DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This Final Office Action is responsive to Applicant's amendment filed on 11 August 2026. Applicant’s amendment on 11 August 2026 amended Claims 1, 2, 4, 5, 11-16, and 18. Currently Claims 1-6, and 8-20 are pending and have been examined. Claim 7 was previously canceled. The Examiner notes that the 101 rejection has been maintained.
Claim Objections
Applicant’s amendments and arguments with respect to the rejection of the claims 1-6 and 8-20 user U.S.C. 103 have been fully considered and, in view of the amendment are persuasive. The rejection under U.S.C. 101 is withdrawn. Claims 1-6 and 8-20 would be in condition for allowance if rewritten or amended to overcome the rejection under U.S.C. 101 as set forth in the Office Action.
Especially, the prior art of record, taken alone or in combination, fails to teach or fairly suggest, and it would not have been obvious to one of ordinary skill in the art before the effective filing date to combine the references to arrive at, concurrently combining each type of the first observation data from the multiple workers that have the same date of the observation, wherein the combining transforms entries from yes-no entries into combined entries that include a number of observation for each type organized by date of the observation, together with generating a training data set from the combined entries by linking the date of the observation of each combined entry to the date of the reported unsafe incident and categorizing each combined entry that is not linked to a date of a reported unsafe incident as being associated with no unsafe incident, and training the machine learning model with that training data set by adjusting coefficients associated with the identified features, in combination with the remaining limitations of independent claims 1 and 12. With respect to independent claim 18, the prior art of record likewise fails to teach or fairly suggest the recited concurrently combining step together with associating the second observation data with the incident data by linking date of the observation data to a date of the unsafe incidents, identifying the predictive features, and adjusting coefficients associated with the identified predictive features according to their importance to the occurrence of the unsafe incident.
Response to Arguments
Applicant's arguments filed 11 August 2026 have been fully considered but they are not persuasive.
The Applicant argues on pages 12-13 that “Regarding Step 2A, Prong One, the Examiner has identified the "predicting… the likelihood of an occurrence of a first future unsafe incident" and "identifying… the features that are most indicative of the occurrence of the reported unsafe incident" limitations as allegedly reciting mathematical concepts and mental processes, respectively. Advisory Action, page 2. Applicant does not repeat the "ordered combination" framing at Prong One. Rather, even if these individual limitations recite abstract ideas at Prong One, the analysis properly proceeds to Prong Two, where the additional elements particularly the specific data transformation recited in the amended combining step, integrate any such exception into a practical application. The Examiner has acknowledged that the "concurrently combining" step itself "presents a challenge under the mental processes grouping" because it "cannot practically be performed in the human mind." Advisory Action, page 2. This concession is significant: the combining step is an additional element beyond the recited judicial exception that must be evaluated at Prong Two for whether it integrates the exception into a practical application”.
The Examiner respectfully disagrees.
With respect to the argument the Examiner notes that the Applicant's Argument effectively concedes that the core limitations of the independent claims specifically, "predicting the likelihood of an occurrence of a first future unsafe incident" and "identifying the features that are most indicative of the occurrence of the reported unsafe incident" recite abstract ideas at Step 2A, Prong One, and reframes the argument as one directed exclusively to Step 2A, Prong Two. Applicant argues that because the Examiner acknowledged the "concurrently combining" step cannot practically be performed in the human mind, that step is necessarily an "additional element beyond the recited judicial exception" that must be given full weight at Prong Two and that, in Applicant's view, integrates the abstract idea into a practical application. This argument mischaracterizes both the legal framework for Step 2A, Prong Two and the nature of the "concurrently combining" step as claimed and as described in the specification.
The Examiner agrees with the procedural premise that a step which cannot practically be performed in the human mind is appropriately characterized as an additional element beyond the recited judicial exception, not as itself reciting the judicial exception under the mental process grouping. The Examiner further agrees, consistent with MPEP 2106.04(d) and the 2019 PEG, that all additional elements including those that are well-understood, routine, and conventional must be given weight and evaluated at Step 2A, Prong Two. The Examiner has given the "concurrently combining" step full consideration as an additional element in the Prong Two analysis and has determined, for the reasons set forth in the rejection, that it does not integrate the abstract idea into a practical application. The mere fact that an additional element is not itself an abstract idea does not mean that it integrates the exception into a practical application; the question is whether it applies, relies on, or uses the judicial exception in a manner that imposes a meaningful limit on the exception such that the claim is more than a drafting effort designed to monopolize it. MPEP 2106.04(d); Alice Corp. Pty. Ltd. v. CLS Bank Int'l.
The "concurrently combining" step, as recited and as described in the specification, does not impose such a meaningful limit. The specification at paragraph [0049] characterizes the combining step in express terms as a technique for reorganizing observation data "for more optimal use in training" the machine-learning model by reducing the number of entries while "correctly representing the types of observation data that may be most indicative of an unsafe incident." This language establishes that the entire purpose and effect of the combining step is to improve the accuracy and efficiency of the downstream statistical prediction which is the abstract idea itself. An improvement to the abstract idea is not an improvement to technology. In re Bd. of Trs. of Leland Stanford Junior Univ. (Stanford II); MPEP 2106.05(a), subsection II. Because the combining step operates as pre-solution data preprocessing whose benefit flows entirely to the downstream abstract prediction, and because it does not improve the functioning of any computer, any technological component, or any other technology or technical field, it does not integrate the abstract idea into a practical application regardless of whether it is characterized as a separate additional element rather than as part of the judicial exception. Furthermore, the combining step does not effect a transformation of a particular article to a different state or thing within the meaning of MPEP 2106.05(c), as the reorganization of intangible data records into a different format does not constitute the type of physical or quasi-physical transformation the courts have found meaningful under that consideration. See Alice Corp. Accordingly, even giving the "concurrently combining" step full weight as an additional element and evaluating it both individually and in combination with all other additional elements in the claims, the Examiner concludes that the claims as a whole do not integrate the judicial exception into a practical application. The rejection is therefore maintained.
The Applicant argues on pages 13-14 that “Regarding Step 2A, Prong Two, even assuming arguendo that the claims recite a judicial exception, the claims as amended integrate any such exception into a practical application. The Examiner has characterized the combining step as "pre-solution data organization" that merely prepares inputs for the downstream predictive process. Advisory Action, page 3. Applicant respectfully disagrees with this characterization. The transformation recited in the amended claims is not neutral data preparation. It fundamentally changes the unit of analysis for the machine- learning model from per-worker-per-day granularity to date-level aggregated entries that include a number of observations for each type. This structural change enables the machine-learning model to detect date-level patterns across multiple workers that would otherwise be obscured by worker- level granularity in the data.
The specification describes that this transformation "can reduce the number of overall entries of observation data 26… while correctly representing the types of observation data 26 ... that may be most indicative of an unsafe incident." As-Filed Specification, paragraph [0049]. Critically, the specification further explains that this reorganization changes the data "from a 'yes- no' entry into a 'number of yeses/number of nos' entry." Id. This is not merely collecting or presenting data. It is a specific structural transformation that produces a fundamentally different data representation. The transformation changes what information the machine-learning model receives and how it can process that information. Specifically, converting yes-no entries to observation counts changes the representational structure of the data from a categorical format to a numerical aggregate format, which is a different data type that enables the machine-learning model to perform numerical statistical operations directly rather than first having to derive aggregate statistics from redundant individual records. The specification at paragraph [0063] confirms that this combining step "allows for the compression and/or reorganization of observation data, weather data, worker data, and/or incident data for potentially easier and/or more suitable use by machine-learning model 72." As-Filed Specification, paragraph [0063]. The phrase "more suitable use" indicates that the transformation is specifically designed to change how the machine- learning model operates on the data, not merely to reduce storage requirements, but to provide the model with a data structure that is better suited to its feature extraction and prediction operations. Furthermore, the transformation eliminates the unit-of-analysis mismatch between per-worker observation entries and per-project-per-date incident data, directly addressing a structural problem in how the machine-learning model correlates inputs to outputs”.
The Examiner respectfully disagrees.
With respect to the argument the Examiner notes that while the Applicant argues that the "concurrently combining" step is not mere "pre-solution data organization" but rather a "fundamental" and "structural" transformation that: (1) changes the unit of analysis from per-worker-per-day granularity to date-level aggregated entries; (2) enables detection of date-level patterns across multiple workers that would otherwise be obscured; (3) converts categorical yes-no data into a numerical aggregate format enabling different statistical operations; and (4) eliminates a unit-of-analysis mismatch between per-worker observation entries and per-project-per-date incident data. Applicant further relies upon paragraphs [0049] and [0063] of the specification to characterize this step as one that changes how the machine-learning model "operates on the data" and makes the data "more suitable" for use by the model. The Examiner respectfully disagrees with this characterization and its legal implications, and the rejection is maintained.
The threshold question under the improvements consideration at Step 2A, Prong Two is not whether an additional element changes the format or structure of the data that is fed to a computational system, but whether the claimed invention improves the functioning of a computer itself or improves some other technology or technical field. MPEP 2106.04(d)(1) and 2106.05(a); Ex Parte Desjardins, Appeal No.; December 5, 2025 Memorandum from the Deputy Commissioner for Patents. The combining step, however specifically characterized, does not do either. It does not improve how the computer processor executes instructions, does not improve network communications, does not improve memory architecture, does not improve the machine-learning model's internal algorithmic structure or training process, and does not solve any technical problem arising from the limitations of the computing environment itself. What Applicant describes as an improvement changing the representational format of data so that the machine-learning model receives numerical aggregates instead of individual binary entries is precisely an improvement to the accuracy and efficiency of the downstream statistical prediction, which is the abstract idea. As the MPEP and controlling case law make clear, an improvement to the abstract idea itself is not an improvement to technology. In re Bd. of Trs. of Leland Stanford Junior Univ. (Stanford II); MPEP 2106.05(a), subsection II. Applicant's argument that converting yes-no entries to numerical aggregate counts enables the machine-learning model to "perform numerical statistical operations directly rather than first having to derive aggregate statistics from redundant individual records" is, at its core, a description of how the abstract idea statistical feature extraction and probabilistic prediction is made to operate more efficiently. The fact that this efficiency is achieved through a preprocessing step rather than a modification to the model itself does not transform the nature of the improvement, which remains an improvement to the predictive process rather than to any technological component.
The specification passages Applicant relies upon confirm rather than undermine this conclusion. Paragraph [0049] states that the reorganization can reduce "the number of overall entries of observation data while correctly representing the types of observation data that may be most indicative of an unsafe incident," and paragraph [0063] states the combining step "allows for the compression and/or reorganization of observation data...for potentially easier and/or more suitable use by machine-learning model 72." These passages describe benefits that are measured entirely in terms of the statistical prediction's accuracy and the machine-learning model's training efficiency not in terms of any improvement to a computer, network, processor, or other technological system. This is analogous to Trading Technologies Int'l, Inc. v. IBG, LLC, where the court determined that a claimed user interface that provided a trader with more and better-organized information to facilitate market trades improved the business process of trading, but did not improve computers or technology. Similarly, here, providing the machine-learning model with better-organized, numerically aggregated data improves the process of predicting safety incidents the abstract idea but does not improve computers or technology.
Furthermore, Applicant's characterization of the combining step as eliminating a "unit-of-analysis mismatch" between per-worker observation entries and per-project-per-date incident data is a description of a methodological data alignment problem arising from the particular domain being modeled worksite safety incident prediction not a technological problem arising from the limitations of the computing environment. The problem being solved is one of data science methodology: how to align the granularity of input data with the granularity of the output data for purposes of statistical modeling. Solving a methodological problem in a statistical modeling domain by preprocessing data is an application of statistical and data management concepts, not a technological improvement to any computing system. See Electric Power Group, LLC v. Alstom S.A., (finding that steps directed to collecting, analyzing, and presenting information, however specifically tailored to a domain, do not improve computer technology). The Examiner also notes that the Desjardins decision, properly understood under the December 5, 2025 Memorandum, found eligibility because the specification disclosed improvements to how the machine learning model itself functions specifically, overcoming catastrophic forgetting by adjusting parameter values during successive training tasks, which changed the model's internal operational behavior. That improvement was intrinsic to the model's computational architecture. In contrast, the combining step here is entirely external to the machine-learning model's internal functioning; it reorganizes inputs before they enter the model and does not alter how the model processes those inputs, adjusts its parameters, or performs its predictions once the data is received. Accordingly, the combining step does not satisfy the improvements consideration under Step 2A, Prong Two, and the rejection is therefore maintained.
The Applicant argues on page 14 that “The Examiner distinguished Ex parte Desjardins on the basis that the improvement there was "to the model's internal operation against the technical problem of 'catastrophic forgetting."' Advisory Action, page 3. While the specific technical problem differs, the specification of the current application describes a specific improvement to how the machine-learning model operates by changing the data structure it receives. The transformation from yes-no entries to date- aggregated entries that include a number of observations for each type is not merely an optimization for efficiency. It changes the fundamental structure of the input data, enabling the model to identify correlations at the date level across workers rather than at the individual worker level. This is an improvement to the technological process of machine-learning-based incident prediction, not merely an improvement to the abstract idea of prediction itself”.
The Examiner respectfully disagrees.
With respect to the argument the Examiner notes that while the Argument attempts to analogize the "concurrently combining" step to the improvement found eligible in Ex Parte Desjardins by arguing that changing the structural format of the input data from per-worker binary yes-no entries to date-aggregated count-based entries constitutes "an improvement to the technological process of machine-learning-based incident prediction" rather than merely an improvement to the abstract idea of prediction itself. The Examiner has carefully considered this argument under the framework established by Ex Parte Desjardins, Appeal No. 2024-000567 (precedential), as incorporated into the MPEP by the December 5, 2025 Memorandum from the Deputy Commissioner for Patents, and finds it unpersuasive for two independent and dispositive reasons.
First, the argument fails at the threshold step of the improvements analysis because the specification does not describe the combining step as providing a technological improvement that would be apparent to one of ordinary skill in the art, but rather describes it as a data organization technique serving the accuracy and efficiency of the statistical prediction. MPEP 2106.05(a); December 5, 2025 Memorandum. The two-step improvements analysis requires: (1) the specification must describe the invention such that a technological improvement would be apparent to one of ordinary skill in the art, and (2) the claim must reflect that disclosed technological improvement. Id. The specification at paragraph [0049] describes the combining step as reorganizing data "for more optimal use in training" the machine-learning model, and at paragraph [0063] as permitting "potentially easier and/or more suitable use by machine-learning model 72." Applicant characterizes the phrase "more suitable use" as indicating that the transformation changes how the machine-learning model "operates on the data." However, "more suitable use by the model" is not equivalent to an improvement in how the model itself operates. A model receiving better-organized inputs may perform a prediction more accurately, but that improvement in predictive accuracy is an improvement to the outcome of the abstract idea the statistical prediction not to the functioning of the machine learning model as a technological artifact. The specification discloses no technical problem with the machine-learning model's internal architecture, no identified failure mode in the model's computational processes, and no mechanism by which the combining step alters the model's parameters, training algorithm, loss function, network architecture, or any other internal operational characteristic. Contrast this with Desjardins, where the specification specifically identified the technical problem of "catastrophic forgetting" a known failure mode intrinsic to the computational architecture of continual learning systems whereby a model's previously learned parameter values are overwritten during subsequent training and disclosed a specific mechanism operating within the model's parameter adjustment process to overcome that failure. The improvements credited in Desjardins were, as the December 5, 2025 Memorandum explains, "tantamount to how the machine learning model itself would function in operation and therefore not subsumed in the identified mathematical calculation." No analogous disclosure exists in the present specification. The present specification does not identify any technical problem intrinsic to the incident predictor's computational architecture that the combining step overcomes; it describes a data organization benefit whose measure is the downstream prediction's accuracy.
Second, even accepting arguendo that the specification could be read as suggesting some form of improvement to the machine-learning-based prediction process, Applicant's own description of that improvement confirms that it is an improvement to the accuracy of the abstract prediction rather than to a technology or technical field. Applicant argues that converting yes-no entries to numerical aggregates enables the machine-learning model to "identify correlations at the date level across workers rather than at the individual worker level." This is a description of what statistical patterns the model can detect which is entirely a function of the abstract idea of feature extraction and statistical modeling not a description of how the model's computational machinery has been improved. The ability to detect date-level correlations rather than worker-level correlations is a consequence of how the input data is structured for analysis, not a consequence of any change to the model's internal algorithmic or computational processes. In Recentive Analytics, Inc. v. Fox Corp., the Federal Circuit emphasized that applying machine learning to a new domain or improving the inputs to a generic machine-learning model does not, without more, constitute a technological improvement to the machine-learning technology itself. The present claims do precisely that: they apply a conventional machine-learning model to worksite safety data that has been preprocessed in a particular way. The preprocessing step, however described, does not improve the machine-learning technology it applies data organization concepts to prepare inputs for an otherwise unchanged, generic, and conventionally operating model. The Examiner therefore maintains the rejection and reaffirms that Desjardins is distinguishable not merely because "the specific technical problem differs," as Applicant acknowledges, but because the nature and locus of the improvement are categorically different in Desjardins, the improvement operated within the model's own computational architecture to address a technical failure intrinsic to that architecture, while here the asserted improvement operates externally to the model and improves only the accuracy of the abstract prediction the model performs. The rejection is therefore maintained.
The Applicant argues on pages 14-15 that “The Examiner also distinguished Examples 47 (Anomaly Detection) and 48 (Speech Separation), stating that those examples involved "operative technical steps that produced a tangible technological output" whereas the combining step merely "reorganizes data records into a different structural format as input for a machine-learning prediction." Advisory Action, page 3. The amended claims do more than reorganize records into a different structural format. Rather, the claims recite a transformation that fundamentally changes the data type, from yes-no entries to entries that include a number of observations for each type, and the unit of analysis, from per- worker-per-day to per-date aggregated entries. This changes what the machine-learning model operates on, analogous to how the speech separation example in Example 48 changed what waveform the model operated on by synthesizing and combining separate speech waveforms. In both cases, the transformation produces a structurally different input that changes how the downstream model processes the data”.
The Examiner respectfully disagrees.
With respect to the argument the Examiner notes that while the Applicant argues that the "concurrently combining" step is analogous to the synthesizing and combining steps of Example 48 (Speech Separation, Claim 2) because, in Applicant's view, both transformations produce "a structurally different input that changes how the downstream model processes the data." This analogy is inapt and rests on a fundamental mischaracterization of why Example 48's steps (f) and (g) were found to integrate the abstract idea into a practical application. A careful reading of Example 48 as published reveals that the eligibility of steps (f) and (g) did not turn on the mere fact that those steps produced a structurally different input for downstream processing. Rather, eligibility turned on three specific features of those steps that are entirely absent from the present claims: (1) the specification in Example 48 identified a defined technical problem in the field of speech separation namely, the inability of existing systems to separate speech from different sources belonging to the same class without requiring speaker-specific training data; (2) steps (f) and (g) produced a new and distinct technological artifact a synthesized mixed speech signal in the time domain that affirmatively excluded audio from an unwanted source that itself constituted a concrete technological output usable for audio playback, social media upload, or downstream transcription, none of which was achievable using the raw input data; and (3) the specification explained with technical specificity how the claimed combination of steps embedding vector determination, clustering, binary masking, waveform synthesis, and signal combination solved the identified technical problem in a particular way. The Office's analysis in Example 48 expressly found that steps (f) and (g) were "directed to creating a new speech signal that no longer contains extraneous speech signals from unwanted sources" and that these steps "reflect the improvement described in the disclosure" rather than being "insignificant extra-solution activity, mere instructions to apply the exception, or mere field of use limitations." None of these three features is present in the instant claims.
The present claims do not identify any technical problem in the field of computing, data processing, or machine-learning architecture. The specification identifies a workplace safety management problem the occurrence of unsafe incidents resulting in injuries, fines, and property damage not a technical problem arising from the limitations of any computational system. The combining step does not produce a new and distinct technological artifact usable independently of the downstream prediction. The reorganized date-aggregated observation data is not itself a new output product; it is internal intermediate data whose sole function is to serve as input to the machine-learning model's feature extraction and prediction process. This is categorically different from the new, usable speech signal produced in Example 48, which could be stored, transmitted, played back, and used for transcription independently of any further abstract processing. And the specification does not explain with technical specificity how the combining step solves a technical problem in the manner that Example 48's disclosure explained how the particular speech-separation technique overcame the prior art's requirement for speaker-specific training. Applicant's assertion that the combining step is analogous because it "changes what the machine-learning model operates on" could be said of any data preprocessing step in any machine-learning system normalizing data, removing outliers, imputing missing values, or encoding categorical variables all change what the model operates on, yet none of these standard preprocessing techniques would be found to integrate an abstract idea into a practical application on that basis alone. The distinguishing characteristic of Example 48 was not that the input to downstream processing was changed in structure, but that the transformation produced a new technological artifact with independent utility that reflected a specific technical improvement over existing technology disclosed with particularity in the specification. Because neither the combining step nor the specification of the present application satisfies any of the features that drove the eligibility finding in Example 48, the analogy fails and the rejection is therefore maintained.
The Applicant argues on page 15 that “Regarding Step 2B, the Examiner has cautioned against conflating the novelty inquiry with the eligibility inquiry. Advisory Action, page 4. Applicant acknowledges this distinction and does not rely on the absence of the transformation from the prior art to establish that the step is unconventional. Rather, the specific transformation recited in the amended claims, converting yes- no observation entries into combined entries that include a number of observations for each type of observation data organized by date, is an unconventional application of data transformation in the context of machine-learning-based safety incident prediction. The Examiner has alleged that this transformation "represents a form of data normalization and aggregation" that is "a well- understood technique in the field of data processing." Advisory Action, page 4. However, the specific transformation claimed here is not generic data aggregation. The claims recite a particular transformation that specifically addresses the structural mismatch between per-worker observation data and per-date incident data in safety prediction systems, converting yes-no entries from multiple workers into date-aggregated entries that include a number of observations for each type to enable the machine-learning model to correlate observation patterns with incident occurrences at the date level. This specific combination of elements for multi-worker safety observation data in the context of incident prediction amounts to significantly more than the judicial exception”.
The Examiner respectfully disagrees.
With respect to the argument the Examiner notes that at the outset, the Examiner acknowledges Applicant's concession that the Step 2B analysis is distinct from the novelty inquiry, and that Applicant is not relying solely on the absence of the transformation from the prior art to establish that it is unconventional. The Examiner has carefully considered Applicant's affirmative argument that the specific combination of elements converting yes-no observation entries from multiple workers into date-aggregated combined entries that include a number of observations for each type of observation data, in the context of machine-learning-based safety incident prediction amounts to significantly more than the abstract idea. This argument is not persuasive.
The Step 2B analysis requires that the additional elements, considered individually and in combination, amount to significantly more than the judicial exception itself. MPEP 2106.05; Alice Corp. Pty. Ltd. v. CLS Bank Int'l. An inventive concept "cannot be furnished by the unpatentable abstract idea itself," and the search for an inventive concept at Step 2B is distinct from both novelty and obviousness analyses. Synopsys, Inc. v. Mentor Graphics Corp., ("a claim for a new abstract idea is still an abstract idea"). As a threshold matter, the Examiner's characterization of the combining step as a form of data normalization and aggregation that is well-understood in the data processing arts is supported by the specification itself, which is a form of evidence the Office may properly rely upon under MPEP 2106.07(a), subsection III. The specification at paragraph [0049] describes the combining/compression step in entirely matter-of-fact terms presenting it as one organizational technique among others that system 10 "can use" for "more optimal use in training" without identifying it as a novel, unconventional, or technically groundbreaking technique. The specification provides no indication that this type of data aggregation would be surprising or unexpected to a person of ordinary skill in the art, no discussion of why it differs from known data preprocessing methods, and no explanation of any technical difficulty that had to be overcome to implement it. This is precisely the type of disclosure that the MPEP identifies as insufficient to demonstrate unconventionality a bare description of a technique without the "detail necessary to be apparent to a person of ordinary skill in the art" as a technical departure from the conventional. MPEP 2106.07(a), subsection III(A); December 5, 2025 Memorandum from the Deputy Commissioner for Patents.
Applicant's recharacterization of the combining step as specifically addressing a "structural mismatch between per-worker observation data and per-date incident data in safety prediction systems" does not transform the nature of the technique itself. The structural alignment of input data granularity with output data granularity for purposes of statistical modeling for example, aggregating individual observations to a date-level unit of analysis before correlating with date-level incident outcomes is a standard data preprocessing consideration in the field of machine learning and statistical modeling, not an unconventional technical advance. Practitioners in the data science and machine-learning arts routinely perform analogous data-level alignment and aggregation operations as a standard part of feature engineering and data preparation pipelines. The fact that this technique is applied in the specific context of multi-worker worksite safety observation data does not render the technique itself unconventional; applying a known data aggregation technique to a new domain or dataset does not supply the inventive concept required under Step 2B. Mayo Collaborative Servs. v. Prometheus Labs., Inc.; SAP Am., Inc. v. InvestPic, LLC. Furthermore, while the Examiner acknowledges that a combination of individually conventional elements can in certain circumstances amount to significantly more as illustrated by BASCOM Global Internet Servs. v. AT&T Mobility, LLC, the BASCOM principle requires that the combination itself be non-conventional and non-generic, not merely that it applies known techniques to a new field. Here, the combination of receiving multi-worker observation data, aggregating it by date, and feeding it into a generic machine-learning model for safety incident prediction reflects a straightforward application of conventional data collection, conventional data aggregation, and conventional machine-learning techniques to a new subject-matter domain. This is precisely the type of application of established methods to a new field that the courts have consistently found ineligible, and the combination of these elements does not produce a non-conventional arrangement that amounts to significantly more than the judicial exception. The rejection is therefore maintained.
The Applicant argues on pages 15-16 that “Regarding the dependent claims, the Examiner has characterized claims 5, 6, 11, 16, 19, and 20 as further limiting data analysis, claims 2-4, 9, 13-15, and 17 as reciting insignificant extra- solution activity, and claims 8 and 10 as reciting non-functional data labels. Final Office Action, pages 9-10. Because independent claims 1, 12, and 18 are patent-eligible for the reasons stated above, the dependent claims that depend from those eligible independent claims are likewise eligible. Moreover, the dependent claims add further technical specificity that reinforces integration into a practical application. Claims 5, 6, 16, 19, and 20 recite specific machine-learning techniques, including random forest, extreme gradient boosting, and time series forecasting, that define how the machine-learning model operates on the transformed data structure. Claims 2-4, 9,13-15, and 17 recite additional data types, including worker data and weather data, that enrich the inputs to the machine-learning model and further define the technical data processing architecture. Claims 8, 10, and 11 recite severity parameters and predictive feature storage that specify how the system processes and utilizes incident data. These additional limitations are not mere field-of-use restrictions or insignificant extra-solution activity, but instead define specific technical components and operations that work together with the data transformation recited in the independent claims to provide a complete technical solution for machine-learning-based safety incident prediction”.
The Examiner respectfully disagrees.
With respect to the argument the Examiner notes that while the Applicant's argument for the dependent claims rests entirely on the premise that independent claims 1, 12, and 18 are patent-eligible for the reasons stated in the preceding arguments. As set forth in the Examiner's responses to Arguments 1 through 5, that premise is incorrect. The independent claims remain directed to an abstract idea and fail to integrate the judicial exception into a practical application at Step 2A, Prong Two, and do not provide an inventive concept at Step 2B. Accordingly, the dependent claims cannot derive eligibility by virtue of dependence from the independent claims alone. Each claim must be independently evaluated for compliance with 35 U.S.C. 101 based on its particular elements, and Applicant's blanket eligibility argument by dependency is legally insufficient. MPEP 2106.07(a) ("Claims should not be judged to automatically stand or fall with similar claims in an application"). The Examiner has independently evaluated each group of dependent claims and finds that none of the additional limitations they recite supply either integration into a practical application at Step 2A, Prong Two, or an inventive concept at Step 2B.
Regarding claims 5, 6, 16, 19, and 20, which recite that the machine-learning model uses a random forest technique, an extreme gradient boosting technique, or a time series forecasting technique, Applicant argues that these limitations "define how the machine-learning model operates on the transformed data structure" and therefore provide technical specificity reinforcing integration into a practical application. This argument is not persuasive. The recitation of specific, named conventional machine-learning algorithms does not transform otherwise abstract subject matter into patent-eligible subject matter. Random forest, extreme gradient boosting (XGBoost), and time series forecasting are each expressly identified in the specification at paragraph [0052] as conventional techniques that the machine-learning model "can" employ, listed alongside numerous other conventional techniques including support vector machines, discriminant analysis, naïve Bayes, nearest neighbor, linear regression, GLM, SVR, GPR, neural networks, and logistic regression. Specifying which among a list of known, conventional algorithms is used does not render a claim patent-eligible, because selecting a particular conventional mathematical tool to execute an abstract idea does not itself provide significantly more than the abstract idea. SAP Am., Inc. v. InvestPic, LLC, (specifying particular mathematical techniques for performing a statistical analysis does not supply the inventive concept); Electric Power Group, LLC v. Alstom S.A. Furthermore, the specification's matter-of-fact identification of these algorithms as known options without identifying any technical problem with any of them or any unconventional manner of their deployment confirms that their recitation reflects well-understood, routine, conventional activity in the machine-learning arts, satisfying the Examiner's evidentiary burden under MPEP 2106.07(a), subsection III(A). Applicant's characterization of these limitations as defining "how the machine-learning model operates on the transformed data structure" is a functional restatement of what any conventional machine-learning algorithm does it operates on whatever inputs it is given and does not establish that any of these algorithms is being deployed in an unconventional manner.
Regarding claims 2, 3, 4, 9, 13, 14, 15, and 17, Applicant argues that these claims "enrich the inputs to the machine-learning model" and "further define the technical data processing architecture" by adding worker data, weather data, additional data association steps, memory storage, and multi-project prediction. The Examiner maintains that these limitations constitute insignificant extra-solution activity and field-of-use limitations that do not provide significantly more than the abstract idea. The receiving steps for worker data and weather data like the receiving steps in the independent claims are pre-solution data-gathering steps necessary to any use of the abstract idea of predicting safety incidents from the identified categories of data. OIP Techs., Inc. v. Amazon.com, Inc. The specification at paragraphs [0018]–[0019] describes worker data and weather data as routine informational inputs daily headcounts, ages, experience levels, and weather readings collected in the same routine manner as the observation data. The data-linking steps in claims 3, 4, and 14 associating observation, worker, and weather data with incident data by linking dates are date-alignment operations that are incidental to and necessary for any use of the abstract idea of correlating observational inputs with incident outputs, not an additional element that independently provides an inventive concept. The multi-project prediction of claim 9, applying previously identified predictive features to a second remote project, is simply a further application of the core abstract prediction to a new dataset, which does not supply significantly more. The memory storage limitation of claim 17 storing project data, observation data, and incident data in memory is a generic computer function that courts have consistently recognized as well-understood, routine, and conventional activity. Versata Dev. Grp. v. SAP Am., Inc.; MPEP 2106.05(d), subsection II.
Regarding claims 8, 10, and 11, Applicant argues that severity parameters and predictive feature storage "specify how the system processes and utilizes incident data" beyond mere field-of-use restrictions. The Examiner disagrees. Claim 8's limitation that the identified predictive features "are saved by the incident predictor as predictive features" specifies a label and intended use for stored data a non-functional data label that describes only the category of the stored information, not any particular technical mechanism by which it is stored, retrieved, or utilized. Claims 10 and 11's specification of OSHA-recordable, DART, and lost-time incident severity categories identifies known, officially-defined real-world administrative classifications for the value of the "severity" data type and does not add any technical content to the claims. The Examiner further notes that adding a further abstract prediction predicting the severity of the future incident in claim 11 to a claim already directed to an abstract prediction of occurrence does not supply significantly more; it simply extends the scope of the abstract idea's output. RecogniCorp, LLC v. Nintendo Co., ("Adding one abstract idea to another abstract idea does not render the claim non-abstract"). The dependent claims accordingly do not provide any limitations that integrate the judicial exception into a practical application or supply an inventive concept, and the rejection is maintained as to all pending claims 1–6 and 8–20.
The Applicant argues on pages 17-18 that “The Examiner has also cited Samson's disclosure that "a CSV file can be produced by the system." Advisory Action, page 5. However, the generation of a CSV file for reporting purposes is fundamentally different from the claimed transformation. A CSV file is merely a file format for storing tabular data. The production of a CSV file does not teach or suggest transforming yes-no observation entries into combined entries that include a number of observations for each type organized by date to reduce entries for machine-learning processing”.
The Examiner further argues that paragraphs [0021] and [0026] of Samson should be read together, asserting that the matrix-organized data from paragraph [0021] "flows into the AI/ML engine as input" described in paragraph [0026], and that this establishes the combining step. Advisory Action, page 6. Applicant respectfully disagrees. Even reading these paragraphs together, Samson does not disclose transforming yes-no entries into combined entries that include a number of observations for each type. Samson's AI module described in paragraph [0026] "collects data from the other modules" and performs predictive analysis, but there is no disclosure that the matrix or CSV from paragraph [0021] involves transforming individual yes-no observation entries into aggregated entries that include a number of observations for each type. Samson's matrix is built from categorized observations, housekeeping, dangerous work practice, dangerous equipment, best practice, and non-compliant procedures, which are categories assigned to individual observations, not a transformation of yes-no entries into observation counts as recited in the amended claims. See Samson, paragraph [0021]”.
The Examiner respectfully disagrees.
With respect to the argument the Examiner notes that Applicant's broader argument that Samson fails entirely to teach or suggest the claimed combining limitation is not persuasive, and the rejection under 35 U.S.C. 103 is maintained for the following reasons. The Examiner concedes that the generation of a CSV file for reporting purposes, standing alone, does not disclose the specific transformation of converting individual yes-no observation entries from multiple workers into combined entries that include a number of observations for each type of observation data organized by date.
However, turning to Applicant's argument that even reading paragraphs [0021] and [0026] of Samson together fails to establish the combining limitation, the Examiner respectfully disagrees. Applicant argues that Samson's matrix described in paragraph [0021] is composed of "categories assigned to individual observations" specifically, housekeeping, dangerous work practice, dangerous equipment, best practice, and non-compliant procedures and that assigning categorical labels to individual observations is not the same as transforming yes-no entries from multiple workers into date-organized combined entries that include observation counts for each type. The Examiner acknowledges that Samson's paragraph [0021] does not expressly describe a step-by-step algorithmic process of converting binary per-worker entries into numerical aggregates organized by date in the precise language used in the amended claims. However, for purposes of a 35 U.S.C. 103 obviousness analysis, a reference need not disclose every element in the exact manner claimed; it is sufficient that the reference discloses subject matter that would render the claimed combination obvious to a person of ordinary skill in the art. KSR Int'l Co. v. Teleflex Inc. Samson's paragraph [0021] discloses that the system receives daily accident and incident reports and worksite assessments and creates a "matrix based risk evaluation with location specifics" from the reported data. The creation of a matrix from multiple daily observation reports each of which would individually contain categorical entries regarding the type of safety condition observed inherently involves organizing and structuring those individual per-observation entries into a consolidated matrix format. Paragraph [0026] further discloses that the artificial intelligence module "collects data from the other modules" and performs predictive analysis and problem solving on that collected data. Reading paragraphs [0021] and [0026] together, as the Examiner previously stated and continues to maintain, the matrix of organized, collected daily observation data from paragraph [0021] serves as the structured input that flows into the AI/ML module of paragraph [0026] for predictive analysis. While Samson does not use the precise terminology of "yes-no entries" being transformed into "combined entries that include a number of observations for each type," the underlying function of organizing individual categorical observation records from a worksite into a structured, consolidated matrix format for use in predictive AI analysis is what Samson discloses, and one of ordinary skill in the art would recognize that consolidating individual categorical observation records by date into a matrix would involve aggregating the counts of each category of observation occurring on each date which is the functional equivalent of the claimed combining step when applied to the same type of safety observation data. The Examiner maintains that this disclosure, in combination with Gattie and Grant, renders the claimed combining step obvious, and the rejection is accordingly maintained.
Furthermore, even accepting Applicant's narrower reading of Samson for the sake of argument, the rejection does not depend solely on Samson providing an explicit written-out disclosure of every operational detail of the claimed combining step. Under KSR, a combination of familiar elements according to known methods that yields predictable results is obvious, and the Examiner's position is that combining the known technique of aggregating categorical observation records into date-organized count-based matrices which is a standard and well-known data organization technique in the safety management and data analytics arts as evidenced by Samson's matrix-based risk evaluation with the predictive safety incident analysis systems of Gattie and Grant would have been obvious to one of ordinary skill in the art. The rejection under 35 U.S.C. 103 is therefore maintained.
The Applicant argues on pages 18-19 that “Furthermore, the Examiner's motivation to combine analysis does not provide adequate reasoning for why one of ordinary skill in the art would have been motivated to modify Gattie's system to include the specific transformation recited in the amended claims. In the Advisory Action, the Examiner provided a more detailed motivation analysis, alleging that Gattie's system creates a "data management problem" because "when multiple workers submit separate safety observation forms on the same date each containing binary categorical entries, . . . the analytics database accumulates one row per observation per worker per date" and that "Samson directly addresses this exact data management need." Advisory Action, page 7. This reconstruction is based on hindsight. Gattie does not disclose or identify any such data management problem. Gattie is silent as to any inefficiency in its data structure or any need to transform per-worker observation entries into date-aggregated entries. The Examiner has imputed a problem to Gattie that Gattie itself does not recognize, and then used Samson to solve that imputed problem. Moreover, even if such a problem existed, Samson does not disclose the specific transformation recited in the amended claims, converting yes-no entries into combined entries that include a number of observations for each type of observation data organized by date. Samson's matrix-based risk evaluation organizes categorized observations (housekeeping, dangerous work practice, dangerous equipment, etc.) into a risk assessment structure, but does not disclose transforming individual yes- no observation entries into aggregated entries that include a number of observations for each type. The Examiner has also alleged that Grant paragraph [0019] provides independent motivation to combine, citing Grant's disclosure that "using raw data fields collected from the workplace, a new variable was created for the standard deviation of the worksite's time to resolve a hazard" and that "active and passive data points can be combined to reflect dynamic data points." Advisory Action, page 7. Applicant respectfully submits that Grant's disclosure is fundamentally different from the claimed transformation. Grant paragraph [0019] describes deriving a statistical metric (standard deviation) from timing data about hazard resolution, not transforming yes-no observation entries into combined entries that include a number of observations for each type organized by date. Similarly, Grant's "combining" of "active and passive data points to reflect dynamic data points" refers to integrating different sources of real-time data to create a Risk Score, not the specific structural transformation of converting per-worker yes-no entries into date-aggregated count entries as recited in the amended claims. The transformation claimed is fundamentally different in kind from what Grant discloses. The claims recite transforming the format and granularity of observation data from yes-no entries to entries that include a number of observations for each type, whereas Grant describes deriving derivative statistical variables from existing data fields. Grant therefore does not provide independent motivation to perform the specific transformation recited in the amended claims”.
The Examiner respectfully disagrees.
With respect to the argument the Examiner notes that the amended claims would have been obvious to a person of ordinary skill in the art under KSR Int'l Co. v. Teleflex Inc., for reasons supported by the references themselves rather than by applicant's own disclosure.
The Examiner maintains that a person of ordinary skill in the art of construction safety analytics and machine-learning-based predictive modeling having familiarity with the systems of Gattie, Grant, and Samson would have been motivated to aggregate and consolidate multi-worker daily observation records into date-organized summary entries without needing to identify any explicit "data management problem" in Gattie. The motivation arises independently from the nature of the predictive modeling task itself and from the teachings of the references in combination. Gattie at paragraphs [0003], [0005], and [0044] discloses a system that collects safety-related data from construction projects and applies analytics to drive decision-making regarding safety outcomes. Grant at paragraphs [0018]–[0019] and [0077] discloses a workplace risk determination and scoring system that receives observation data from multiple observers and uses machine-learning models including random forest techniques to predict incident likelihood. Critically, Grant at paragraph [0019] itself discloses that in building a predictive model, "using raw data fields collected from the workplace, a new variable was created" meaning that the process of feature engineering, including the derivation of new variables from raw collected data records, is expressly taught by Grant as a step in building the predictive model. The derivation of new, transformed variables from raw multi-observer data is precisely what a person of ordinary skill in the art would understand to be necessary when building a machine-learning model that correlates multiple workers' daily observations with date-specific incident outcomes, because incident data is inherently organized at the date level while per-worker observation data is organized at the individual-observation level. The motivation to align these levels of granularity does not derive from applicant's disclosure but from the fundamental technical requirements of machine-learning model construction that Grant itself teaches and that a person of ordinary skill in the art would recognize. KSR, 550 U.S. at 418 ("a person of ordinary skill is also a person of ordinary creativity, not an automaton").
Regarding Applicant's argument that Grant's paragraph [0019] disclosure of deriving "standard deviation of the worksite's time to resolve a hazard" and combining "active and passive data points to reflect dynamic data points" is fundamentally different in kind from the claimed transformation, the Examiner agrees that Grant paragraph [0019] does not literally disclose the specific step of converting binary yes-no observation entries into date-organized combined entries that include observation counts for each type. The Examiner does not rely on Grant paragraph [0019] as an express disclosure of the specific combining transformation. Rather, Grant paragraph [0019] is relevant to the motivation analysis for two reasons: first, it expressly teaches the concept of creating new derived variables from raw collected data fields as part of building a predictive safety model, which is the same underlying data transformation concept at issue in the claimed combining step; and second, Grant's disclosure that "active and passive data points can be combined to reflect dynamic data points" establishes that the concept of combining multiple types of collected data points into unified summary representations for use in predictive modeling was well within the knowledge of a person of ordinary skill in the art and was recognized as a standard step in building such models. The Examiner maintains that a person of ordinary skill in the art, faced with the task of combining Gattie's construction safety analytics system with Grant's multi-observer predictive risk scoring model, would have recognized without the need for applicant's disclosure that aligning the granularity of multi-worker observation data with the date-level granularity of incident data through aggregation is a standard and predictable data preparation step in machine-learning model construction, and would have been motivated to implement such aggregation based on the combination of Gattie's data collection framework, Grant's feature engineering teachings, and Samson's matrix-based organization of categorical safety observation records. The rejection under 35 U.S.C. 103 is accordingly maintained, with the motivation to combine analysis refined as set forth herein to address Applicant's hindsight objection.
The remaining Applicant's arguments filed 11 August 2026 have been fully considered but they are moot in view of new grounds of rejection as necessitated by amendment.
Claim Rejections - 35 USC 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-6, and 8-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The claims recite mathematical concepts and, in the alternative as to certain limitations, mental processes: tallying worksite safety observations by date, pairing the tallies with incident outcomes, fitting a statistical model to that data by weighting features with coefficients, and calculating the likelihood of a future unsafe incident. This judicial exception is not integrated into a practical application because the additional elements (receiving data, transmitting the result, and generic computer components including a machine-learning model used as a tool) amount to insignificant extra-solution activity, mere instructions to apply the exception on a computer, and a field-of-use limitation, and do not reflect an improvement to the functioning of a computer or to any other technology or technical field. The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements, individually and in combination, are generic computer implementation and well-understood, routine, conventional data gathering and output, as supported by the evidence identified below.
STEP 1
Regarding Step 1 of the Subject Matter Eligibility Test for Products and Processes, claims 1-6, 8-11, and 18-20 are directed to a method (process), and claims 12-17 are directed to a system (machine). Therefore, the claims fall within the statutory categories of invention.
STEP 2A Prong One
The claims recite an abstract idea. Specifically, independent claim 1 recites the following limitations (claim 12 recites the same operations as functions of the data linking module and incident predictor):
• (a) “combining each type of the first observation data from the multiple workers that have the same date of the observation, wherein the combining transforms entries of the first observation data from yes-no entries into combined entries that include a number of observations for each type of the first observation data organized by the date of the observation, thereby reducing a number of overall entries of the first observation data”;
• (b) “generating a training data set from the combined entries of the first observation data by linking the date of the observation of each combined entry to the date of the reported unsafe incident of the first incident data and by categorizing each combined entry that is not linked to a date of a reported unsafe incident as being associated with no unsafe incident”;
• (c) training with the training data set, “wherein the training comprises identifying … features that are most indicative of the occurrence of the reported unsafe incident and adjusting coefficients associated with the identified features to increase or decrease a weighting of each identified feature according to an importance of the identified feature to the occurrence of the reported unsafe incident”; and
• (d) “from features that include at least one of the types of first observation data and the types of first project data that are most indicative of the occurrence of the reported unsafe incident, predicting … the likelihood of an occurrence of a first future unsafe incident at the first project.”
Independent claim 18 recites limitation (a); “extracting … features from the types of project data and the types of observation data that are most indicative of the occurrence of the future unsafe incident,” including “associating the second observation data with the first incident data by linking a date of the observation data to a date of the unsafe incidents” and “identifying predictive features … that are most indicative of an occurrence of the unsafe incident at the second project”; “adjusting coefficients associated with the identified predictive features to increase or decrease a weighting of each identified predictive feature according to an importance of the identified predictive feature”; and “predicting … the likelihood of the future unsafe incident occurring at a first worksite of a first project.” Claim 18 does not recite limitation (b).
Abstract Idea Grouping Analysis
Mathematical Concepts (primary basis)
These limitations recite mathematical calculations and mathematical relationships. A mathematical concept need not be expressed in symbols; it may be described in words, and a claim need not use the word “calculating.” See MPEP 2106.04(a)(2), subsection I.
Limitation (a) converts individual yes-no entries into “a number of observations for each type” for each date. That is counting, a mathematical calculation, and it organizes information through a mathematical correlation. See Digitech Image Techs., LLC v. Elecs. for Imaging, Inc. The stated reduction in the number of entries is the arithmetic consequence of that tally.
Limitation (b) pairs each date’s tallies with an outcome value (unsafe incident or no unsafe incident). This sets forth the mathematical relationship between the input variables and the output variable of the statistical model. The specification describes the incident data as “the dependent variable(s)” the system is aimed at predicting (paragraph [0034]) and includes “a variable indicating if no unsafe incident occurred … on a particular date” (paragraph [0035]).
Limitation (c) recites “adjusting coefficients” to “increase or decrease a weighting” of each feature according to its importance. This describes, in words, a numerical calculation performed on the parameters of a statistical model. The specification describes it the same way: the model weights “the importance of each input using coefficients that are increased and/or decreased to refine the accuracy of the prediction” (paragraphs [0052], [0069]). This limitation therefore does more than involve mathematics. Unlike the bare “training the neural network” limitation of Example 39, it sets forth the calculation itself, as did the training limitation of claim 2 of Example 47. See August 4, 2025 Memorandum, section II.A; July 2024 Subject Matter Eligibility Examples, Example 47. This is also consistent with Ex Parte Desjardins, Appeal No. 2024-000567 (PTAB Sept. 26, 2025) (precedential), in which the finding that claims reciting adjustment of model parameter values recited a mathematical concept was upheld at Prong One. See December 5, 2025 Memorandum.
Limitation (d), and the identifying and extracting of “most indicative” features, recite statistical evaluation of data to produce a probability. The specification describes the output as a percentage or “a value between 0 and 1” (paragraph [0056]). See SAP Am., Inc. v. InvestPic, LLC; In re Bd. of Trs. of Leland Stanford Junior Univ.
Mental Processes (independent alternative basis for limitations (b) and (d) and the feature-identifying limitations)
In the alternative, limitation (b), limitation (d), and the identifying of indicative features, under their broadest reasonable interpretation, cover evaluations and judgments that can practically be performed in the human mind or with pen and paper. Limitation (b) compares the date of each date-level entry with incident dates and labels the entries that have no match; the claims recite no number of entries and require only multiple workers. Identifying which kinds of observations are most indicative of past incidents and estimating the likelihood of a future incident is the evaluation a safety professional makes when reviewing worksite records. The claims do not limit how the identifying or predicting is performed. See MPEP 2106.04(a)(2), subsection III; Example 47, claim 2 (discretizing training data, and detecting and analyzing “using the trained ANN,” recite mental processes). The recitation that these steps are performed by a machine-learning model or processor does not take them out of this grouping; those components are addressed below as additional elements.
Consistent with MPEP 2106.04(a)(2), subsection III(A), and the August 4, 2025 Memorandum, the Examiner does not rely on the mental-process grouping for limitation (a) or for the coefficient-adjusting portion of limitation (c). The mathematical concepts grouping is the primary basis and stands independently. The groupings are not mutually exclusive, and the limitations are considered together as a single abstract idea for the remainder of the analysis. See MPEP 2106.04, subsection II.B.
STEP 2A Prong Two
Identification of Additional Elements
The claims recite the following additional elements beyond the identified abstract idea:
• receiving first project data, first observation data “from multiple workers,” and first incident data of the enumerated types (claim 1; recited as data of the system in claim 12); receiving first project data and first observation data, and receiving second project data, second observation data, and second incident data (claim 18);
• “transmitting the predicted likelihood … to a computing device associated with the first worksite” (claims 1, 12, and 18);
• “a machine-learning model of an incident predictor” that performs the training, identifying, extracting, and predicting (claims 1, 12, and 18);
• “a data linking module, which includes at least one computer processor,” and “an incident predictor, which includes the at least one computer processor” (claim 12); and
• performing the combining “concurrently,” to the extent that term requires computer processing (claims 1, 12, and 18).
Analysis of Additional Elements
Particular Machine and Particular Transformation (MPEP 2106.05(b) and (c))
The claims do not recite a particular machine. The incident predictor may be a cloud-based system, server, smartphone, tablet, or mobile application (paragraphs [0040]-[0042]); the processor may be “any or more than one of a processor, a microprocessor, a controller, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other equivalent discrete or integrated logic circuitry” (paragraph [0043]); and the data linking module “can include any software and/or hardware” (paragraph [0047]). The “computing device associated with the first worksite” is not further limited. The claims also do not transform a particular article to a different state or thing; reorganizing data records is not such a transformation.
Mere Instructions to Apply the Exception (MPEP 2106.05(f))
The machine-learning model, incident predictor, data linking module, and computer processor are recited at a high level of generality as the tools on which the abstract idea is performed. The claims do not recite any particular model structure or training technique, and the specification states that the model may use any of a long list of classification, regression, clustering, and other techniques (paragraph [0052]). Performing the combining “concurrently” by a processor is likewise a generic computer function. These elements amount to no more than instructions to implement the abstract idea on a generic computer, which is equivalent to adding the words “apply it.” See Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 573 U.S. 208, 225-26 (2014); Example 47, claim 2 (“using the trained ANN”).
Insignificant Extra-Solution Activity (MPEP 2106.05(g))
The receiving limitations are mere data gathering. They obtain the information on which the calculations operate, are recited without any limit on how the data is received, and are required by every use of the abstract idea. The transmitting limitation is mere output of the result. It recites no particular manner of transmission, and no action is taken on the prediction; the specification describes it only as providing the prediction “to the necessary persons, computers, etc. via wired or wireless communication” (paragraph [0024]). That the transmitted data is the output of the trained model does not make the step significant. See Example 47, claim 2 (“outputting the anomaly data from the trained ANN” is insignificant extra-solution activity); compare Example 47, claim 3 (dropping packets and blocking traffic reflected a disclosed improvement); Apple, Inc. v. Ameranth, Inc.; Electric Power Group, LLC v. Alstom S.A.
Field of Use (MPEP 2106.05(h))
The enumerated types of project, observation, and incident data, and the limitation to a worksite of a project, restrict the abstract idea to worksite safety data. Selecting a particular type of data to be analyzed and limiting the analysis to a particular field does not integrate the exception. See Electric Power Group, 830 F.3d at 1354-55; Parker v. Flook, 437 U.S. 584 (1978).
Alternative Consideration of Limitation (b)
Even if limitation (b) were treated as an additional element instead of part of the abstract idea, the result at this step would be the same. It prepares and labels the data supplied to the model, is performed before training, does not alter how the model or computer operates, and reflects no disclosed improvement to technology. It would be insignificant pre-solution activity under MPEP 2106.05(g).
Each limitation has been considered individually, and the claim has been considered as a whole without disregarding the specific requirements of the claims. See MPEP 2106.05(a); December 5, 2025 Memorandum. As an ordered combination, the claims gather data, tally it, label it, fit a model to it, compute a likelihood, and send the result, which is the ordinary sequence of a supervised statistical prediction carried out on generic computer components. The specification describes no unconventional arrangement and states that the steps can be performed “in any order” (paragraph [0058]). The additional elements do not impose a meaningful limit on the abstract idea. Accordingly, the claims are directed to an abstract idea.
STEP 2B
As discussed with respect to Step 2A Prong Two, the machine-learning model, incident predictor, data linking module, and processor amount to no more than mere instructions to apply the exception using generic computer components. The same analysis applies in Step 2B; mere instructions to apply an exception using generic computer components cannot provide an inventive concept. See MPEP 2106.05(f). The field-of-use limitations likewise cannot provide an inventive concept. See MPEP 2106.05(h).
The inventive concept must be furnished by an element or combination of elements recited in addition to the judicial exception; it cannot be furnished by the exception itself. See MPEP 2106.05, subsection I; Synopsys, Inc. v. Mentor Graphics Corp., (“a claim for a new abstract idea is still an abstract idea”). Limitations (a) through (d) are part of the abstract idea. The Examiner therefore makes no finding that they are well-understood, routine, or conventional, and whether they are new is not relevant to this inquiry.
Well-Understood, Routine, Conventional Activity Analysis
The extra-solution activity identified at Prong Two has been re-evaluated. The following additional elements, individually and in combination, are well-understood, routine, and conventional, as supported under MPEP 2106.07(a), subsection III:
• Receiving the project, observation, and incident data, and transmitting the predicted likelihood to a computing device. These limitations are recited at a high level of generality. The courts have recognized receiving or transmitting data over a network as well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II, citing Intellectual Ventures I LLC v. Symantec Corp.; TLI Communications LLC v. AV Auto. LLC; OIP Techs., Inc. v. Amazon.com, Inc.; buySAFE, Inc. v. Google, Inc. See also Example 47, claim 2.
• The computer processor, data linking module, and incident predictor. The specification describes these components generically and without particulars, as any processor or equivalent circuitry, any computing device or mobile application, and “any software and/or hardware” (paragraphs [0041]-[0043], [0047]). This description indicates that the components are sufficiently well-known that their particulars need not be described to satisfy 35 U.S.C. 112(a) (subsection III(A)).
The Examiner does not rely on any silence of the specification, and does not rely on the prior art references of record, for these findings.
Considered as an ordered combination, the additional elements receive data, carry out the abstract idea on generic computer components, and transmit the result. The combination adds nothing that is not already present when the elements are considered individually, and it is not a non-conventional and non-generic arrangement. Compare BASCOM Global Internet Servs., Inc. v. AT&T Mobility LLC, 827 F.3d 1341, 1350 (Fed. Cir. 2016). The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claims are not patent eligible.
Dependent Claims Analysis
Each dependent claim has been considered individually and in combination with the claim from which it depends. The dependent claims do not add limitations that integrate the judicial exception into a practical application or provide an inventive concept:
• Claims 2 and 13: These claims add receiving worker data and weather data of enumerated types and basing the prediction on features that include those types. The receiving is additional data gathering recited at a high level of generality (MPEP §2106.05(g), 2106.05(d), subsection II), the enumerated types are a field-of-use selection of data (MPEP 2106.05(h)), and the prediction remains part of the abstract idea.
• Claim 3: This claim adds associating the observation, worker, and weather data with the incident data by linking dates, and identifying the most indicative features by the incident predictor. Date-linking and feature identification are part of the abstract idea for the reasons given for limitations (b) through (d). The incident predictor is a generic component used as a tool (MPEP 2106.05(f)).
• Claims 4 and 14: These claims add combining incident data having the same date into one entry, associating observation data with incident data by date, and identifying or determining the most indicative features or types of data. The combining is a further aggregation of the kind recited in limitation (a), and the remaining limitations are part of the abstract idea. The data linking module is a generic component (MPEP 2106.05(f)).
• Claims 5, 16, and 19: These claims add determining an importance of each feature and weighting the features accordingly in predicting the likelihood. Determining importance and applying weights are mathematical calculations and further define the abstract idea.
• Claims 6 and 20: These claims recite that the model uses a random forest technique, an extreme gradient boosting technique, or a time series forecasting technique. Each is a mathematical technique referred to by name, and the claims recite no detail of its use. Naming the mathematical technique further defines the abstract idea. See August 4, 2025 Memorandum, section II.A; Example 47, claim 2. The specification lists these among many interchangeable techniques (paragraph [0052]).
• Claim 8: This claim adds that the most indicative types of features are saved by the incident predictor as predictive features. Saving the result of the analysis is insignificant extra-solution activity (MPEP 2106.05(g)). Storing and retrieving information in memory is well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II, citing Versata Dev. Group, Inc. v. SAP Am., Inc.; OIP Techs.
• Claim 9: This claim adds receiving project and observation data from a second, physically remote project and predicting the likelihood of an incident there from the predictive features. The receiving is further data gathering, and the second prediction applies the same abstract idea to another data set.
• Claim 10: This claim specifies that the severity is an OSHA-recordable, DART, or lost-time incident. It describes the content of the data and is a selection of a type of data within the field of use (MPEP §2106.05(g), 2106.05(h)).
• Claim 11: This claim adds predicting, by the incident predictor, the severity of the future unsafe incident. This is a further statistical evaluation of the same kind as limitation (d). Adding one abstract idea to another does not render the claim non-abstract. See RecogniCorp, LLC v. Nintendo Co.
• Claim 15: This claim adds that the machine-learning model is configured to extract the features it identifies as most indicative. Feature extraction is part of the abstract idea, and the model is recited generically as the tool that performs it (MPEP 2106.05(f)).
• Claim 17: This claim adds memory in communication with the incident predictor for storing the project, observation, and incident data. The memory is a generic component described as any volatile or non-volatile storage (paragraphs [0045]-[0046]), and storing information in memory is well-understood, routine, conventional activity. See MPEP 2106.05(d), subsection II, citing Versata.
For the foregoing reasons, claims 1-6 and 8-20 are rejected under 35 U.S.C. 101 as being directed to non-statutory subject matter.
Conclusion
The prior art made of record and not relied upon considered pertinent to Applicant’s disclosure.
Alsahlawai et al. (U.S. Patent 10,959,056 B1) discloses monitoring system for site safety and tracking.
Goel et al. (U.S. Patent Publication 2020/0117903 A1) discloses architecture, engineering and construction (AEC) construction safety risk analysis system and method for interactive visualization and capture.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to STEPHEN S SWARTZ whose telephone number is (571)270-7789. The examiner can normally be reached Mon-Fri 9:00 - 6:00.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Boswell Beth can be reached at 571 272-6737. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/S.S.S/Examiner, Art Unit 3625
/BETH V BOSWELL/Supervisory Patent Examiner, Art Unit 3625
Prosecution Insights