DCT
3:25-cv-09640
Adaptive Classification Tech LLC v. Arctera US LLC
Key Events
Complaint
I. Executive Summary and Procedural Information
- Parties & Counsel:
- Plaintiff: Adaptive Classification Technologies LLC (Texas)
- Defendant: Arctera US LLC (Delaware, PPOB California)
- Plaintiff’s Counsel: Bradford Black P.C.
- Case Identification: 3:25-cv-09640, N.D. Cal., 11/07/2025
- Venue Allegations: Venue is alleged to be proper in the Northern District of California because Defendant maintains its principal place of business in the district, has a regular and established place of business there, and has allegedly committed acts of infringement within the district.
- Core Dispute: Plaintiff alleges that Defendant’s Arctera eDiscovery Platform infringes a patent related to methods for determining when to terminate a technology-assisted review (TAR) of electronic documents.
- Technical Context: The technology operates within the electronic discovery (e-discovery) industry, where machine learning is used to classify massive volumes of digital documents for legal review, a process that represents a significant component of modern litigation costs.
- Key Procedural History: The complaint alleges that Plaintiff provided Defendant with notice of the patent-in-suit and its alleged infringement via a letter on October 8, 2025, approximately one month prior to filing the lawsuit.
Case Timeline
| Date | Event |
|---|---|
| 2015-06-19 | U.S. Patent No. 10,445,374 Priority Date |
| 2019-10-15 | U.S. Patent No. 10,445,374 Issues |
| 2025-10-08 | Plaintiff sends letter to Defendant alleging infringement |
| 2025-11-07 | Complaint Filed |
II. Technology and Patent(s)-in-Suit Analysis
- Patent Identification: U.S. Patent No. 10,445,374, "Systems and Methods for Conducting and Terminating a Technology-Assisted Review," issued October 15, 2019.
The Invention Explained
- Problem Addressed: In large-scale document reviews, such as e-discovery, it is prohibitively expensive to manually review every document. While TAR systems can prioritize documents, a critical problem is determining when the review is "complete"—that is, when a sufficiently high percentage (recall) of all relevant documents has been found, and the review can be defensibly stopped. (’374 Patent, col. 3:25-34, 56-65).
- The Patented Solution: The invention proposes a quality assurance method to terminate a TAR process. The method first requires identifying a "target set" of known relevant documents from the collection. ('374 Patent, col. 6:63-67). Then, a separate and "independent" classification process is executed to classify documents in the entire collection. ('374 Patent, col. 7:20-30). The process is terminated based on a comparison between the performance of this independent process and the pre-identified target set, such as when the process has successfully found a sufficient number of the target documents. ('374 Patent, FIG. 1, steps 1040, 1060). This provides a statistical basis for concluding that the process has achieved a desired level of recall with a certain probability. ('374 Patent, Abstract).
- Technical Importance: The claimed method provides a structured, statistically-grounded framework for stopping a TAR process, intended to give legal teams a defensible basis for asserting that their review was sufficiently thorough. (Compl. ¶39).
Key Claims at a Glance
- The complaint asserts infringement of at least independent claim 1. (Compl. ¶54).
- The essential elements of independent claim 1 include:
- Receiving an identification of a target set of documents in a document collection.
- Executing a classification process that uses a second, iterative search strategy that "does not distinguish between documents in the target set and documents in the document collection."
- Terminating the classification process based upon a comparison between the results of the second search strategy and a characteristic of the target set.
- Wherein the process achieves a target level of recall with a certain probability upon termination.
- The complaint does not explicitly reserve the right to assert dependent claims.
III. The Accused Instrumentality
Product Identification
- The Arctera eDiscovery Platform, specifically its "Transparent Predictive Coding" workflow. (Compl. ¶47).
Functionality and Market Context
- The Accused Instrumentality is a software platform designed for technology-assisted review in e-discovery. (Compl. ¶48). The "Transparent Predictive Coding" workflow allows users to train a machine learning model to classify documents. (Compl. ¶48). The process involves users first identifying an initial "Training Set" of documents. (Compl. ¶50). The system then enters an iterative "Train and Test" cycle, applying the classifier to the document population, and using a "Control Set" to test the model's accuracy. (Compl. ¶51; Compl. p. 13). The complaint cites Arctera's user documentation, which shows a multi-step workflow for setting up folders, training the model, and testing its performance. (Compl. p. 12). The workflow concludes when performance metrics, such as a recall objective, are met, at which point a "Best Prediction Rank Threshold" is applied to classify the remaining documents. (Compl. ¶52; Compl. p. 11).
- The platform is marketed as a tool to "defensibly reduce the time and cost of document review" by automatically classifying documents. (Compl. ¶48).
IV. Analysis of Infringement Allegations
U.S. Patent No. 10,445,374 Infringement Allegations
| Claim Element (from Independent Claim 1) | Alleged Infringing Functionality | Complaint Citation | Patent Citation |
|---|---|---|---|
| receive an identification of a target set of documents in a document collection... | The system receives an identification of a "Training Set" created by users who locate, review, and tag documents for responsiveness. | ¶50 | col. 6:63-7:14 |
| execute the classification process that enables training of a classifier using documents in the target set, wherein the classification process utilizes a second iterative search strategy which does not distinguish between documents in the target set and documents in the document collection to classify documents in the document collection; | The platform executes a "Train and Test" workflow, which trains a classifier based on the user-identified training set and then applies that classifier "uniformly across the document collection to assign prediction scores and classify documents, without using target-set membership as a factor." | ¶51 | col. 18:35-39 |
| and terminate the classification process based upon a comparison between the results of the second search strategy and a characteristic of the target set of documents... | The platform terminates the classification process when a "tested threshold from the Results By Outcome report meets the recall objective for the predictive tag." The complaint cites a screenshot of this report, which details metrics such as "Calculated Recall." | ¶52 | col. 18:40-42 |
| wherein the classification process achieves a target level of recall with a certain probability upon termination. | The complaint alleges this is met because the platform's termination condition—meeting a recall objective based on a tested threshold—results in achieving a target level of recall with a certain probability. | ¶52 | col. 18:42-45 |
Identified Points of Contention
- Scope Questions: The complaint alleges the user-created "Training Set" is the claimed "target set." (Compl. ¶50). However, the accused workflow also appears to use a separate "Control Set" for testing. (Compl. p. 12). This raises the question of whether the accused product's use of distinct sets for training and testing can be read on a claim that recites a single "target set" used as the basis for the termination comparison.
- Technical Questions: A central issue may be the interpretation of the limitation that the search strategy "does not distinguish between documents in the target set and documents in the document collection." (Compl. ¶51). The complaint alleges the accused classifier is applied "uniformly." (Compl. ¶51). A counterargument may be that a model trained on the target set (or training set) inherently "distinguishes" those documents because its internal parameters are derived from them, suggesting a potential mismatch with the patent's requirement for an independent search strategy.
V. Key Claim Terms for Construction
The Term: "target set of documents"
- Context and Importance: The definition is critical because the accused product's workflow involves multiple document sets, including a "Training Set," a "Control Set," and an "Initial Sample." (Compl. p. 12). Whether one or a combination of these meets the definition of "target set" will be central to the infringement analysis.
- Intrinsic Evidence for Interpretation:
- Evidence for a Broader Interpretation: The specification describes the target set simply as one created when "relevant documents are identified in a document collection." ('374 Patent, col. 6:63-65). This could support an argument that any pre-identified set of relevant documents used in the workflow qualifies.
- Evidence for a Narrower Interpretation: Claim 1 links the same target set to both training the classifier and the final termination comparison. The detailed description also describes termination as being based on when "a sufficient number of documents in the target set have been classified as relevant by the independent search strategy." ('374 Patent, col. 4:37-40). This language may support a narrower construction requiring a single set to be used for both purposes.
The Term: "second iterative search strategy which does not distinguish between documents in the target set and documents in the document collection"
- Context and Importance: This term defines the required level of independence between the classification process and the target set. Practitioners may focus on this term because it appears to be the primary technical distinction asserted by the inventors. The infringement question may turn on whether a process trained on the target set can still be said to "not distinguish."
- Intrinsic Evidence for Interpretation:
- Evidence for a Broader Interpretation: This could be interpreted to mean that the final classification rule is applied identically to every document, regardless of its origin, as the complaint alleges. (Compl. ¶51). Under this view, the training phase is distinct from the application phase.
- Evidence for a Narrower Interpretation: The patent states the classification process uses "an iterative search strategy that is independent of the strategy that identified the target set of documents." ('374 Patent, col. 4:24-26). This could support a requirement for greater separation, suggesting that using the target set to train the classifier inherently makes the strategy "distinguish" those documents from the rest of the collection.
VI. Other Allegations
- Indirect Infringement: The complaint alleges induced infringement under 35 U.S.C. § 271(b). (Compl. ¶55). It asserts that Arctera provides product documentation, such as the "Transparent Predictive Coding User Guide," which instructs customers on how to perform the allegedly infringing steps of creating training sets, running the iterative workflow, and using the results to terminate review. (Compl. ¶57).
- Willful Infringement: Willfulness is alleged based on Arctera's purported knowledge of the '374 Patent since at least October 8, 2025, the date it allegedly received a notice letter from the Plaintiff. (Compl. ¶56). The complaint alleges that Arctera continued to engage in infringing activities after receiving notice, demonstrating at least reckless disregard of Plaintiff's patent rights. (Compl. ¶¶58-59).
VII. Analyst’s Conclusion: Key Questions for the Case
- A core issue will be one of definitional scope: does the accused platform's workflow, which utilizes distinct "Training Sets" for model building and "Control Sets" for testing, meet the claim limitation of a single "target set" that serves as the basis for the termination comparison?
- A key technical question will be one of functional independence: can the accused classification process, which is explicitly trained using the user-identified "Training Set," be considered a "second iterative search strategy which does not distinguish" between those documents and the broader collection, as required by the patent?