Mis-Mapping, Not Mini-Trial: A Case Study in the USPTO’s New Pre-Order Paper Procedure

Executive Summary: The USPTO’s new pre-order paper procedure gives patent owners a narrow but potentially consequential opportunity to challenge an ex parte reexamination request before the Office orders reexamination, but the article emphasizes that this opportunity must be framed consistently with the low SNQ threshold: the patent owner need not, and should not, try to disprove unpatentability; rather, it should show that the request misidentifies the cited reference’s teaching or mis-maps that teaching to the claim, such that the asserted teaching is not the kind of important, claim-relevant teaching that would raise an SNQ. MPEP § 2242; Official Gazette Notice, Pre-order Procedure Regarding Substantial New Question Determination in Ex Parte Reexamination Proceedings 1–3 (Apr. 1, 2026). The article develops that point through the 90/016,330 case study, where the Patent Owner argued that the Request improperly recast Barnard’s fixed-feature pruning and thresholding, supplemented by Knecht’s Fourier vectors and Lawrence’s threshold-estimation theory, as the ’639 patent’s ordered select → generate → extract → use → output feature-generation pipeline. Patent Owner Pre-Order Paper Providing Information Useful in Making the SNQ Determination, Reexamination Control No. 90/016,330, at 2, 17–30. Because the ’639 patent expired on May 17, 2022, the article also explains that the Office should apply the Phillips ordinary-and-customary-meaning framework rather than broadest reasonable interpretation, making the Patent Owner’s record-based constructions of “generating,” “target specific feature information,” and “background clutter” central to the SNQ mapping dispute. MPEP § 2258.I.G; Phillips v. AWH Corp., 415 F.3d 1303, 1312–17 (Fed. Cir. 2005).

I.             Introduction

The USPTO’s new “pre-order” paper procedure gives patent owners a limited opportunity to influence the threshold decision in ex parte reexamination: whether the request raises a substantial new question of patentability (“SNQ”). The opportunity is important, but its limits matter. The SNQ threshold is intentionally low. Under MPEP § 2242, an SNQ exists where the cited patent or printed publication contains a teaching that a reasonable examiner would consider important in deciding patentability, and where the same question of patentability has not already been decided or previously raised. MPEP § 2242. Critically, the MPEP states that a prima facie case of unpatentability is not required. Id.; see also In re Etter, 756 F.2d 852, 857 n.5 (Fed. Cir. 1985).

That threshold should shape how patent owners use the new procedure. A pre-order paper should not argue that the patent owner has already disproved unpatentability. It should argue something narrower: because the requester has misidentified what the reference teaches, or has mis-mapped that teaching to the claim, the asserted teaching is not the kind of important, claim-relevant teaching that would raise an SNQ in the first place. That formulation preserves the low SNQ threshold while giving patent owners a principled way to challenge requests that depend on technical or claim-mapping category errors.

Reexamination Control No. 90/016,330 provides a useful case study. The paper does not merely contend that U.S. Patent No. 6,894,639 is patentable over Barnard, Knecht, and Lawrence. Instead, it argues that the request recasts fixed-feature thresholding as the claimed feature-generation pipeline. Patent Owner Pre-Order Paper Providing Information Useful in Making the SNQ Determination, Reexamination Control No. 90/016,330, at 2 (June 20, 2026) (“Pre-Order Paper”).

II.          The Procedure: A Narrow Paper for a Threshold Determination

The USPTO’s Official Gazette notice permits a patent owner to file a pre-order paper before the Office decides whether to order ex parte reexamination. The paper may be filed without a petition or fee, is limited to thirty pages, and must be filed as soon as possible but no later than thirty days from service of the reexamination request. Official Gazette Notice, Pre-order Procedure Regarding Substantial New Question Determination in Ex Parte Reexamination Proceedings 1–3 (Apr. 1, 2026) (“Pre-Order Notice”).

The notice also defines the substantive boundaries of the submission. The paper should be limited to arguments or facts supporting the patent owner’s position that, despite the request’s alleged teachings, the Office should maintain the patentability determination made during examination. The notice further states that the paper should not address matters outside the request and should not argue discretionary denial under 35 U.S.C. § 325(d), which the Office treats as an issue taken up after the SNQ determination. Pre-Order Notice, supra, at 2.

The procedure is therefore best understood as a focused SNQ-stage intervention. It is not a preliminary response in the PTAB sense, and it is not a full merits brief. The patent owner’s strongest argument will usually be that the request’s asserted teaching, as actually disclosed and actually mapped, is not an important claim-relevant teaching under MPEP § 2242. In the 90/016,330 paper, that meant arguing that fixed features, pruning, and thresholding do not map onto the claimed sequence of selecting, generating, extracting, using, and outputting target-specific feature information.

III.      Claim Construction in Reexamination: The Expired-Patent Standard Controls

The case study also turns on claim construction. In ex parte reexamination, unexpired patent claims are generally given their broadest reasonable interpretation consistent with the specification. In re Yamamoto, 740 F.2d 1569, 1571–72 (Fed. Cir. 1984); MPEP § 2258.I.G. But that is not the standard here. The ’639 patent expired on May 17, 2022, and the USPTO applies the Phillips ordinary-and-customary-meaning framework in reexamination proceedings involving expired claims because those claims can no longer be amended. Phillips v. AWH Corp., 415 F.3d 1303, 1312–16 (Fed. Cir. 2005) (en banc); MPEP § 2258.I.G; see also CSB-Sys. Int’l, Inc. v. SAP Am., Inc., 832 F.3d 1335, 1340–41 (Fed. Cir. 2016).

That posture matters. The Patent Owner’s reliance on ordinary meaning, the specification, and the prosecution record is not simply district-court claim-construction doctrine imported into reexamination. It follows from the expired status of the claims. Under Phillips, claim terms are read in light of the intrinsic record, including the claims, specification, and prosecution history, from the perspective of a person of ordinary skill in the art. Phillips, 415 F.3d at 1313–17. For an expired patent, that framework gives the Office a narrower and more record-anchored lens than broadest reasonable interpretation.

This does not mean the Office must conduct a full Markman analysis before deciding whether to order reexamination. The SNQ threshold remains low, and no prima facie case of unpatentability is required. But the asserted SNQ still must be evaluated against the claim as properly understood. If the claim requires “generating said target feature information from said data statistics,” the requester cannot establish an SNQ merely by pointing to any statistical calculation in the prior art. The cited teaching must be important to the claim limitation as construed under the expired-patent standard.

In the 90/016,330 pre-order paper, this distinction is central. The Patent Owner’s argument depends on construing the “generating” limitation and related feature-information language in view of the claim’s ordered sequence and the intrinsic record. The paper argues that the request maps that limitation to Barnard’s fixed-feature thresholding, rather than to the generation of target feature information from data statistics. Patent Owner Pre-Order Paper Providing Information Useful in Making the SNQ Determination, Reexamination Control No. 90/016,330, at 8–14, 21–23. Because the ’639 patent is expired, the Office should evaluate that mapping under the ordinary-and-customary-meaning standard, not under the broader interpretive latitude that would apply to amendable claims.

IV.       The Case Study: Fixed-Feature Thresholding Versus Feature Generation

The pre-order paper’s central move is to define the challenged claim as an ordered technical pipeline rather than as a generic target-detection result. Claim 1, as the paper frames it, requires calculating data statistics, using those statistics to select target-specific feature information, generating target feature information from the statistics, extracting that generated feature information from the data, using it to distinguish targets from background clutter, and outputting target and background-clutter information. Pre-Order Paper, supra, at 2–3.

That framing matters because the request allegedly mapped the claims to a different architecture. According to the pre-order paper, Barnard begins with fixed, hand-engineered features such as brightness, size, and quietness; uses statistics to decide whether those fixed features are reliable enough to use; and then places a Bayes or decision threshold in that fixed feature space. Id. at 2, 14–23.

The patent owner’s point is not simply that Barnard is weak prior art. It is that Barnard’s asserted teaching is the wrong kind of teaching. A reasonable examiner might consider fixed-feature thresholding important for some claims directed to target detection. But the patent owner argues that fixed-feature thresholding is not important in the relevant SNQ sense for a claim that requires generating target feature information from data statistics and then extracting and using that generated information.

V.          The Technical Record: Why “Generating” Does the Work

The supporting exhibits reinforce the mapping argument. The exhibit list identifies Sanger as EX2006, the Majumder declaration as EX2007, Dr. Majumder’s CV as EX2008, the inventor-authored Technical Report as EX2009, and “Robust Classifiers Without Robust Features” as EX2010. Patent Owner’s Exhibit List, Reexamination Control No. 90/016,330.

The Technical Report is central. It states that the work used generalized Hebbian learning (“GHA”) to generate distinguishing target characteristics and defines clutter as non-target image regions with target-like characteristics. A.J. Katz, Hebbian Learning, Principal Components and Automatic Target Recognition, Tex. Instruments Tech. Rep. TR 08-91-14, at 4 (May 31, 1991) (“Technical Report”). The report further explains that target statistics, rather than clutter statistics, were used for feature generation because background clutter could change significantly from scene to scene. Id.

Sanger supplies the algorithmic foundation. Sanger describes the Generalized Hebbian Algorithm as a method for training a neural network to find eigenvectors of the input correlation matrix, ordered by decreasing eigenvalue. Terence D. Sanger, Optimal Unsupervised Learning in a Single-Layer Linear Feedforward Neural Network, 2 Neural Networks 459, 460 (1989). Dr. Majumder’s declaration ties that concept to the ’639 patent, explaining that a person of ordinary skill would understand the patent to use GHA to learn principal components of target-containing image patches and treat those components as discriminative filters for automatic target recognition. Declaration of Dr. Aditi Majumder ¶¶ 36–42, IPR2025-00340, Ex. 2007.

Katz, Gately, and Collins provide the clutter-variability backdrop. In Robust Classifiers Without Robust Features, Figure 1 depicts target and clutter distributions changing between image sets, with the decision surface shifting between the two; Figure 2 depicts a parameter-driven routing architecture for feature data. Alan J. Katz, Michael T. Gately & Dean R. Collins, Robust Classifiers Without Robust Features, 2 Neural Computation 472, 473–74 figs. 1–2 (1990). Those figures visually support the Technical Report’s point that clutter statistics are environment-dependent and may not transfer across scenes.

Together, the exhibits draw a technical line between two processes. On one side is generating feature information from data statistics, such as learning principal-component or receptive-field filters. On the other side is applying a threshold to predefined features after the features already exist. The pre-order paper argues that Barnard, Knecht, and Lawrence fall on the latter side.

VI.       The Reference-by-Reference Mapping Argument

The pre-order paper’s reference analysis works by assigning each reference a role and then showing why that role does not correspond to the claimed role.

Barnard is the primary mismatch. The paper characterizes Barnard as using predefined measures such as brightness, size, and quietness, followed by feature pruning and thresholding. Pre-Order Paper, supra, at 14–23. Under the patent owner’s view, Barnard’s use of statistics to determine whether fixed features are stable enough to use is not the same as generating target feature information from data statistics. Id. at 19–23.

Knecht is treated as an attempted patch that does not fill the relevant hole. Knecht uses Fourier power-spectrum feature vectors and stored decision rules for classifying scanned images, but the pre-order paper argues that those Fourier vectors are not generated from Barnard’s brightness, size, and quietness statistics and are not the same target-specific feature information mapped by the request. Id. at 15–16, 21–22.

Lawrence is treated similarly. The request allegedly uses Lawrence as a neural-network or Bayesian-estimation mechanism for threshold calculation. The patent owner’s response is that changing how a threshold is calculated does not transform a threshold into target feature information generated from data statistics. Id. at 22–23, 27–30.

This is the core SNQ-stage mapping argument. The patent owner is not asking the Office to decide every obviousness dispute. It is asking the Office to recognize that the request’s own mapping places the cited teachings in the wrong claim roles.

VII.    “Background Clutter” as a Claim-Relevant Limitation

The paper also argues that the request improperly treats “background clutter” as undifferentiated background. The Technical Report defines clutter as non-target regions that have target-like characteristics. Technical Report, supra, at 4. The pre-order paper uses that definition to argue that the claim is concerned with target-like non-target content that must be distinguished using generated target-specific feature information, not merely anything on the non-target side of a generic threshold. Pre-Order Paper, supra, at 23–25.

That point reinforces the mapping theme. If “background clutter” means target-like non-target content, then a generic target/background threshold may not supply the claimed treatment of background clutter. The issue is not whether Barnard has “background” in some broad sense. The issue is whether the request identifies a teaching corresponding to the claimed use and output of target and background-clutter information through the ordered feature-generation pipeline.

The argument also explains why the “generate” limitation matters downstream. If the request fails to identify generated target feature information, then it may also fail to identify the later extraction, use, and output of that same generated information. The pre-order paper thus links the alleged failure at the generation step to limitations [1.4] through [1.6]. Pre-Order Paper, supra, at 25–27.

VIII.                  The Related IPR: Procedural History Without Overclaiming

The same patent and related Barnard theories were involved in Tesla, Inc. v. Intellectual Ventures II LLC, IPR2025-00340. Tesla challenged claims 1–7 of the ’639 patent, and the asserted grounds included Barnard alone, Barnard with Knecht, Barnard with Lawrence, and Barnard with Lawrence and Knecht. Tesla, Inc. v. Intellectual Ventures II LLC, IPR2025-00340, Paper 13, at 6–7 (P.T.A.B. Aug. 25, 2025). The Board initially instituted review of all challenged claims on all asserted grounds. Id. at 3.

That institution decision did not become a final merits ruling. The Director later granted Director Review, vacated the institution decision, and denied institution. Tesla, Inc. v. Intellectual Ventures II LLC, IPR2025-00340, Paper 18, at 1, 4 (Dir. Nov. 5, 2025). The Director’s order focused on inconsistent claim-construction positions: in district court, Tesla argued that the “generating” limitation was indefinite, while in the IPR it asked the Board to apply plain and ordinary meaning. Id. at 2–4. The USPTO designated the Director Review decision informative.

This procedural history supports the case study in a limited way. It underscores the importance of the “generating” limitation and the challenger’s mapping theory. But it does not establish that the PTAB finally adjudicated Barnard’s technical teachings in the patent owner’s favor. The Director Review decision vacated institution because of a claim-construction-position problem, not because the Board reached a final merits determination that Barnard failed to teach the limitation. The pre-order paper therefore must stand on the request’s own asserted teachings and mappings.

IX.        The Limits of the Mapping Strategy in an Asymmetric Record

The mapping strategy has real force, but it must be understood against the procedural posture of a pre-order paper. The requester ordinarily does not get to file a response to the patent owner’s pre-order submission. The USPTO notice permits a requester response only by exception, such as to address alleged misrepresentations of fact or law or other improper arguments that materially impede the SNQ determination. Even then, the requester must file a petition under 37 C.F.R. § 1.182, pay the required fee, and limit the responsive paper to ten pages. Official Gazette Notice, Pre-order Procedure Regarding Substantial New Question Determination in Ex Parte Reexamination Proceedings 2–3 (Apr. 1, 2026).

That asymmetry matters. The practical question is not what arguments a requester might later develop if given a full reply. It is whether the request as filed identifies a new, important, claim-relevant teaching and maps that teaching to the challenged claims in a way that raises an SNQ. A patent owner’s pre-order paper can therefore focus on defects in the request’s existing theory, while the Office remains free to evaluate the request, the patent owner’s paper, and any permitted requester response under the low SNQ threshold. Pre-Order Notice, supra, at 3.

First, the request as filed may assert that Barnard’s fixed feature set and thresholding process constitute “feature information.” That position draws some force from the low SNQ threshold: the request need not prove unpatentability, and no prima facie case is required for an SNQ. The patent owner’s answer, however, is that claim 1 does not recite feature information in isolation. It recites feature information selected, generated, extracted, used, and output through an ordered sequence. The mapping dispute is therefore not whether Barnard contains “features” in a general sense. It is whether the request identifies Barnard teachings that correspond to the claimed generated target feature information. The 90/016,330 paper frames the mismatch this way: Barnard uses fixed features such as brightness, size, and quietness, applies statistical screens, and places a threshold in that fixed feature space; the ’639 patent, by contrast, claims a select → generate → extract → use → output pipeline. Patent Owner Pre-Order Paper Providing Information Useful in Making the SNQ Determination, Reexamination Control No. 90/016,330, at 2, 17–23.

Second, the Office need not resolve every claim-construction or obviousness dispute before ordering reexamination. That limitation is important because the pre-order procedure should not become a mini-trial. But the patent owner’s mapping argument need not ask for a final merits determination. It can ask only that the Office recognize that the request’s asserted teaching is materially mismatched to the claim limitation. If the request maps “generating” to threshold placement, the Office can determine that the asserted teaching does not raise the relevant SNQ without deciding every possible obviousness theory that could have been constructed.

Third, secondary references and combinations matter only to the extent the request actually articulates their role. The pre-order paper need not defeat a better request that could have been written. It need only show that the filed request does not identify and map the missing claim-relevant teaching. In the case study, the patent owner argues that Knecht supplies Fourier vectors and stored decision rules, and Lawrence supplies, at most, a neural-network or Bayesian-estimation mechanism for threshold calculation. The paper’s point is that neither reference, as used in the request, supplies target feature information generated from Barnard’s data statistics or the claimed ordered pipeline. Pre-Order Paper, supra, at 15–16, 21–23, 27–30.

This procedural posture sharpens the patent owner’s burden. The pre-order paper should not try to anticipate every conceivable reply. Instead, it should identify the teachings actually asserted in the request, show how the request maps those teachings to the claim, and explain why that mapping does not present an important claim-relevant teaching under the SNQ standard. Because the requester ordinarily has no right to respond, the patent owner should be precise and restrained: overstatements may invite an exceptional requester response, while a focused mapping critique is more likely to assist the Office within the compressed SNQ decision window.

X.          Lessons for Patent Owners

The principal lesson is that a pre-order paper should be drafted around the request’s alleged teaching, not around the patent owner’s complete validity defense. The 90/016,330 paper repeatedly returns to the same technical mismatch: Barnard supplies fixed features and thresholding; Knecht supplies Fourier vectors and decision rules; Lawrence supplies, at most, threshold estimation; none supplies generation of target feature information from data statistics followed by extraction, use, and output. Pre-Order Paper, supra, at 17–30.

Claim construction can be used, but it should be tied to SNQ mapping. Assuming the ’639 patent is expired, the ordinary-and-customary-meaning framework likely applies. MPEP § 2258; Phillips, 415 F.3d at 1316. That makes the specification, prosecution record, and contemporaneous technical record especially relevant. But the claim-construction discussion should remain instrumental: it should explain why the request’s asserted teaching does or does not correspond to the claim language.

Declarations and exhibits should support the paper, not replace it. The Pre-Order Notice permits declarations, but the Office will rely on arguments made in the paper and will not permit incorporation by reference. Pre-Order Notice, supra, at 2. The 90/016,330 paper uses Sanger, the Technical Report, Katz/Gately/Collins, and Dr. Majumder’s declaration to explain the technical difference between generated feature operators and fixed-feature thresholds. That is the proper role for supporting evidence at the SNQ stage.

XI.       Lessons for Requesters

The case study also offers lessons for requesters. A reexamination request should not assume that broad functional similarity will be enough. If the claim recites a sequence of technical roles, the request should identify which reference performs each role and why the teaching is important to patentability under the SNQ standard.

A requester relying on Barnard-style fixed features should confront the “generated feature information” issue directly. It should explain whether the fixed features themselves are the claimed feature information, whether the threshold is part of that information, or whether a secondary reference supplies a generated feature operator. If the theory depends on a combination, the request should articulate the combination in the request rather than leave the role mapping to inference.

Claim-construction consistency also matters. The IPR2025-00340 Director Review decision did not decide the Barnard merits, but it illustrates the procedural risk of inconsistent positions across forums. A party that argues indefiniteness in district court and plain meaning before the Board must explain why those positions are warranted. Tesla, IPR2025-00340, Paper 18, at 2–4.

XII.   Conclusion

The 90/016,330 pre-order paper is a useful model because it treats the SNQ inquiry as a technical mapping problem. It does not need to prove that the claims are patentable over every possible combination. It needs to show that the filed request has not identified an important, claim-relevant teaching that would raise an SNQ.

That distinction is key to using the pre-order procedure without overstating it. The SNQ threshold remains low, and no prima facie case of unpatentability is required. But a low threshold is not no threshold. A request still must identify a patent or printed-publication teaching that a reasonable examiner would consider important to patentability of the challenged claim.

Where the request maps a generated-feature limitation to fixed-feature thresholding, maps background clutter to generic background, or relies on a secondary reference for a role the request does not actually articulate, the patent owner can argue that the request has not raised the right patentability question at all. The resulting fight is not a mini-obviousness trial. It is a focused front-end dispute over whether the requester has identified the correct teaching and mapped it to the claim in a way that matters under the SNQ standard.