DCT
1:25-cv-01802
Adaptive Classification Tech LLC v. Kldiscovery Ontrack LLC
Key Events
Complaint
I. Executive Summary and Procedural Information
- Parties & Counsel:
- Plaintiff: Adaptive Classification Technologies LLC (Texas)
- Defendant: KLDiscovery Ontrack, LLC (Delaware)
- Plaintiff’s Counsel: Bradford Black P.C.
- Case Identification: 1:25-cv-01802, W.D. Tex., 11/07/2025
- Venue Allegations: Plaintiff alleges venue is proper because Defendant maintains a regular and established place of business in the Western District of Texas and has committed the alleged acts of infringement within the district.
- Core Dispute: Plaintiff alleges that Defendant’s Nebula eDiscovery Platform infringes a patent related to methods for reliably terminating a technology-assisted document review process.
- Technical Context: The technology at issue involves machine learning algorithms, specifically Continuous Active Learning (CAL), used in the legal industry to automate the review of massive volumes of electronic documents for litigation.
- Key Procedural History: The complaint alleges that Defendant had pre-suit knowledge of the patent-in-suit as of July 14, 2025, via a notice letter, which forms the basis for Plaintiff's willfulness allegations.
Case Timeline
| Date | Event |
|---|---|
| 2015-06-19 | ’374 Patent Priority Date |
| 2019-10-15 | ’374 Patent Issue Date |
| 2025-07-14 | Alleged Pre-Suit Notice of Infringement |
| 2025-11-07 | Complaint Filing Date |
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 technology-assisted review (TAR), a key challenge is determining when to stop the review process while ensuring that a sufficiently high percentage of all relevant documents (i.e., high recall) has been found with a high degree of reliability (’374 Patent, col. 3:28-40, 61-65). The patent notes that conventional methods for measuring recall can be problematic due to "effort, bias, and imprecision associated with sampling" (’374 Patent, col. 3:45-46).
- The Patented Solution: The invention proposes a method to terminate a TAR process with statistical reliability. First, a "target set" of known relevant documents is created (e.g., by randomly sampling and reviewing documents) (’374 Patent, col. 6:53-65). Then, a separate, independent TAR process (e.g., Continuous Active Learning) is executed to classify the entire document collection. The process is terminated when this independent TAR process has successfully identified a predetermined portion of the documents in the initial "target set," thereby providing a reliable measure that a desired level of recall has been achieved across the whole collection (’374 Patent, col. 4:10-32; Fig. 1).
- Technical Importance: This approach aims to provide a provably reliable stopping point for TAR workflows, moving beyond subjective reviewer decisions or statistically biased control sets to create a defensible and efficient review process (’374 Patent, col. 6:31-34, 57-60).
- Key Claims at a Glance:
- The complaint asserts at least independent claim 1 (Compl. ¶55).
- Essential elements of claim 1 include:
- A system with a processor and memory.
- Receiving an identification of a "target set of documents" consisting of documents identified as relevant by a first search strategy.
- Executing a classification process using a second iterative search strategy which "does not distinguish between documents in the target set" and other documents when classifying.
- 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 KLDiscovery Nebula eDiscovery Platform (the "Accused Instrumentality"), specifically its "Predictive Coding & Advanced Text Analytics" and "Prioritized Review/CAL" features (Compl. ¶¶48-49).
- Functionality and Market Context:
- The Accused Instrumentality is a technology-assisted review system that uses machine learning to classify documents and prioritize them for human review (Compl. ¶49). The platform's workflow allegedly begins with reviewers coding an initial "seed set" of documents, which can be a random sample (Compl. ¶51). The system then uses a "Continuous Active Learning (CAL)" process, described as a "Train & Classify" loop, to analyze reviewer coding decisions, generate predictive scores for all documents, and prioritize the remaining documents for review (Compl. ¶¶51-52). The complaint includes a screenshot of the platform's user interface for initiating this "Train & Classify" step (Compl. p. 13).
- The platform allows users to set termination criteria using a "Control Set," where they can define a "Target Cut-off value" for recall, a "Confidence Level," and a "Target Error Margin" (Compl. ¶53). A screenshot of this "Control Set" configuration dialog is provided in the complaint (Compl. p. 17). The process allegedly terminates when a score threshold meets the user-defined recall objective (Compl. ¶53).
IV. Analysis of Infringement Allegations
- ’374 Patent 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, wherein the target set of documents consists of documents identified as relevant as part of a first search strategy... | The Nebula platform receives an identification of a "seed set," which consists of documents identified as relevant by reviewers during an initial search-and-review step, such as coding a random sample. | ¶51 | 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 & Classify" loop, which is a continuous active-learning process. This process allegedly applies a classifier uniformly across the entire collection to assign prediction scores to both coded and uncoded documents without using "target-set membership as a factor." | ¶52 | col. 7:20-29 |
| 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, wherein the classification process achieves a target level of recall with a certain probability upon termination. | The platform terminates the process when a score cutoff, tested against a "Control Set," meets a user-defined recall objective (e.g., 80%) at a specified confidence level (e.g., 95%) and margin of error. This is shown in a screenshot of the "Control Set" settings and performance report. | ¶53 | col. 7:39-56 |
- Identified Points of Contention:
- Scope Questions: A central dispute may be whether the Accused Instrumentality’s "seed set" (used for initial training) and its separate "Control Set" (used for performance measurement and termination) together meet the definition of the claimed "target set". The patent describes a single "target set" being used as the benchmark for termination, raising the question of whether the accused product's bifurcated use of a seed set and a control set falls within the claim's scope.
- Technical Questions: Does the Accused Instrumentality's termination logic, which allegedly compares model performance against a recall goal on a "Control Set" (Compl. p. 15), perform the specific function of the claimed "comparison between the results of the second search strategy and a characteristic of the target set"? The complaint alleges it does, but a technical question is whether measuring recall on a control set is functionally the same as the patent’s described method of terminating only after re-identifying a specific quantum of known-relevant documents from the original "target set".
V. Key Claim Terms for Construction
The Term: "target set"
Context and Importance: This term is foundational to the claim. Its construction will determine whether the accused product's "seed set" or "Control Set"—or some combination—can be considered the claimed "target set". The patent's concept relies on the independence of the "target set" from the subsequent classification process it is used to measure.
Intrinsic Evidence for Interpretation:
- Evidence for a Broader Interpretation: The specification states that documents in the target set "may be used to train a classifier" (’374 Patent, col. 6:43-44), which could support an argument that a "seed set" used for training is a "target set".
- Evidence for a Narrower Interpretation: The patent repeatedly describes the termination step as a comparison of the independent search results against the "target set" itself (e.g., "terminated when a sufficient number of documents in the target set have been classified as relevant") (’374 Patent, col. 4:37-40; Fig. 1, step 1060). This could support a narrower reading where the "target set" must be the direct object of the comparison for termination, not just an initial training input.
The Term: "terminate...based upon a comparison between the results of the second search strategy and a characteristic of the target set"
Context and Importance: This limitation defines the core inventive step of how the process stops. The dispute will likely focus on whether measuring statistical recall on a control set, as the Accused Instrumentality allegedly does, constitutes the specific "comparison" described in the patent.
Intrinsic Evidence for Interpretation:
- Evidence for a Broader Interpretation: The claim language is somewhat general ("a characteristic of the target set"), which could be argued to encompass statistical properties like recall derived from a set of documents like a control set.
- Evidence for a Narrower Interpretation: The patent specification provides a specific example: "the classification process may be terminated when the independent search strategy has substantially identified the documents in T" (’374 Patent, col. 7:49-51). This suggests a direct comparison where the system checks for the re-identification of specific documents, a potentially narrower function than calculating a general recall percentage on a control set.
VI. Other Allegations
- Indirect Infringement: The complaint alleges induced infringement under 35 U.S.C. § 271(b), stating that KLDiscovery publishes documentation and training materials that instruct customers to perform the claimed steps, such as creating a seed set, running the "Train & Classify" process, and using the Control Set features to terminate review (Compl. ¶58).
- Willful Infringement: The complaint alleges willful infringement based on Defendant having pre-suit knowledge of the ’374 Patent since at least July 14, 2025, from a notice letter, and continuing to engage in the allegedly infringing activity thereafter (Compl. ¶¶57, 59-60).
VII. Analyst’s Conclusion: Key Questions for the Case
- A core issue will be one of definitional scope: does the claim term "target set", described in the patent as a single set used as a benchmark for termination, read on the accused system’s alleged use of a "seed set" for training and a separate "Control Set" for performance validation?
- A key evidentiary question will be one of functional operation: does the accused product's termination logic—which relies on achieving a user-defined recall percentage against a Control Set—perform the specific "comparison" required by Claim 1, or is there a fundamental mismatch between this statistical validation and the patent's teaching of re-identifying documents from the original "target set"?