharmaceutical Laboratories, Ltd., 476 F.3d 1321 (Fed. Cir. 2007), accepted it. But a definition costs a sentence:

As used herein, "about," when modifying a pH value, means within plus or minus 0.2 pH units of the stated value, as measured by Test Method B. When modifying a concentration, "about" means within plus or minus 5% of the stated value.

Three warnings. First, once written, the definition controls, so pick the tolerance with your prior art in mind. Second, keep your examples consistent with it. Third, do not define in one place and contradict in another. Sloppy use of "i.e." and "e.g." can matter: Interval Licensing LLC v. AOL, Inc., 766 F.3d 1364 (Fed. Cir. 2014), found that one embodiment described with examples did not supply a clear definition.

Anchor subjective words. "Unobtrusive" failed in Interval and "aesthetically pleasing" failed in Datamize, LLC v. Plumtree Software, Inc., 417 F.3d 1342 (Fed. Cir. 2005). "Visually negligible" passed in Sonix Technology Co. v. Publications International, Ltd., 844 F.3d 1370 (Fed. Cir. 2017), since specific examples and requirements in the specification gave an objective baseline. "Resilient" and "pliable" passed in Niazi Licensing Corp. v. St. Jude Medical S.C., Inc., 30 F.4th 1339 (Fed. Cir. 2022). Your job is to supply comparison examples, objective tests, or a threshold.

Use functional language with care, but do not fear it. Functional language is not per se indefinite under BASF Corp. v. Johnson Matthey Inc., 875 F.3d 1360 (Fed. Cir. 2017), and Cox Communications, Inc. v. Sprint Communication Co., 838 F.3d 1224 (Fed. Cir. 2016). The risks lie elsewhere: enablement across the class, and 112(f).

Be wary of coining terms to beat prior art. In the nonprecedential Network-1 Technologies, Inc. v. Google LLC, No. 24-1893 (Fed. Cir. Apr. 23, 2026), "non-exhaustive search" was added years after filing, to get around art, and could not be pinned down from the specification. If the amendment needs a new word, you may also have a new matter problem. See below.

Step 9: Check claim form for traps

• Every "or" is a promise. Satius Holding, LLC v. Samsung Electronics Co., No. 25-1446 (Fed. Cir. Oct. 1, 2026), invalidated a claim that recited transmitting "electric or electromagnetic signals over air" because one alternative was physically impossible. A claim also covering inoperative embodiments is not automatically invalid under Atlas Powder, but an express alternative that cannot work is a serious problem. Read every disjunctive as if it stood alone. In re Starrett, No. 22-2209 (Fed. Cir. June 8, 2023) (nonprecedential), involved a claim whose 47 "or" clauses covered over 140 trillion embodiments.
• No mixed claims. An apparatus claim that also recites a user's method step is indefinite under IPXL Holdings, L.L.C. v. Amazon.com, Inc., 430 F.3d 1377 (Fed. Cir. 2005). Write capability language ("configured to receive a selection") and not user action ("wherein the user selects").
• Check antecedents and sequence. Not fatal in themselves, as the nonprecedential MPH Technologies Oy v. Apple Inc., No. 25-1069 (Fed. Cir. Aug. 3, 2026), and Indect USA Corp. v. Park Assist, LLC, No. 24-1023 (Fed. Cir. Jan. 7, 2026), both show, but why pay to litigate it?
• Avoid negative limitations without support. A limitation that excludes something needs a basis: Novartis Pharmaceuticals Corp. v. Accord Healthcare Inc., 38 F.4th 1013 (Fed. Cir. 2022), says silence is generally not disclosure. If the invention works without a component, say so, and say why.
• Jepson preambles need support as well. In re Xencor, Inc., No. 24-1870 (Fed. Cir. Mar. 13, 2025).
• Be careful with undefined parameter claims. If you invent a metric, as in Brita, the claim reaches anything that hits the number. Consider combining the metric with structure at some rung.
• Do not limit a method-of-treatment claim to what you cannot teach. Wyeth's "unit dosage" limitation, adopted from the specification's definition and perhaps needed to distinguish art, is what the patent could not support.

Step 10: Consider how a method-of-use claim changes the analysis

Teva Pharmaceuticals International GmbH v. Eli Lilly & Co., No. 24-1094 (Fed. Cir. Apr. 16, 2026), held that a jury could find a method of treating headache with humanized anti-CGRP antagonist antibodies described and enabled, because the genus was well known and was not itself the invention, and a skilled reader would understand all of it works in the method. The court denied rehearing en banc on September 30, 2026, with Judge Dyk dissenting, and further review is possible. It remains prudent to file a method claim alongside composition claims when a new use of a known class is the real discovery.

To use this line, the specification should do what Teva's did. State plainly that the class as a whole is useful for the method, give the reason, cite the art showing the class and how to make it, and disclose examples. To stay out of University of Rochester v. G.D. Searle & Co., 358 F.3d 916 (Fed. Cir. 2004), make sure you are not claiming a method of doing X using "something that does X," with no such thing described or known. Where the class's efficacy varies, Idenix Pharmaceuticals LLC v. Gilead Sciences Inc., 941 F.3d 1149 (Fed. Cir. 2019), remains the governing caution.

Step 11: Map priority before you file anything

Every claim's date depends on the earliest application that supports it under § 112(a). 35 U.S.C. § 120 and § 119(e) both require it. A claim that loses priority can be anticipated by intervening art, including the very product you are suing on, as happened in Seagen Inc. v. Daiichi Sankyo Co., No. 23-2424 (Fed. Cir. Dec. 2, 2025).

Practical rules:

• Treat the provisional as a real application. A thin provisional provides a thin date. Preparing and Filing a Provisional Patent Application: A Practical Guide and the Provisional Application Toolkit: Invention Capture Templates, Written Description Audits, and Conversion Calendars show how to capture the invention at filing.
• Write down what later claims might need. If a specific subgenus is valuable, single it out in the text, with reasons for preferring it. In re Ruschig, quoted in Seagen, asks for blaze marks. A list of 83 amino acids at each of four positions is a forest with no marks.
• Before filing a continuation aimed at a competitor's product, make the chart. Each new limitation should map to specific text in the earliest application. Claiming a competitor's product is allowed if the original application supports it. If it does not, you will have bought an anticipation problem and an inventor deposition you will not enjoy.
• Add new matter only by a new filing. Under 35 U.S.C. § 132(a) and 37 C.F.R. § 1.121(f), an amendment cannot add new matter. A continuation-in-part can, but claims relying on it carry the later date.
• Incorporate by reference with care. Check 37 C.F.R. § 1.57, since essential material may be incorporated only from certain sources.
• Deposit where useful. Biological material that cannot be described in words may be deposited under 37 C.F.R. §§ 1.801 to 1.809, and Enzo Biochem, Inc. v. Gen-Probe Inc., 323 F.3d 956 (Fed. Cir. 2002), accepts a deposit's role in description.

Part 3: Prosecution

Step 12: Answer Section 112 rejections on their terms

Enablement. The Office's 2024 guidelines, 89 Fed. Reg. 1563 (Jan. 10, 2024), say examiners will keep using the Wands factors. Check that the rejection addresses them, then respond factor by factor, tied to the specific claim scope. Both In re PEN, No. 23-2282 (Fed. Cir. June 12, 2024) (nonprecedential), and Starrett show what happens when an applicant answers only with assertions, or addresses some embodiments rather than the full scope: the Board's decision stands, and arguments not made below are forfeited. A declaration under 37 C.F.R. § 1.132 from a skilled person can be very useful, since it puts facts in the record about skill, routine practice, and predictability. In Starrett, the applicant's bare claim that a specification was "fully enabled" got nowhere.

Written description. The examiner has the initial burden. Respond by mapping the claim to specific text and figures, or explain what a skilled person would understand from them. If the support is really absent, amend or file a new application. Do not argue obviousness of the missing feature: Ariad says that does not satisfy the requirement.

Indefiniteness. Remember In re Packard, 751 F.3d 1307 (Fed. Cir. 2014): the Office can reject a claim containing words whose meaning is unclear, and it is cheap to fix them now. Fixing it now is usually better than winning.

§ 112(f) interpretation. The Office applies a three-prong test under Examining Computer-Implemented Functional Claim Limitations for Compliance With 35 U.S.C. 112, 84 Fed. Reg. 57 (Jan. 7, 2019), and MPEP Section 2181. If an examiner says a limitation invokes 112(f), consider whether to contest, amend, or accept. Do not ignore it.

Sample response paragraph:

The Office Action does not address factors (2), (3), or (7) of In re Wands*. As to factor (2), paragraphs [0041] to [0055] teach that candidates with [feature] exhibit [function], and Table 3 reports that correlation across [n] diverse examples. As to factor (3), Examples 1 to 9 span the full recited range of [parameter]. As to factor (7), the attached declaration of Dr. [Name] explains that, given [feature], the result is predictable because [reason]. The claims do not call for a person of ordinary skill to screen candidates at random; they are limited to candidates identified by the disclosed measurement.*

Step 13: Think about how today's arguments read later

Two dangers appear again and again.

Arguments that distinguish prior art can make the claim narrower or vaguer. In Enviro Tech, inconsistent use of "about" in responses contributed to indefiniteness, consistent with Infinity Computer Products, Inc. v. Oki Data Americas, Inc., 987 F.3d 1053 (Fed. Cir. 2021). If you amend to distinguish art, say clearly what the amendment means.

Arguments about unpredictability can come back as enablement problems. If you tell the examiner the art is so unpredictable that a combination would not have been obvious, you have made statements a challenger will quote. If the specification also describes the art as complex, as Wyeth noted, you will need correspondingly greater guidance. It is possible to thread this: the invention may be non-obvious because the correlation was not known, and enabled because the specification now teaches it. Say it that way. For the obviousness side, see Overcoming Obviousness Rejections: A Comprehensive Guide to Section 103 Analysis and for general response practice Responding to Patent Office Actions: Strategies for Overcoming Rejections.

Step 14: Run a final audit before issue, and keep something pending

Before paying the issue fee, read the allowed claims against the specification as if you were a challenger. Then decide whether to keep a continuation pending. It gives you a place to put claims that respond to the market and to fix § 112 problems in light of new law. Reissue under 35 U.S.C. § 251 remains available for some defects, but broadening must be sought within two years of issue, and it invites scrutiny. A wider prosecution reference set is in the Patent Prosecution Toolkit: A Roadmap and Research Guide.

What the choices cost

Fee schedules and per-page translation costs change, so check current figures with counsel and the USPTO's fee schedule rather than relying on any number in a guide.

Part 4: Attacking a patent under Section 112

Step 15: Triage the claim

Who: litigation counsel with a technical expert. When: in the first 30 days after a notice letter or complaint if you can, because local rule deadlines come quickly.

Ask four questions in order. What is the broadest reading the patent owner needs to reach your product? Where does the specification teach that reading? Which words have no objective anchor? And which claim limitations are written as function?

A hypothetical shows why order matters. Suppose Ashgrove Therapeutics sues Corvane Labs on a claim to "a pharmaceutical composition comprising a kinase inhibitor that binds residue 412 and reduces tumor volume by at least 30% in a mouse xenograft model." The patent has four compounds, filed in 2018, and it is a 2023 continuation of a 2012 application. Corvane's product is a structurally distant molecule, first disclosed in a 2017 publication.

Corvane's team would look at (1) priority: does the 2012 text describe binding residue 412 at all? If not, its own 2017 publication may anticipate. Then (2) written description of the genus, with a distant accused compound as evidence under Juno and Amgen Inc. v. Sanofi, Aventisub LLC, 872 F.3d 1367 (Fed. Cir. 2017). Then (3) enablement across all compounds with that function, under Amgen and Baxalta. Finally (4) definiteness: is the xenograft model specified, including dose, line, and time point? If there are several reasonable protocols with different results, Teva and Dow apply.

Step 16: Pick the forum

• District court: All grounds are available, at clear and convincing. Microsoft Corp. v. i4i Ltd. Partnership, 564 U.S. 91 (2011).
• Post-grant review: Available within nine months of issue for first-inventor-to-file patents, under 35 U.S.C. § 321. It allows § 112 grounds on a preponderance standard. Pioneer Hi-Bred shows it working. Institution is discretionary, and Office practice has been shifting, so confirm current practice before planning around it.
• Inter partes review: Limited by 35 U.S.C. § 311(b) to prior art grounds. You can still use § 112 indirectly by arguing a claim is not entitled to priority, as in Regents of the University of Minnesota v. Gilead Sciences, Inc., 61 F.4th 1350 (Fed. Cir. 2023). Consider also what you say there: Teva used Lilly's IPR assertions that the art was replete and humanization routine to uphold the verdict against it. See Filing or Defending an Inter Partes Review: A Practical Guide.
• ITC: Section 337 respondents may raise § 112, as Brita shows.
• Hatch-Waxman or BPCIA: Section 112 is a staple in these cases, as discussed in Running an ANDA Litigation: A Practical Guide and Biologics and the BPCIA: The Patent Dance, Biosimilar Approval, and Exclusivity.

Step 17: Plead and contend specifically

A complaint or answer must have a factual basis under Fed. R. Civ. P. 11, and courts increasingly expect invalidity counterclaims to say more than a list of section numbers. Many patent local rules require written disclosure of Section 112 grounds with invalidity contentions; N.D. Cal. Patent L.R. 3-3 and E.D. Tex. P.R. 3-3 are the familiar models, and you should check the current version in your court. Failing to disclose can mean losing the ground.

Your contention should do what McRO, Inc. v. Bandai Namco Games America Inc., 959 F.3d 1091 (Fed. Cir. 2020), demands: concretely identify embodiments that fall within the claim and are not enabled, and explain how much experimentation it would take. "The claims are broad and the specification does not enable them" lost that case. The toolkit in this series has fuller contention language.

Mind the tension with noninfringement. If you argue your product is outside the claim, and also that the patent does not enable claims reaching your product, you need to plead in the alternative and manage how that looks. The standard framing: if the claim is read broadly enough to reach us, it is invalid.

Step 18: Use claim construction as a Section 112 event

Indefiniteness and § 112(f) issues are usually decided at or around the claim construction hearing, sometimes with expert testimony. Under Teva Pharmaceuticals USA, Inc. v. Sandoz, Inc., 574 U.S. 318 (2015), fact findings on extrinsic evidence receive clear-error deference, so a well-built record matters. Preparing for a Markman Hearing: A Practical Guide and the Claim Construction Toolkit: Charts, Briefs, Expert Declarations, and Hearing Materials cover the mechanics.

If you want § 112(f) treatment, you carry the burden. TrackTime says the proponent must show the phrase qualifies, with any underlying fact by a preponderance, and it vacated a judgment where the court had done too little with field usage. So retain an expert who can testify about what the words meant to practitioners at the time. Zeroclick, LLC v. Apple Inc., 891 F.3d 1003 (Fed. Cir. 2018), and Dyfan both reversed findings that lacked evidence or ignored it. Parse the phrase, identify the function, and show how general it is.

If you can use the squeeze, do. Accept the patent owner's broad construction, in the alternative, and then attack enablement across that scope. That's how the defendants won in Liebel-Flarsheim and Seagen.

Do not call inoperability indefiniteness. After Satius, argue impossibility primarily as enablement, and preserve it. There the trial court treated the enablement ground as waived at claim construction and the Federal Circuit reached it anyway, but no one should plan on that generosity.

Offer an alternative. In Gramm v. Deere & Co., No. 24-1598 (Fed. Cir. Mar. 11, 2026), a defendant's alternative construction, and its concession about how a prior model worked, were used in reversing its indefiniteness win. Think through what each alternative concedes.

Step 19: Prove it

Inventors first. The most damaging evidence in recent cases came from them. In Seagen, inventors said they had first seen the claimed subgenus in the accused product or the claims themselves. In Brita, they said they had changed technology, and that other media would require new technology. In Liebel, they could not build it. In Wyeth, their later publication described the dosing problem. Take those depositions with the specification in hand, asking what they had made by the filing date, what they had tried, and what they knew about alternatives. Handling Objections, Coaching, and Misconduct at a Deposition: A Practical Guide covers conduct issues.

Then the patent owner's own later work. Post-filing data about embodiments that could have existed at the filing date is admissible, under Amgen (2017) and Pioneer Hi-Bred. Request it. Limit the request to embodiments that existed or could have, since In re Entresto, 125 F.4th 1090 (Fed. Cir. 2025), and In re Hogan, 559 F.2d 595 (C.C.P.A. 1977), bar reaching back with later-arising technology.

Then the expert. Prepare a declaration or report that walks each Wands factor with facts, under Fed. R. Evid. 702. Identify the embodiments, explain what work they would take, and state the hit rate if you can. Avoid conclusory statements about meaning: Satius discounted such testimony as conclusory where it contradicted the party's own position. On disclosure and cross-examination generally, see Expert Witnesses After the 2023 Amendment to Rule 702: Daubert, Disclosure, and Cross-Examination and Expert Depositions: Taking Apart the Opinion, Protecting Your Own Expert, and Building the Daubert Record.

Move at the right time. Baxalta was decided on summary judgment under Fed. R. Civ. P. 56 because the screening facts were not disputed. Seagen, Wyeth, and Juno were JMOL cases. Move under Fed. R. Civ. P. 50(a) before the case goes to the jury, and renew under Rule 50(b), on each ground. An issue not preserved can't be rescued.

Part 5: Defending a patent under Section 112

Step 20: Decide what you need the claim to mean, then test it

Before you press a construction, ask the question Liebel-Flarsheim answers so painfully. If the claim covers the accused product, does the specification teach and describe that product's approach? If not, a narrower construction that still reaches a real market might be worth more than a broader one that invites invalidity. Run the Claim Construction Checklist: A Practical Checklist with this in mind.

Step 21: Frame the invention correctly

Claim framing wins these cases. Teva won because the court accepted that the invention was the use of antibodies to treat headache, and that the antibodies themselves were a known ingredient, in a line going back through Ajinomoto Co. v. International Trade Commission, 932 F.3d 1342 (Fed. Cir. 2019). Novartis won in Entresto because the invention was the combination and not the later complex. If your invention is a new use, a new combination, or an improvement in which a known class plays a known role, say so from the opening statement through the appeal.

Then address each factor and each theory in terms of the actual claim:

• Burden. Hold the challenger to clear and convincing evidence, and object when it speaks of breadth in the abstract. Alcon says Wands is not a free-floating breadth test, and McRO requires concrete embodiments.
• Routine. Show through skilled witnesses and literature what work practitioners consider ordinary. Beware that Baxalta and Idenix reject the idea that routine screening is enough when it amounts to searching for what works.
• Common quality. If the specification has one, build the case around it. If the expert can only say "you make them and test them," you have just given the Baxalta testimony.
• Function correlation. If you assert a structural genus, test whether structure predicts function before you assert it. Pioneer Hi-Bred lost on its own experiments.

Step 22: Prepare your witnesses for the questions that have ended cases

Experts. The "straightforward leap" testimony in Seagen is the textbook warning. An expert who explains that a skilled person could get to the claim by modification concedes the specification is not there. The right question for written description is what the document discloses to a skilled reader. And the right question for enablement is how much work it takes. Make sure the expert knows which is being asked. Do not let an expert testify that the specification need not teach what is novel: Genentech, Inc. v. Novo Nordisk A/S, 108 F.3d 1361 (Fed. Cir. 1997), and Brita will be quoted back. Rebut the other side's expert: Dyfan turned on unrebutted testimony, and so did Wyeth's toxicity evidence, which went against the patent owner.

Inventors. Prepare them honestly and thoroughly for "When did you first conceive of this exact subgenus?" and "Did you ever make one without the jacket?" You cannot and should not script the answers. You can make sure they have reread the application and understand the question. If the answers are bad, you need to know before you decide what to assert. See Expert Deposition Preparation Checklist: A Practical Checklist for the expert side.

Consistency. Line up your positions across every proceeding. Check what you told the examiner about predictability, what you told the Board on obviousness, and what you tell the jury about routine practice. Teva shows how powerful such statements are when they favor the other side.

Step 23: Protect the verdict you hope to win

The substantial evidence standard is your friend, but only for facts. Written description is a fact question, and Teva treats the jury as having made the supporting findings, yet Juno, Seagen, and Ariad show verdicts falling for lack of substantial evidence. So ask for findings the record supports, make a full record on each Wands factor, and offer dependent claims at trial that provide a separate path. Assert the rungs. Do not win on one claim and leave the others behind. For the appellate side of damages exposure, see Patent Damages: Reasonable Royalties, Lost Profits, Apportionment, and Enhancement.

Common mistakes

1. Claiming the slide. Claims that recite the marketing goal ("detect fraud in real time") and specifications that repeat it.
2. Treating "any suitable" as disclosure. It describes nothing, and Aristocrat Technologies Australia Pty Ltd. v. International Game Technology, 521 F.3d 1328 (Fed. Cir. 2008), makes this clear in software as to "appropriate programming."
3. Assuming that a screening protocol is an enabling disclosure of what it finds. That is the whole lesson of Amgen and Baxalta.
4. Thinking only biotech is at risk. Brita (water filters), Sitrick (movies), Liebel (injectors), LizardTech (image compression), and Satius (radio) say otherwise.
5. Using "module," "handler," "logic," or "unit" as the noun and hoping. After Williamson, Rain, Fintiv, and Egenera, Inc. v. Cisco Systems, Inc., 972 F.3d 1367 (Fed. Cir. 2020), that is a bet against the odds, especially when the disclosure repeats the claim.
6. Believing a 112(f) win is a validity win. A patentee who escapes § 112(f) must then enable and describe the full functional scope.
7. Leaving "about" undefined next to close art. See Enviro Tech.
8. Reciting a parameter with no method of measuring it. See Teva (2015), Dow, and HZNP.
9. Filing a continuation to cover a competitor without charting the earliest application. See Seagen and PowerOasis, Inc. v. T-Mobile USA, Inc., 522 F.3d 1299 (Fed. Cir. 2008).
10. Writing an expert report that says "a skilled person could get there." See Seagen, again.
11. Attacking "breadth" without naming an embodiment. See McRO.
12. Calling the art routine in one forum and unpredictable in another. See Teva for how that ends.
13. Ignoring the Section 101 overlap. A result-focused claim is exposed on eligibility as well. The Section 101 Toolkit: Claim Drafting, Motions to Dismiss, and Expert Support is a companion for that.
14. Forgetting to preserve. A JMOL motion on each ground, before and after verdict, is required for appellate rescue.

Before you go

For a fast version, use the Section 112 Patent Review Checklist: A Practical Checklist. For working papers, including claim support charts, the Wands worksheet, an algorithm template, and contention and declaration outlines, use the Section 112 Toolkit: Claim Support Charts, Invalidity Contentions, and Expert Declarations. For forum questions outside patent law, including removal and pleading jurisdiction, Getting Into or Staying Out of Federal Court: A Practical Guide to Jurisdiction and Removal helps. This guide is general information and not legal advice for a specific situation. If you are drafting a set of claims or facing one, MC Law can review it with you.

Related Documents

Related Articles

• Enablement, Written Description, and Indefiniteness After Amgen v. Sanofi: What Section 112 Demands: the doctrine and case readings behind each step here.
• How to Read a Patent: Claims, Specification, and Prosecution History: an orientation for inventors new to claims.
• Provisional Patent Applications: What They Protect, What They Do Not, and How to Keep the Priority Date: priority basics.
• Claim Construction and Markman Hearings: How Courts Decide What a Patent Actually Covers: how construction interacts with validity.
• Hatch-Waxman Litigation: The Orange Book, Paragraph IV Certifications, and the 30-Month Stay: where dose and use claims are tested.

Related Guides

• Preparing and Filing a Provisional Patent Application: A Practical Guide: how to make the first filing count.
• Prosecuting and Defending a Section 101 Challenge: A Practical Guide: the overlapping eligibility fight.
• Preparing for a Markman Hearing: A Practical Guide: where the definiteness record is made.
• Filing or Defending an Inter Partes Review: A Practical Guide: priority fights and the risk of admissions.
• Running an ANDA Litigation: A Practical Guide: Section 112 in generic drug cases.
• Handling Objections, Coaching, and Misconduct at a Deposition: A Practical Guide: rules for the inventor depositions described above.
• Getting Into or Staying Out of Federal Court: A Practical Guide to Jurisdiction and Removal: forum basics.

Related Checklists

• Section 112 Patent Review Checklist: A Practical Checklist: the short version for review before filing or suing.
• Preparing an Invention Disclosure: A Practical Checklist: intake for Step 1.
• Responding to a Patent Office Action: A Practical Checklist: general response practice.
• Expert Deposition Preparation Checklist: A Practical Checklist: preparing technical witnesses.

Related Toolkits

• Section 112 Toolkit: Claim Support Charts, Invalidity Contentions, and Expert Declarations: charts, worksheets, and templates that implement these steps.
• Provisional Application Toolkit: Invention Capture Templates, Written Description Audits, and Conversion Calendars: written description audits at the first filing.
• Claim Construction Toolkit: Charts, Briefs, Expert Declarations, and Hearing Materials: materials for the Markman stage.
• Patent Prosecution Toolkit: A Roadmap and Research Guide: prosecution research and practice references.