Multi-agent framework for synthetic data generation and verification

Abstract

The disclosure includes a computer-implemented method including generating a dataset of questions that mimic user queries by a multi-agent framework, wherein the multi-agent framework includes deployment of a plurality of artificial intelligence (AI) agents, running a subset of the dataset of questions through a first agent of the plurality of AI agents where each running of a question produces a trajectory which is a sequence of tools that the first agent calls, selecting a most common trajectory (MCT) for each question across all runnings, wherein the MCT for each question is annotated to indicate a correctness of the MCT, reverse engineering alternate questions from the responses for trajectory verification, extracting features from the MCT for each question that has been annotated and the alternate questions, and training a discriminative machine learning model on the features.

Description

Docket No. 102454.9082PCT -1- C / P / 1064223 / WO / SEC / 1MULTI-AGENT FRAMEWORK FOR SYNTHETIC DATA GENERATION AND VERIFICATIONCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of priority to U.S. Patent Application No. 19 / 402,998, filed November 27, 2025, which claims the benefit of priority to U.S. Provisional Application No. 63 / 726,145, filed on November 27, 2024, the entire contents of both are incorporated herein by reference.FIELD

[0002] The present disclosure relates to extending the capabilities of large language models (LLMs) with functions or tools through the use of artificial intelligence (Al) agents. More particularly, embodiments of the disclosure are directed to a multi-agent framework configured to generate a dataset of questions that mimic customer queries, and reverse engineer alternate questions from the responses for trajectory verification in order to improve Al agent performance on actual user queries.BACKGROUND

[0003] Recent advances in generative-text modeling have enabled agents, i.e., Al applications that use LLMs for planning and reasoning and leverage external functions (tools) to solve complex tasks and promote improved user interaction. Pushing the boundary even further are multi -agent systems that synergize the communication between multiple agents to achieve a common goal by distributing responsibilities across agents.

[0004] Building intelligent assistants to provide better customer experiences sits at the heart of every technology-driven enterprise. However, training (even fine-tuning) LLMs per user, or enterprise, is often infeasible due to insufficient data and privacy concerns. This creates the perfect opportunity for using agents which can reason over private data by relying only on the zero-shot abilities of its underlying LLM.

[0005] One issue with the current technology is that, before deploying custom assistants, a test suite of queries representative of actual user questions is needed. While asking users to interact with the assistant is an option that directly provides gold data, it is time consuming and thus, does not scale well.23103175Docket No. 102454.9082PCT -2- C / P / 1064223 / WO / SEC / 1

[0006] When addressing a question, an agent decides if it can be answered using the LLM’s internal knowledge or, requires information from the external environment. For the latter case, the agent invokes tools, i.e., functions performing dedicated tasks such as retrieving weather, converting currency, etc., to gather the necessary world knowledge. The sequence of tools that an agent calls is referred to as its trajectory (T), i.e., TQ = [ti, ..., tn], where TQ is the trajectory given question Q and ti are individual tool calls.

[0007] A second issue with the current technology pertains to determining the correctness of the agent trajectory. A tool call ti consists of two parts, viz., the tool name and the arguments. For example, in the call get_current_weather(city = “Boston”) the tool name is get current weather and the argument is city = “Boston”. Determining the correctness of the agent trajectory is a non-trivial problem and forms the first part of response verification. While using strong LLMs as-a-judge may be used in some instances, the approach faces drawbacks such as LLM sensitivity to the input prompt and inconsistent behavior of API (Application Programming Interface)-based models; while the latter is mitigated by altering generation temperature, it is not completely eliminated.SUMMARY OF THE INVENTION

[0008] To address both issues discussed above, the disclosure provides a multi-agent framework (“MAG-V”) for generating questions mimicking customer queries and verifying trajectories deterministically, i.e., without using LLMs to provide feedback. In some examples, verifying the trajectories may involve classical ML approaches such as distant supervision and discriminative models such as Support Vector Machines (SVM), etc., in combination with recent advances in LLM response verification such as SelfVerification Prompting.

[0009] Overall, the disclosure provides at least three novel contributions to the technical field of Al and particularly to that of LLMs including: (1) the use of agents configured to create synthetic data aligned with specific requirements; (2) the use of a deterministic method for verifying agent trajectories that does not rely on using LLMs for feedback; and (3) the use of simple ML baselines with feature engineering that matches the performance of more expensive and capable models.23103175Docket No. 102454.9082PCT -3- C / P / 1064223 / WO / SEC / 1

[0010] Data Generation with Multi-Agent Systems Synthetic data generation using LLMs has become a standard practice to address data scarcity and has been utilized for various tasks as inductive reasoning and question answering, the latter also utilizing ICL. However, using LLM agents for more complex data generation is a nascent and active area of study. Some experts have utilized multiple agents (2 for generation and at most 3 for evaluation) to create a preference alignment dataset. The issue with these strategies is the high reliance on LLM-as-a-judge, which is preferably avoided, and there is no human-in-the-loop to gauge the quality of generations.

[0011] In one current approach, a two-agent system is utilized to produce a conversation dataset for solving linear programming word problems and also relies on human evaluators for feedback. However, the novel multi -agent framework approach discussed here differs from such a current approach due to the multi -agent framework requiring “questions” and not dialog data, and prioritizing tool-calling to that end. Finally, other current approaches test the limits of multi -agent data generation by using a minimum of 29 agents to transform the seed to synthetic data while another creates “personas” to apply specific transformations to the seed data, creating 25M and IB samples respectively. Although impressive, these methods are still experimental, incur high costs, potentially introduce noise at such scales, and are unsuitable for many domains due to large seed data requirements.

[0012] Trajectory Verification Evaluating LLM response is an active area of study as there is no standardized method yet for gauging their correctness. Early attempts include LLM-as-a-judge like frameworks, which measures a machine learning model’s win rate (WR), which is the number of times a judge model prefers its response over a reference by considering factors such as consistency, relevancy, etc. Extending this idea for trajectory verification, some current approaches propose using a judge model (GPT-4) to determine WR and pass rate, i.e., whether a model is able to reach the goal in a certain number of steps. However, the drawback with such an approach is the over reliance on LLM-as-a-judge for trajectory evaluation.

[0013] Yet another current approach used for trajectory verification includes performing step-by-step tool call evaluation. Although this approach computes similar metrics as those computed with respect to the multi-agent framework, such a current approach ultimately23103175Docket No. 102454.9082PCT .4.C / P / 1064223 / WO / SEC / 1incorporates LLM-based evaluation which is orthogonal to the objective of the novel multiagent framework disclosed herein.BRIEF DESCRIPTION OF DRAWINGS

[0014] The above, and other, aspects, features, and advantages of several embodiments of the present disclosure will be more apparent from the following description as presented in conjunction with the following several figures of the drawings.

[0015] FIG. 1 is a conceptual illustration of data generation and trajectory verification operations in accordance with various embodiments of the disclosure;

[0016] FIG. 2 illustrates a first graphical representation of accuracy of a set of machine learning models and a second graphical representation of Fl scores of the set of machine learning models in accordance with various embodiments of the disclosure;

[0017] FIG. 3 provides a table providing metrics of a set of machine learning models trained using the best trajectory feature subset compared to GPT-baselines that displays the best accuracy and Fl across all random seeds with each score being the average from a set of trials in accordance with various embodiments of the disclosure;

[0018] FIG. 4 is a conceptual block diagram of a device suitable for configuration with logic of a multi-layer anomaly detection subsystem in accordance with various embodiments of the disclosure;

[0019] FIG. 5 is a block diagram illustrating an example computing environment that includes a data intake and query system in accordance with various embodiments of the disclosure;

[0020] FIG. 6 is a block diagram illustrating in greater detail an example of an indexing system of a data intake and query system, such as the data intake and query system of FIG.5 in accordance with various embodiments of the disclosure;

[0021] FIG. 7 is a block diagram illustrating in greater detail an example of the search system of a data intake and query system, such as the data intake and query system of FIG.5 in accordance with various embodiments of the disclosure; and23103175Docket No. 102454.9082PCT .5.C / P / 1064223 / WO / SEC / 1

[0022] FIG. 8 illustrates an example of a self-managed network that includes a data intake and query system in accordance with various embodiments of the disclosure.

[0023] Corresponding reference characters indicate corresponding components throughout the several figures of the drawings. Elements in the several figures are illustrated for simplicity and clarity and have not necessarily been drawn to scale. For example, the dimensions of some of the elements in the figures might be emphasized relative to other elements for facilitating understanding of the various presently disclosed embodiments. In addition, common, but well -understood, elements that are useful or necessary in a commercially feasible embodiment are often not depicted in order to facilitate a less obstructed view of these various embodiments of the present disclosure.DETAILED DESCRIPTION

[0024] Aspects of the present disclosure may be embodied as an apparatus, system, method, or computer program product. Accordingly, aspects of the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, micro-code, or the like) or an embodiment combining software and hardware aspects that may all generally be referred to herein as a “function,” “module,” “apparatus,” or “system.”. Furthermore, aspects of the present disclosure may take the form of a computer program product embodied in one or more non-transitory computer-readable storage media storing computer-readable and / or executable program code. Many of the functional units described in this specification have been labeled as functions, in order to emphasize their implementation independence more particularly. For example, a function may be implemented as a hardware circuit comprising custom VLSI circuits or gate arrays, off-the-shelf semiconductors such as logic chips, transistors, or other discrete components. A function may also be implemented in programmable hardware devices such as via field programmable gate arrays, programmable array logic, programmable logic devices, or the like.

[0025] Functions may also be implemented at least partially in software for execution by various types of processors. An identified function of executable code may, for instance, comprise one or more physical or logical blocks of computer instructions that may, for instance, be organized as an object, procedure, or function. Nevertheless, the executables of an identified function need not be physically located together but may comprise disparate 23103175Docket No. 102454.9082PCT -6- C / P / 1064223 / WO / SEC / 1instructions stored in different locations which, when joined logically together, comprise the function and achieve the stated purpose for the function.

[0026] Indeed, a function of executable code may include a single instruction, or many instructions, and may even be distributed over several different code segments, among different programs, across several storage devices, or the like. Where a function or portions of a function are implemented in software, the software portions may be stored on one or more computer-readable and / or executable storage media. Any combination of one or more computer-readable storage media may be utilized. A computer-readable storage medium may include, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing, but would not include propagating signals. In the context of this document, a computer readable and / or executable storage medium may be any tangible and / or non-transitory medium that may contain or store a program for use by or in connection with an instruction execution system, apparatus, processor, or device.

[0027] Computer program code for carrying out operations for aspects of the present disclosure may be written in any combination of one or more programming languages, including an object-oriented programming language such as Python, Java, Smalltalk, C++, C#, Objective C, or the like, conventional procedural programming languages, such as the “C” programming language, scripting programming languages, and / or other similar programming languages. The program code may execute partly or entirely on one or more of a user's computer and / or on a remote computer or server over a data network or the like.

[0028] A component, as used herein, comprises a tangible, physical, non-transitory device. For example, a component may be implemented as a hardware logic circuit comprising custom VLSI circuits, gate arrays, or other integrated circuits; off-the-shelf semiconductors such as logic chips, transistors, or other discrete devices; and / or other mechanical or electrical devices. A component may also be implemented in programmable hardware devices such as field programmable gate arrays, programmable array logic, programmable logic devices, or the like. A component may comprise one or more silicon integrated circuit devices (e.g., chips, die, die planes, packages) or other discrete electrical devices, in electrical communication with one or more other components through electrical lines of a printed circuit board (PCB) or the like. Each of the functions and / or modules described23103175Docket No. 102454.9082PCT -7- C / P / 1064223 / WO / SEC / 1herein, in certain embodiments, may alternatively be embodied by or implemented as a component.

[0029] A circuit, as used herein, comprises a set of one or more electrical and / or electronic components providing one or more pathways for electrical current. In certain embodiments, a circuit may include a return pathway for electrical current, so that the circuit is a closed loop. In another embodiment, however, a set of components that does not include a return pathway for electrical current may be referred to as a circuit (e.g., an open loop). For example, an integrated circuit may be referred to as a circuit regardless of whether the integrated circuit is coupled to ground (as a return pathway for electrical current) or not. In various embodiments, a circuit may include a portion of an integrated circuit, an integrated circuit, a set of integrated circuits, a set of non-integrated electrical and / or electrical components with or without integrated circuit devices, or the like. In one embodiment, a circuit may include custom VLSI circuits, gate arrays, logic circuits, or other integrated circuits; off-the-shelf semiconductors such as logic chips, transistors, or other discrete devices; and / or other mechanical or electrical devices. A circuit may also be implemented as a synthesized circuit in a programmable hardware device such as field programmable gate array, programmable array logic, programmable logic device, or the like (e.g., as firmware, a netlist, or the like). A circuit may comprise one or more silicon integrated circuit devices (e.g., chips, die, die planes, packages) or other discrete electrical devices, in electrical communication with one or more other components through electrical lines of a printed circuit board (PCB) or the like. Each of the functions and / or modules described herein, in certain embodiments, may be embodied by or implemented as a circuit.

[0030] Reference throughout this specification to “one embodiment,” “an embodiment,” or similar language means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present disclosure. Thus, appearances of the phrases “in one embodiment,” “in an embodiment,” and similar language throughout this specification may, but do not necessarily, all refer to the same embodiment, but mean “one or more but not all embodiments” unless expressly specified otherwise. The terms “including,” “comprising,” “having,” and variations thereof mean “including but not limited to”, unless expressly specified otherwise. An enumerated listing of items does not imply that any or all of the items are mutually exclusive and / or23103175Docket No. 102454.9082PCT -8- C / P / 1064223 / WO / SEC / 1mutually inclusive, unless expressly specified otherwise. The terms “a,” “an,” and “the” also refer to “one or more” unless expressly specified otherwise.

[0031] Further, as used herein, reference to reading, writing, storing, buffering, and / or transferring data can include the entirety of the data, a portion of the data, a set of the data, and / or a subset of the data. Likewise, reference to reading, writing, storing, buffering, and / or transferring non-host data can include the entirety of the non-host data, a portion of the non-host data, a set of the non-host data, and / or a subset of the non-host data. Lastly, the terms “or” and “and / or” as used herein are to be interpreted as inclusive or meaning any one or any combination. Therefore, “A, B or C” or “A, B and / or C” mean “any of the following: A; B; C; A and B; A and C; B and C; A, B and C ”. An exception to this definition will occur only when a combination of elements, functions, steps, or acts are in some way inherently mutually exclusive.

[0032] Aspects of the present disclosure are described below with reference to schematic flowchart diagrams and / or schematic block diagrams of methods, apparatuses, systems, and computer program products according to embodiments of the disclosure. It will be understood that each block of the schematic flowchart diagrams and / or schematic block diagrams, and combinations of blocks in the schematic flowchart diagrams and / or schematic block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a computer or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor or other programmable data processing apparatus, create means for implementing the functions and / or acts specified in the schematic flowchart diagrams and / or schematic block diagrams block or blocks.

[0033] It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. Other steps and methods may be conceived that are equivalent in function, logic, or effect to one or more blocks, or portions thereof, of the illustrated figures. Although various arrow types and line types may be employed in the flowchart and / or block diagrams, they are understood not to limit the scope of the corresponding embodiments. For instance, an arrow23103175Docket No. 102454.9082PCT .9.C / P / 1064223 / WO / SEC / 1may indicate a waiting or monitoring period of unspecified duration between enumerated steps of the depicted embodiment.

[0034] In the following detailed description, reference is made to the accompanying drawings, which form a part thereof. The foregoing summary is illustrative only and is not intended to be in any way limiting. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features will become apparent by reference to the drawings and the following detailed description. The description of elements in each figure may refer to elements of proceeding figures. Like numbers may refer to like elements in the figures, including alternate embodiments of like elements.1. Methodology

[0035] The disclosure deconstructs the operability and construction of a novel multi-agent framework referred to herein as “MAG-V,” which is illustrated by FIG. 1. The operability of MAG-V may begin with a dataset construction (discussed further below in Section 1.1), which is followed by a trajectory verification (discussed further below in Section 1.2). In one example embodiment as shown in FIG. 1, three agents have been configured for deployment within MAG-V. One may be referred to as an “investigator” and is responsible for query generation, a second may be referred to as an “assistant” and is responsible for answering queries, and a third may be referred to as a “reverse engineer” and is responsible for creating questions based on a given response. All three agents may use an API-based LLM including large models (such as Claude 3.5 Sonnet and GPT4o), and small models (such as Claude 3 Haiku, GPT4o-mini, and Gemini -fl ash). The investigator and reverseengineer agents may use GPT-4o-mini while the assistant agent may use GPT-4o-mini during data-generation and GPT-4o during verification. Such a configuration may be due to the fact that 4o-mini may time-out for many reverse queries (discussed below in Section 1.2).1.1 Data Generation

[0036] The purpose of the synthetic generation phase is to create a dataset of queries indicative of what a user might ask the assistant agent. This phase may include generation or receipt of a set of questions (e.g., 5, 10, 15, 20, etc.) that have been manually written by an expert to meet certain product requirements, e.g., historical questions. These questions,23103175Docket No. 102454.9082PCT -10- C / P / 1064223 / WO / SEC / 1along with their verified trajectories, form a seed dataset from which synthetic samples are generated.

[0037] The questions are generated to stress test the assistant agent, e.g., causing the assistant agent to invoke a variety of tools to handle complex user queries. To that end, InContext Learning (ICL) (predictions based on the given examples without parameter update) is performed to illustrate to the assistant agent the types of trajectories and questions desired. Each ICL sample has a trajectory Tl, a question that was answered using that trajectory QI, and another trajectory T2 randomly sampled from the seed dataset. The investigator agent executes T2 and is tasked with writing 10 questions similar in style to QI by studying the tool responses, which results in a total of 190 synthetic questions, which may then be filtered (discussed below in Section 3) down to a smaller sample size, e.g., 20, 45, 50, etc., questions (“subset of sample questions”).

[0038] To account for fluctuations in API calls, each of the questions in the subset of sample questions are run a plurality of times (e.g., 5 times in some embodiments) via the assistant agent and the most common trajectory (MCT) across all trials is determined for annotation. A response using the MCT may be randomly and added to the dataset.

[0039] In some instances, one or more independent domain experts may annotate the dataset. Given a question and its corresponding trajectory, each trajectory may be annotated (e.g., 0 or 1). In some examples, ‘0’ represents that trajectory was incorrect (the trajectory made incorrect calls / makes assumptions beyond the question requirements, etc.) and ‘1’ represents that the trajectory was correct (all steps are logically correct, and the overall trajectory takes into account all of the user requirements). Annotating the trajectories in such a manner essentially casts the problem as a binary classification.1.2 Trajectory Verification

[0040] The second phase of MAG-V is trajectory verification, which includes ascertaining whether the assistant agent calls the correct sequence of tools to reach the response. This forms the first part of response verification. One example methodology is based on the hypothesis that, “If questions similar to a base question follow the same trajectory, we can establish a degree of confidence in the assistant’s reasoning pathway. Else, high trajectory23103175Docket No. 102454.9082PCT -11- C / P / 1064223 / WO / SEC / 1variance equates to lower confidence.” Such a hypothesis may inspired by two ideas, viz.: distant supervision and backward reasoning.

[0041] Referring first to distant supervision, this idea includes methods to make predictions about unlabeled data using auxiliary knowledge sources. For example, when building a relation extraction model, knowledge graphs may be used to make relationship inferences on unstructured text. In a related fashion, the trajectories of similar questions may be used to make predictions about our annotated trajectories.

[0042] Second, referring to backward reasoning, under this idea, reverse question answering (RQA) and self-verification prompting involve verifying LLM-responses by generating questions from the final response and asking the LLM to answer them. RQA is a jeopardy -like technique where the generated question’s answer must be the response, while the latter generates questions whose answers contain all facts entailed in the original question. In some examples of methods disclosed herein, a backward question generation strategy may be deployed focusing on trajectory verification instead of response.

[0043] To verify the MCT of a question Q, the reverse engineer agent is asked to create a set of alternate questions (AQ) (e.g., 3 AQs) based on the assistant agent’s response using the MCT. The reverse engineer agent was instructed to write questions that captured all of the main points of the response. The assistant agent then answers the AQ’s which leads to alternate trajectories (AT), e.g., 3 alternate trajectories for 3 AQs.

[0044] Using the annotated (or base) trajectory (BT) and the AT’s, features are extracted to determine the similarity therebetween. These features are a combination of statistical and embedding-based measures. In some instances, six features may be computed between the BT and AT’s (discussed below in Section 4), viz., EM (Exact Match - 0 / 1 measure of equality); EDIT (minimum number of string edits needed to convert AT to BT); GEDIT (Graph Edit Distance between the trajectories formulated as graphs); SS (Semantic (Cosine) Similarity between BT and AT); AO (Argument Overlap - count of common arguments between BT and AT) and LCSS (Longest Common Sequence from Starting Call to measure the extent of commonality between BT and AT from a common point). Additionally, TF-IDF (Term Frequency-Inverse Document Frequency) features may also be extracted from the base questions to provide grounding context to the models as a trajectory can only be analyzed in relation to its question.23103175Docket No. 102454.9082PCT -12- C / P / 1064223 / WO / SEC / 1

[0045] Finally, using these features, discriminative ML models are trained using stratified ten-fold cross-validation to account for label imbalance. Overall, a plurality of machine learning models (e.g., Random Forest, Logistic Regression, Naive Bayes, k-Nearest Neighbours (k-NN), Support Vector Machines (SVM), Decision Tree, XGBoost) may be tested with the mean accuracy and Fl reported across all folds and 3 random seeds. Apart from AO, the influence of including and excluding the tool arguments for each measure for trajectory verification is tested. All models may use default settings except for k-NN, which was found to have the best performance at k = 5 (without arguments verification) and k = 4 (with arguments verification).1.2.1 GPT-baseline

[0046] To compare against a strong LLM-as-a-judge, GPT-4 and GPT-4o were selected. Each machine learning model was tested on the same test splits and random seeds from the cross-validation to provide consistent evaluation. The same annotation rubric was provided to the models, but the position of the labels may be switched to account for recency bias, i.e., the tendency of LLMs to repeat the last seen values in the prompt. Before generating its score, each LLM was asked to provide a rationale as this tactical has been shown to lead to better evaluation. Additionally, one example each of a correct and incorrect Q+T pair from the seed dataset was provided to further help the machine learning models. Finally, the performance with and without using the default system prompt (“You are a helpful Al Assistant.”) was tested in the framework used to create the agents.2. Experiments

[0047] The following discusses the benefit of generating synthetic questions and then detail the results from trajectory verification trials.2.1 Utilizing Synthetic Data

[0048] The goal of creating synthetic questions is to simulate user queries to test the capabilities of the assistant agent. As a by-product of this step, the generated data may be used to aid the assistant agent in fixing its mistakes and answering questions better through ICL. The following is an example of a user query (including a grammatical issue): “I see that the paymentservice on production had an increased error rate around 7:59pm EST23103175Docket No. 102454.9082PCT -13- C / P / 1064223 / WO / SEC / 110 / 10 / 2024, can you dive into that service during that time and a half hour and each side and find out what the most common errors associated were?”

[0049] The assistant agent may run GPT-4o as its backbone. However, to test for cost-efficiency, 4o-mini was also considered. To answer the above user query, the assistant agent needs to use the correct time range, which is 7:29 PM to 8:29 PM on October 10. The above query maybe run with both 4o and 4o-mini, and also providing the subset of synthetic questions and their MCT as ICL samples. The 4o-backed assistant agent with and without the samples may fail to use the correct time range: e.g., using {“time_range”:{“start”:“2024-10-10T23:29:00Z”,“stop”:“2024-10-l 1TOO:29:OOZ”}}.

[0050] While the start date is correct in the above, the start / end times and end date are wrong. With 4o-mini, the assistant agent may spiral into a chain of retries, unable to fix itself even after multiple (e.g., 10) attempts. However, by adding the generated questions discussed above and their MCT as ICL samples, the assistant agent is better able to figure out the correct time range on the first try, {“time_range”: {“start”: “2024-10-10T19:29:00Z”, “stop”:“2024-10-10T20:29:00Z”}}}. This provision of the generated questions discussed above provides a technical benefit and potential cost-benefits, i.e., by coupling a cheaper model with ICL samples, we can guide it to perform better or on-par with more expensive LLMs.2.2 Trajectory Verification

[0051] The results from verification trials (accuracy and Fl) are shown in Figure 2. As can be seen, each machine learning model benefits from features that consider tool arguments. The reason for performing tests that removed the arguments is to see if the overall tool calls were consistent across the base and alternate trajectories. However, by considering the arguments, the machine learning models perform better as they have more fine-grained features to learn from. Furthermore, by excluding the arguments, the boundary separating the correct and incorrect trajectories becomes less prominent since many trajectories with the same calls get placed in both classes. For example, if T1 = T2 = [tl, t2, t3], where T1 is incorrect and T2 is correct for their respective questions, the machine learning model learns the same features for each class which in turn distorts its performance. By adding the arguments, the model is able to learn a better decision23103175Docket No. 102454.9082PCT -14- C / P / 1064223 / WO / SEC / 1boundary. Overall, it can be seen that the k-NN model shows the best accuracy and Fl scores across all ML models.

[0052] Referring to FIG. 2, GPT-4o, which is alleged by those having ordinary skill in the art to be a much stronger model than GPT-4, is outperformed by the latter by quite a margin. Second, it can be seen that by simply adding the system prompt (“You are a helpful Al Assistant.”), the accuracy of GPT-4o increases by about 12%, which hearkens on prompt sensitivity, i.e., making such small modifications to a prompt can have tremendous impact on the output. Finally, the k-NN model shows 11% accuracy and 4% Fl improvement over GPT-4o which demonstrates the effectiveness of such a simple model over a more complex LLM (i.e., demonstrating the technical improvement of the inventive concepts in the technical field of LLMs). While the k-NN model (82.33%) matches GPT-4 in accuracy (82.83%), it lags behind in Fl (71.73 vs. 76.06), indicating a reduced alignment of predictions. However, this may be addressed with a larger dataset for the k-NN model to learn from.

[0053] Feature ablations for the machine learning models are considered as shown in Table 1. These trials providing the result of Table 1 only considered the six trajectory features (with arguments) to obtain an understanding of the extent of performance that may be achieved by only conditioning the machine learning models on the trajectories. As the number of features here is much lower than the standard setup, five-fold cross-validation (to avoid overfitting) is run but the same random seeds as before are used. Each feature combination (by considering 1, . . ., 6 features at a time) is tested and scores for those models which had the best score using the same feature subset across each seed are reported. The GPT-baselines shown in Table 1 (FIG. 3) are their best scores, i.e., using the system prompt. In this setup, it can be seen that the best machine learning models come close to GPT-4 and outperform GPT-4o in accuracy. Additionally, the distribution of predictions (Fl) is much more aligned with that of GPT-4. The models utilize only a single feature, i.e., EDIT distance, to make their best predictions, e.g., because EDIT is a more lenient metric than the others, as it determines the amount of work needed to align two trajectories, rather than penalizing mismatches which in turn provides more signal to the models.23103175Docket No. 102454.9082PCT -15- C / P / 1064223 / WO / SEC / 1

[0054] It may be seen that performance for both the novel multi-agent approach discussed herein versus using LLM-as-a-judge are comparable using either question and trajectory features, or just the latter in isolation. Where the novel multi-agent approach discussed herein wins over the GPT benchmark is in determinism, i.e., for the exact same conditions, it will always give the same result, which cannot be said for API-based LLMs. It may be understood that the novel multi-agent approach discussed herein follows a best-of-N-predictions strategy, i.e., the decisions are based on what the majority of the trajectories predict as opposed to a single trajectory. This leads to a model that is more robust and confident in its predictions.

[0055] To summarize Table 1 (FIG. 1), the table illustrates performance metrics of ML models trained using the best trajectory feature subset vs. GPT-baselines. Using EDIT distance, each model displays the best accuracy and Fl across all random seeds. Each score is the average from all trials.3. Filtering Generated Questions

[0056] To avoid having questions of the same type for the test set, the generated questions are filtered as referenced above. The filtering may include embedding the questions using BAAI General Embeddings (BGE), which achieves state-of-the-art performance on MTEB (Massive Text Embedding Benchmark). In one embodiment, this yields R1024 question embeddings. Using UMAP (UniformManifold Approximation and Projection) for dimension-reduction to R2, the resulting embeddings are clustered with HDBSCAN, which results in 11 clusters that are verified (e.g., manually) as being sufficiently diverse in their themes such as requesting information on a given service in one or more data environments, issues related to a hardware component, etc. Considering the mean of all embeddings in a cluster as the centroid, the top-5 questions with the least distance to the centroid are selected, which gives a total of 55 questions. Finally, 10 questions are removed that are either generic and do not require tool calls (or for other reasons), which yields a dataset of 45 questions.4. Traj ectory F eatures

[0057] Six features are extracted across the base and alternate trajectories, described as follows:23103175Docket No. 102454.9082PCT -16- C / P / 1064223 / WO / SEC / 1• EM (Exact Match): A binary measure of equality, i.e., EM = 1, if BT = AT verbatim, else 0.• EDIT: Levenshtein Edit Distance (Miller et al., 2009) between BT and AT, i.e., the minimum number of edits (add, delete, substitute) required to modify the AT to become the BT.• GEDIT (Graph Edit Distance): As a trajectory is a sequence of operations, it can also be viewed as a directed graph. As such, the edit distance required to make the BT and AT graphs isomorphic, i.e., similar in structure, is measured. GEDIT is similar to EDIT, but operates on nodes and edges.• SS (Semantic Similarity): This measures cosine similarity between the BT and AT embedded using BGE-Large embeddings.• AO (Argument Overlap): Count of arguments common to BT and AT for each tool call pair. AO is an Fl score (2PR / (P + R)), where:- overlap (O) = number of common arguments across all calls for BT and AT. - precision (P) = O / total number of arguments in AT.- recall (R) = O / total number of arguments in BT.• LCSS (Longest Common Sequence from Starting Call): The longest tool call sequence common to BT and AT starting from the first call. Similar to AO, LCSS is an Fl score with:- L = Length of longest sequence across BT and AT from the starting call. - P = L / total number of calls in AT.- R = L / total number of calls in BT.5. Summary

[0058] The disclosure introduces MAG-V, a framework for generating synthetic questions using LLM agents and verifying the trajectory taken by an agent to answer questions. The novelty of the verification system lies in its deterministic aspect, i.e., reducing the reliance on API-based LLMs to judge the veracity of results (trajectories).6. Computing Device

[0059] Referring now to FIG. 4, a conceptual block diagram of a device suitable for configuration with logic deployed in accordance with various embodiments of the disclosure is shown. The embodiment of the conceptual block diagram depicted in FIG. 4 23103175Docket No. 102454.9082PCT -17- C / P / 1064223 / WO / SEC / 1can illustrate a conventional server, computer, workstation, desktop computer, laptop, tablet, network appliance, e-reader, smartphone, or other computing device, and can be utilized to execute any of the application and / or logic components presented herein. The embodiment of the conceptual block diagram depicted in FIG. 4 can also illustrate an access point, a switch, or a router in accordance with various embodiments of the disclosure. The device 400 may, in many nonlimiting examples, correspond to physical devices or to virtual resources described herein.

[0060] In many embodiments, the device 400 may include an environment 402 such as a baseboard or “motherboard,” in physical embodiments that can be configured as a printed circuit board with a multitude of components or devices connected by way of a system bus or other electrical communication paths. Conceptually, in virtualized embodiments, the environment 402 may be a virtual environment that encompasses and executes the remaining components and resources of the device 400. In more embodiments, one or more processors 404, such as, but not limited to, central processing units (“CPUs”) can be configured to operate in conjunction with a chipset 406. The processor(s) 404 can be standard programmable CPUs that perform arithmetic and logical operations necessary for the operation of the device 400.

[0061] In a number of embodiments, the processor(s) 404 can perform one or more operations by transitioning from one discrete, physical state to the next through the manipulation of switching elements that differentiate between and change these states. Switching elements generally include electronic circuits that maintain one of two binary states, such as flip-flops, and electronic circuits that provide an output state based on the logical combination of the states of one or more other switching elements, such as logic gates. These basic switching elements can be combined to create more complex logic circuits, including registers, adders-subtractors, arithmetic logic units, floating-point units, and the like.

[0062] In various embodiments, the chipset 406 may provide an interface between the processor(s) 404 and the remainder of the components and devices within the environment 402. The chipset 406 can provide an interface to a random-access memory (“RAM”) 408, which can be used as the main memory in the device 400 in some embodiments. The chipset 406 can further be configured to provide an interface to a computer-readable storage23103175Docket No. 102454.9082PCT -18- C / P / 1064223 / WO / SEC / 1medium such as a read-only memory (“ROM”) 410 or non-volatile RAM (“NVRAM”) for storing basic routines that can help with various tasks such as, but not limited to, starting up the device 400 and / or transferring information between the various components and devices. The ROM 410 or NVRAM can also store other application components necessary for the operation of the device 400 in accordance with various embodiments described herein.

[0063] Additional embodiments of the device 400 can be configured to operate in a networked environment using logical connections to remote computing devices and computer systems through a network, such as the network 440. The chipset 406 can include functionality for providing network connectivity through a network interface card (“NIC”) 412, which may comprise a gigabit Ethernet adapter or similar component. The NIC 412 can be capable of connecting the device 400 to other devices over the network 440. It is contemplated that multiple NICs 412 may be present in the device 400, connecting the device to other types of networks and remote systems.

[0064] In further embodiments, the device 400 can be connected to a storage 418 that provides non-volatile storage for data accessible by the device 400. The storage 418 can, for instance, store an operating system 420, and programs 422. In various embodiments, the storage 418 includes logic modules encompassing logic deployed in embodiments discussed herein and the summary and detection indexes (“data stores 426”) as discussed above.

[0065] The storage 418 can be connected to the environment 402 through a storage controller 414 connected to the chipset 406. In certain embodiments, the storage 418 can consist of one or more physical storage units. The storage controller 414 can interface with the physical storage units through a serial attached SCSI (“SAS”) interface, a serial advanced technology attachment (“SATA”) interface, a fiber channel (“FC”) interface, or other type of interface for physically connecting and transferring data between computers and physical storage units.

[0066] The device 400 can store data within the storage 418 by transforming the physical state of the physical storage units to reflect the information being stored. The specific transformation of physical state can depend on various factors. Examples of such factors can include, but are not limited to, the technology used to implement the physical storage 23103175Docket No. 102454.9082PCT -19- C / P / 1064223 / WO / SEC / 1units, whether the storage 418 is characterized as primary or secondary storage, and the like.

[0067] In many more embodiments, the device 400 can store information within the storage 418 by issuing instructions through the storage controller 414 to alter the magnetic characteristics of a particular location within a magnetic disk drive unit, the reflective or refractive characteristics of a particular location in an optical storage unit, or the electrical characteristics of a particular capacitor, transistor, or other discrete component in a solid-state storage unit, or the like. Other transformations of physical media are possible without departing from the scope and spirit of the present description, with the foregoing examples provided only to facilitate this description. The device 400 can further read or access information from the storage 418 by detecting the physical states or characteristics of one or more particular locations within the physical storage units.

[0068] In addition to the storage 418 described above, the device 400 can have access to other computer-readable storage media to store and retrieve information, such as program modules, data structures, or other data. It should be appreciated by those skilled in the art that computer-readable storage media is any available media that provides for the non-transitory storage of data and that can be accessed by the device 400. In some examples, the operations performed by a cloud computing network, and or any components included therein, may be supported by one or more devices similar to device 400. Stated otherwise, some or all of the operations performed by the cloud computing network, and or any components included therein, may be performed by one or more devices 400 operating in a cloud-based arrangement. By way of example, and not limitation, computer-readable storage media can include volatile and non-volatile, removable and non-removable media implemented in any method or technology.

[0069] By way of example, and not limitation, computer-readable storage media can include volatile and non-volatile, removable and non-removable media implemented in any method or technology. Computer-readable storage media includes, but is not limited to, RAM, ROM, erasable programmable ROM (“EPROM”), electrically-erasable programmable ROM (“EEPROM”), flash memory or other solid-state memory technology, compact disc ROM (“CD-ROM”), digital versatile disk (“DVD”), high definition DVD (“HD-DVD”), BLU-RAY, or other optical storage, magnetic cassettes, magnetic tape,23103175Docket No. 102454.9082PCT -20- C / P / 1064223 / WO / SEC / 1magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information in a non-transitory fashion.

[0070] As mentioned briefly above, the storage 418 can store an operating system 420 utilized to control the operation of the device 400. According to one embodiment, the operating system comprises the LINUX operating system. According to another embodiment, the operating system comprises the WINDOWS® SERVER operating system from MICROSOFT Corporation of Redmond, Washington. According to further embodiments, the operating system can comprise the UNIX operating system or one of its variants. It should be appreciated that other operating systems can also be utilized. The storage 418 can store other system or application programs and data utilized by the device 400.

[0071] In many additional embodiments, the storage 418 or other computer-readable storage media is encoded with computer-executable instructions which, when loaded into the device 400, may transform it from a general -purpose computing system into a specialpurpose computer capable of implementing the embodiments described herein. These computer executable instructions may be stored as program 422 (for example, an application) and transform the device 400 by specifying how the processor(s) 404 can transition between states, as described above. In some embodiments, the device 400 has access to computer-readable storage media storing computer executable instructions which, when executed by the device 400, perform the various processes described above with regard to any of the figures discussed herein. In certain embodiments, the device 400 can also include computer-readable storage media having instructions stored thereupon for performing any of the other computer-implemented operations described herein.

[0072] In still further embodiments, the device 400 can also include one or more input / output controllers 416 for receiving and processing input from a number of input devices, such as a keyboard, a mouse, a touchpad, a touch screen, an electronic stylus, or other type of input device. Similarly, an input / output controller 416 can be configured to provide output to a display, such as a computer monitor, a flat panel display, a digital projector, a printer, or other type of output device. Those skilled in the art will recognize that the device 400 might not include all of the components shown in FIG. 4 and can include23103175Docket No. 102454.9082PCT -21- C / P / 1064223 / WO / SEC / 1other components that are not explicitly shown in FIG. 4 or might utilize an architecture completely different than that shown in FIG. 4.

[0073] As described above, the device 400 may support a virtualization layer, such as one or more virtual resources executing on the device 400. In some examples, the virtualization layer may be supported by a hypervisor that provides one or more virtual machines running on the device 400 to perform functions described herein. The virtualization layer may generally support a virtual resource that performs at least a portion of the techniques described herein.

[0074] Although a specific embodiment for a device suitable for configuration with logic of an Al system for carrying out the various steps, processes, methods, and operations described herein is discussed with respect to FIG. 4, any of a variety of systems and / or processes may be utilized in accordance with embodiments of the disclosure. For example, the device may be in a virtual environment such as a cloud-based network administration suite, or it may be distributed across a variety of network devices or switches. The elements depicted in FIG. 4 may also be interchangeable with other elements of the disclosure as appropriate to realize a particularly desired embodiment.G. Data Intake and Query Architecture and Use Cases

[0075] Entities that operate computing environments need information about their computing environments. For example, an entity may need to know the operating status of the various computing resources in the entity’s computing environment, so that the entity can administer the environment, including performing configuration and maintenance, performing repairs or replacements, provisioning additional resources, removing unused resources, or addressing issues that may arise during operation of the computing environment, among other examples. As another example, an entity can use information about a computing environment to identify and remediate security issues that may endanger the data, users, and / or equipment in the computing environment. As another example, an entity may be operating a computing environment for some purpose (e.g., to run an online store, to operate a bank, to manage a municipal railway, etc.) and may want information about the computing environment that can aid the entity in understanding whether the computing environment is operating efficiently and for its intended purpose.23103175Docket No. 102454.9082PCT -22- C / P / 1064223 / WO / SEC / 1

[0076] Collection and analysis of the data from a computing environment can be performed by a data intake and query system such as is described herein. A data intake and query system can ingest and store data obtained from the components in a computing environment, and can enable an entity to search, analyze, and visualize the data. Through these and other capabilities, the data intake and query system can enable an entity to use the data for administration of the computing environment, to detect security issues, to understand how the computing environment is performing or being used, and / or to perform other analytics.

[0077] FIG. 5 is a block diagram illustrating an example computing environment 500 that includes a data intake and query system 510. The data intake and query system 510 obtains data from a data source 502 in the computing environment 500 and ingests the data using an indexing system 520. A search system 560 of the data intake and query system 510 enables users to navigate the indexed data. Though drawn with separate boxes in FIG. 5, in some implementations the indexing system 520 and the search system 560 can have overlapping components. A computing device 504, running a network access application 506, can communicate with the data intake and query system 510 through a user interface system 514 of the data intake and query system 510. Using the computing device 504, a user can perform various operations with respect to the data intake and query system 510, such as administration of the data intake and query system 510, management and generation of “knowledge objects,” (user-defined entities for enriching data, such as saved searches, event types, tags, field extractions, lookups, reports, alerts, data models, workflow actions, and fields), initiating of searches, and generation of reports, among other operations. The data intake and query system 510 can further optionally include apps 512 that extend the search, analytics, and / or visualization capabilities of the data intake and query system 510.

[0078] The data intake and query system 510 can be implemented using program code that can be executed using a computing device. A computing device is an electronic device that has a memory for storing program code instructions and a hardware processor for executing the instructions. The computing device can further include other physical components, such as a network interface or components for input and output. The program code for the data intake and query system 510 can be stored on a non-transitory computer-readable medium, such as a magnetic or optical storage disk or a flash or solid-state memory, from which the program code can be loaded into the memory of the computing device for execution. “Non-23103175Docket No. 102454.9082PCT -23- C / P / 1064223 / WO / SEC / 1transitory” means that the computer-readable medium can retain the program code while not under power, as opposed to volatile or “transitory” memory or media that requires power in order to retain data.

[0079] In various examples, the program code for the data intake and query system 510 can be executed on a single computing device, or execution of the program code can be distributed over multiple computing devices. For example, the program code can include instructions for both indexing and search components (which may be part of the indexing system 520 and / or the search system 560, respectively), which can be executed on a computing device that also provides the data source 502. As another example, the program code can be executed on one computing device, where execution of the program code provides both indexing and search components, while another copy of the program code executes on a second computing device that provides the data source 502. As another example, the program code can be configured such that, when executed, the program code implements only an indexing component or only a search component. In this example, a first instance of the program code that is executing the indexing component and a second instance of the program code that is executing the search component can be executing on the same computing device or on different computing devices.

[0080] The data source 502 of the computing environment 500 is a component of a computing device that produces machine data. The component can be a hardware component (e.g., a microprocessor or a network adapter, among other examples) or a software component (e.g., a part of the operating system or an application, among other examples). The component can be a virtual component, such as a virtual machine, a virtual machine monitor (also referred as a hypervisor), a container, or a container orchestrator, among other examples. Examples of computing devices that can provide the data source 502 include personal computers (e.g., laptops, desktop computers, etc.), handheld devices (e.g., smart phones, tablet computers, etc.), servers (e.g., network servers, compute servers, storage servers, domain name servers, web servers, etc.), network infrastructure devices (e.g., routers, switches, firewalls, etc.), and “Internet of Things” devices (e.g., vehicles, home appliances, factory equipment, etc.), among other examples. Machine data is electronically generated data that is output by the component of the computing device and reflects activity of the component. Such activity can include, for example, operation status, actions performed, performance metrics, communications with other components, or23103175Docket No. 102454.9082PCT -24- C / P / 1064223 / WO / SEC / 1communications with users, among other examples. The component can produce machine data in an automated fashion (e.g., through the ordinary course of being powered on and / or executing) and / or as a result of user interaction with the computing device (e.g., through the user’s use of input / output devices or applications). The machine data can be structured, semi-structured, and / or unstructured. The machine data may be referred to as raw machine data when the data is unaltered from the format in which the data was output by the component of the computing device. Examples of machine data include operating system logs, web server logs, live application logs, network feeds, metrics, change monitoring, message queues, and archive files, among other examples.

[0081] As discussed in greater detail below, the indexing system 520 obtains machine date from the data source 502 and processes and stores the data. Processing and storing of data may be referred to as “ingestion” of the data. Processing of the data can include parsing the data to identify individual events, where an event is a discrete portion of machine data that can be associated with a timestamp. Processing of the data can further include generating an index of the events, where the index is a data storage structure in which the events are stored. The indexing system 520 does not require prior knowledge of the structure of incoming data (e.g., the indexing system 520 does not need to be provided with a schema describing the data). Additionally, the indexing system 520 retains a copy of the data as it was received by the indexing system 520 such that the original data is always available for searching (e.g., no data is discarded, though, in some examples, the indexing system 520 can be configured to do so).

[0082] The search system 560 searches the data stored by the indexing system 520. As discussed in greater detail below, the search system 560 enables users associated with the computing environment 500 (and possibly also other users) to navigate the data, generate reports, and visualize search results in “dashboards” output using a graphical interface. Using the facilities of the search system 560, users can obtain insights about the data, such as retrieving events from an index, calculating metrics, searching for specific conditions within a rolling time window, identifying patterns in the data, and predicting future trends, among other examples. To achieve greater efficiency, the search system 560 can apply map-reduce methods to parallelize searching of large volumes of data. Additionally, because the original data is available, the search system 560 can apply a schema to the data at search time. This allows different structures to be applied to the same data, or for the structure to23103175Docket No. 102454.9082PCT -25- C / P / 1064223 / WO / SEC / 1be modified if or when the content of the data changes. Application of a schema at search time may be referred to herein as a late-binding schema technique.

[0083] The user interface system 514 provides mechanisms through which users associated with the computing environment 500 (and possibly others) can interact with the data intake and query system 510. These interactions can include configuration, administration, and management of the indexing system 520, initiation and / or scheduling of queries that are to be processed by the search system 560, receipt or reporting of search results, and / or visualization of search results. The user interface system 514 can include, for example, facilities to provide a command line interface or a web-based interface.

[0084] Users can access the user interface system 514 using a computing device 504 that communicates with data intake and query system 510, possibly over a network. A “user,” in the context of the implementations and examples described herein, is a digital entity that is described by a set of information in a computing environment. The set of information can include, for example, a user identifier, a username, a password, a user account, a set of authentication credentials, a token, other data, and / or a combination of the preceding. Using the digital entity that is represented by a user, a person can interact with the computing environment 500. For example, a person can log in as a particular user and, using the user’s digital information, can access the data intake and query system 510. A user can be associated with one or more people, meaning that one or more people may be able to use the same user’s digital information. For example, an administrative user account may be used by multiple people who have been given access to the administrative user account. Alternatively or additionally, a user can be associated with another digital entity, such as a bot (e.g., a software program that can perform autonomous tasks). A user can also be associated with one or more entities. For example, a company can have associated with it a number of users. In this example, the company may control the users’ digital information, including assignment of user identifiers, management of security credentials, control of which persons are associated with which users, and so on.

[0085] The computing device 504 can provide a human-machine interface through which a person can have a digital presence in the computing environment 500 in the form of a user. The computing device 504 is an electronic device having one or more processors and a memory capable of storing instructions for execution by the one or more processors. The23103175Docket No. 102454.9082PCT -26- C / P / 1064223 / WO / SEC / 1computing device 504 can further include input / output (I / O) hardware and a network interface. Applications executed by the computing device 504 can include a network access application 506, such as a web browser, which can use a network interface of the client computing device 504 to communicate, over a network, with the user interface system 514 of the data intake and query system 510. The user interface system 514 can use the network access application 506 to generate user interfaces that enable a user to interact with the data intake and query system 510. A web browser is one example of a network access application. A shell tool can also be used as a network access application. In some examples, the data intake and query system 510 is an application executing on the computing device 504. In such examples, the network access application 506 can access the user interface system 514 without going over a network.

[0086] The data intake and query system 510 can optionally include apps 512. An app of the data intake and query system 510 is a collection of configurations, knowledge objects (a user-defined entity that enriches the data in the data intake and query system 510), views, and dashboards that may provide additional functionality, different techniques for searching the data, and / or additional insights into the data. The data intake and query system 510 can execute multiple applications simultaneously. Example applications include an information technology service intelligence application, which can monitor and analyze the performance and behavior of the computing environment 500, and an enterprise security application, which can include content and searches to assist security analysts in diagnosing and acting on anomalous or malicious behavior in the computing environment 500.

[0087] Though FIG. 5 illustrates only one data source, in practical implementations, the computing environment 500 contains many data sources spread across numerous computing devices. The computing devices may be controlled and operated by a single entity. For example, in an “on the premises” or “on-prem” implementation, the computing devices may physically and digitally be controlled by one entity, meaning that the computing devices are in physical locations that are owned and / or operated by the entity and are within a network domain that is controlled by the entity. In an entirely on-prem implementation of the computing environment 500, the data intake and query system 510 executes on an on-prem computing device and obtains machine data from on-prem data sources. An on-prem implementation can also be referred to as an “enterprise” network, though the term “on-prem” refers primarily to physical locality of a network and who23103175Docket No. 102454.9082PCT -27- C / P / 1064223 / WO / SEC / 1controls that location while the term “enterprise” may be used to refer to the network of a single entity. As such, an enterprise network could include cloud components.

[0088] “ Cloud” or “in the cloud” refers to a network model in which an entity operates network resources (e.g., processor capacity, network capacity, storage capacity, etc.), located for example in a data center, and makes those resources available to users and / or other entities over a network. A “private cloud” is a cloud implementation where the entity provides the network resources only to its own users. A “public cloud” is a cloud implementation where an entity operates network resources in order to provide them to users that are not associated with the entity and / or to other entities. In this implementation, the provider entity can, for example, allow a subscriber entity to pay for a subscription that enables users associated with subscriber entity to access a certain amount of the provider entity’s cloud resources, possibly for a limited time. A subscriber entity of cloud resources can also be referred to as a tenant of the provider entity. Users associated with the subscriber entity access the cloud resources over a network, which may include the public Internet. In contrast to an on-prem implementation, a subscriber entity does not have physical control of the computing devices that are in the cloud, and has digital access to resources provided by the computing devices only to the extent that such access is enabled by the provider entity.

[0089] In some implementations, the computing environment 500 can include on-prem and cloud-based computing resources, or only cloud-based resources. For example, an entity may have on-prem computing devices and a private cloud. In this example, the entity operates the data intake and query system 510 and can choose to execute the data intake and query system 510 on an on-prem computing device or in the cloud. In another example, a provider entity operates the data intake and query system 510 in a public cloud and provides the functionality of the data intake and query system 510 as a service, for example under a Software-as-a-Service (SaaS) model, to entities that pay for the user of the service on a subscription basis. In this example, the provider entity can provision a separate tenant (or possibly multiple tenants) in the public cloud network for each subscriber entity, where each tenant executes a separate and distinct instance of the data intake and query system 510. In some implementations, the entity providing the data intake and query system 510 is itself subscribing to the cloud services of a cloud service provider. As an example, a first entity provides computing resources under a public cloud service model, a second entity23103175Docket No. 102454.9082PCT -28- C / P / 1064223 / WO / SEC / 1subscribes to the cloud services of the first provider entity and uses the cloud computing resources to operate the data intake and query system 510, and a third entity can subscribe to the services of the second provider entity in order to use the functionality of the data intake and query system 510. In this example, the data sources are associated with the third entity, users accessing the data intake and query system 510 are associated with the third entity, and the analytics and insights provided by the data intake and query system 510 are for purposes of the third entity’s operations.

[0090] FIG. 6 is a block diagram illustrating in greater detail an example of an indexing system 620 of a data intake and query system, such as the data intake and query system 510 of FIG. 5. The indexing system 620 of FIG. 6 uses various methods to obtain machine data from a data source 602 and stores the data in an index 638 of an indexer 632. As discussed previously, a data source is a hardware, software, physical, and / or virtual component of a computing device that produces machine data in an automated fashion and / or as a result of user interaction. Examples of data sources include files and directories; network event logs; operating system logs, operational data, and performance monitoring data; metrics; first-in, first-out queues; scripted inputs; and modular inputs, among others. The indexing system 620 enables the data intake and query system to obtain the machine data produced by the data source 602 and to store the data for searching and retrieval.

[0091] Users can administer the operations of the indexing system 620 using a computing device 604 that can access the indexing system 620 through a user interface system 614 of the data intake and query system. For example, the computing device 604 can be executing a network access application 606, such as a web browser or a terminal, through which a user can access a monitoring console 616 provided by the user interface system 614. The monitoring console 616 can enable operations such as: identifying the data source 602 for data ingestion; configuring the indexer 632 to index the data from the data source 632; configuring a data ingestion method; configuring, deploying, and managing clusters of indexers; and viewing the topology and performance of a deployment of the data intake and query system, among other operations. The operations performed by the indexing system 620 may be referred to as “index time” operations, which are distinct from “search time” operations that are discussed further below.

[0092] The indexer 632, which may be referred to herein as a data indexing component, coordinates and performs most of the index time operations. The indexer 632 can be 23103175Docket No. 102454.9082PCT -29- C / P / 1064223 / WO / SEC / 1implemented using program code that can be executed on a computing device. The program code for the indexer 632 can be stored on a non-transitory computer-readable medium (e.g. a magnetic, optical, or solid state storage disk, a flash memory, or another type of non-transitory storage media), and from this medium can be loaded or copied to the memory of the computing device. One or more hardware processors of the computing device can read the program code from the memory and execute the program code in order to implement the operations of the indexer 632. In some implementations, the indexer 632 executes on the computing device 604 through which a user can access the indexing system 620. In some implementations, the indexer 632 executes on a different computing device than the illustrated computing device 604.

[0093] The indexer 632 may be executing on the computing device that also provides the data source 602 or may be executing on a different computing device. In implementations wherein the indexer 632 is on the same computing device as the data source 602, the data produced by the data source 602 may be referred to as “local data.” In other implementations the data source 602 is a component of a first computing device and the indexer 632 executes on a second computing device that is different from the first computing device. In these implementations, the data produced by the data source 602 may be referred to as “remote data.” In some implementations, the first computing device is “on-prem” and in some implementations the first computing device is “in the cloud.” In some implementations, the indexer 632 executes on a computing device in the cloud and the operations of the indexer 632 are provided as a service to entities that subscribe to the services provided by the data intake and query system.

[0094] For a given data produced by the data source 602, the indexing system 620 can be configured to use one of several methods to ingest the data into the indexer 632. These methods include upload 622, monitor 624, using a forwarder 626, or using HyperText Transfer Protocol (HTTP 628) and an event collector 630. These and other methods for data ingestion may be referred to as “getting data in” (GDI) methods.

[0095] Using the upload 622 method, a user can specify a file for uploading into the indexer 632. For example, the monitoring console 616 can include commands or an interface through which the user can specify where the file is located (e.g., on which computing device and / or in which directory of a file system) and the name of the file. The file may be located at the data source 602 or maybe on the computing device where the indexer 632 is23103175Docket No. 102454.9082PCT -30- C / P / 1064223 / WO / SEC / 1executing. Once uploading is initiated, the indexer 632 processes the file, as discussed further below. Uploading is a manual process and occurs when instigated by a user. For automated data ingestion, the other ingestion methods are used.

[0096] The monitor 624 method enables the indexer 632 to monitor the data source 602 and continuously or periodically obtain data produced by the data source 602 for ingestion by the indexer 632. For example, using the monitoring console 616, a user can specify a file or directory for monitoring. In this example, the indexer 632 can execute a monitoring process that detects whenever the file or directory is modified and causes the file or directory contents to be sent to the indexer 632. As another example, a user can specify a network port for monitoring. In this example, a monitoring process can capture data received at or transmitting from the network port and cause the data to be sent to the indexer 632. In various examples, monitoring can also be configured for data sources such as operating system event logs, performance data generated by an operating system, operating system registries, operating system directory services, and other data sources.

[0097] Monitoring is available when the data source 602 is local to the indexer 632 (e.g., the data source 602 is on the computing device where the indexer 632 is executing). Other data ingestion methods, including forwarding and the event collector 630, can be used for either local or remote data sources.

[0098] A forwarder 626, which may be referred to herein as a data forwarding component, is a software process that sends data from the data source 602 to the indexer 632. The forwarder 626 can be implemented using program code that can be executed on the computer device that provides the data source 602. A user launches the program code for the forwarder 626 on the computing device that provides the data source 602. The user can further configure the forwarder 626, for example to specify a receiver for the data being forwarded (e.g., one or more indexers, another forwarder, and / or another recipient system), to enable or disable data forwarding, and to specify a file, directory, network events, operating system data, or other data to forward, among other operations.

[0099] The forwarder 626 can provide various capabilities. For example, the forwarder 626 can send the data unprocessed or can perform minimal processing on the data before sending the data to the indexer 632. Minimal processing can include, for example, adding metadata tags to the data to identify a source, source type, and / or host, among other information, dividing the data into blocks, and / or applying a timestamp to the data.. In some 23103175Docket No. 102454.9082PCT -SI- C / P / 1064223 / WO / SEC / 1implementations, the forwarder 626 can break the data into individual events (event generation is discussed further below) and send the events to a receiver. Other operations that the forwarder 626 may be configured to perform include buffering data, compressing data, and using secure protocols for sending the data, for example.

[0100] Forwarders can be configured in various topologies. For example, multiple forwarders can send data to the same indexer. As another example, a forwarder can be configured to filter and / or route events to specific receivers (e.g., different indexers), and / or discard events. As another example, a forwarder can be configured to send data to another forwarder, or to a receiver that is not an indexer or a forwarder (such as, for example, a log aggregator).

[0101] The event collector 630 provides an alternate method for obtaining data from the data source 602. The event collector 630 enables data and application events to be sent to the indexer 632 using HTTP 628. The event collector 630 can be implemented using program code that can be executing on a computing device. The program code may be a component of the data intake and query system or can be a standalone component that can be executed independently of the data intake and query system and operates in cooperation with the data intake and query system.

[0102] To use the event collector 630, a user can, for example using the monitoring console 616 or a similar interface provided by the user interface system 614, enable the event collector 630 and configure an authentication token. In this context, an authentication token is a piece of digital data generated by a computing device, such as a server, that contains information to identify a particular entity, such as a user or a computing device, to the server. The token will contain identification information for the entity (e.g., an alphanumeric string that is unique to each token) and a code that authenticates the entity with the server. The token can be used, for example, by the data source 602 as an alternative method to using a username and password for authentication.

[0103] To send data to the event collector 630, the data source 602 is supplied with a token and can then send HTTP 628 requests to the event collector 630. To send HTTP 628 requests, the data source 602 can be configured to use an HTTP client and / or to use logging libraries such as those supplied by Java, JavaScript, and .NET libraries. An HTTP client enables the data source 602 to send data to the event collector 630 by supplying the data, and a Uniform Resource Identifier (URI) for the event collector 630 to the HTTP client.23103175Docket No. 102454.9082PCT -32- C / P / 1064223 / WO / SEC / 1The HTTP client then handles establishing a connection with the event collector 630, transmitting a request containing the data, closing the connection, and receiving an acknowledgment if the event collector 630 sends one. Logging libraries enable HTTP 628 requests to the event collector 630 to be generated directly by the data source. For example, an application can include or link a logging library, and through functionality provided by the logging library manage establishing a connection with the event collector 630, transmitting a request, and receiving an acknowledgement.

[0104] An HTTP 628 request to the event collector 630 can contain a token, a channel identifier, event metadata, and / or event data. The token authenticates the request with the event collector 630. The channel identifier, if available in the indexing system 620, enables the event collector 630 to segregate and keep separate data from different data sources. The event metadata can include one or more key -value pairs that describe the data source 602 or the event data included in the request. For example, the event metadata can include key-value pairs specifying a timestamp, a hostname, a source, a source type, or an index where the event data should be indexed. The event data can be a structured data object, such as a JavaScript Object Notation (JSON) object, or raw text. The structured data object can include both event data and event metadata. Additionally, one request can include event data for one or more events.

[0105] In some implementations, the event collector 630 extracts events from HTTP 628 requests and sends the events to the indexer 632. The event collector 630 can further be configured to send events to one or more indexers. Extracting the events can include associating any metadata in a request with the event or events included in the request. In these implementations, event generation by the indexer 632 (discussed further below) is bypassed, and the indexer 632 moves the events directly to indexing. In some implementations, the event collector 630 extracts event data from a request and outputs the event data to the indexer 632, and the indexer generates events from the event data. In some implementations, the event collector 630 sends an acknowledgement message to the data source 602 to indicate that the event collector 630 has received a particular request form the data source 602, and / or to indicate to the data source 602 that events in the request have been added to an index.

[0106] The indexer 632 ingests incoming data and transforms the data into searchable knowledge in the form of events. In the data intake and query system, an event is a single23103175Docket No. 102454.9082PCT -33- C / P / 1064223 / WO / SEC / 1piece of data that represents activity of the component represented in FIG. 6 by the data source 602. An event can be, for example, a single record in a log file that records a single action performed by the component (e.g., a user login, a disk read, transmission of a network packet, etc.). An event includes one or more fields that together describe the action captured by the event, where a field is a key-value pair (also referred to as a name-value pair). In some cases, an event includes both the key and the value, and in some cases the event includes only the value and the key can be inferred or assumed.

[0107] Transformation of data into events can include event generation and event indexing. Event generation includes identifying each discrete piece of data that represents one event and associating each event with a timestamp and possibly other information (which may be referred to herein as metadata). Event indexing includes storing of each event in the data structure of an index. As an example, the indexer 632 can include a parsing module 634 and an indexing module 636 for generating and storing the events. The parsing module 634 and indexing module 636 can be modular and pipelined, such that one component can be operating on a first set of data while the second component is simultaneously operating on a second sent of data. Additionally, the indexer 632 may at any time have multiple instances of the parsing module 634 and indexing module 636, with each set of instances configured to simultaneously operate on data from the same data source or from different data sources. The parsing module 634 and indexing module 636 are illustrated in FIG. 6 to facilitate discussion, with the understanding that implementations with other components are possible to achieve the same functionality.

[0108] The parsing module 634 determines information about incoming event data, where the information can be used to identify events within the event data. For example, the parsing module 634 can associate a source type with the event data. A source type identifies the data source 602 and describes a possible data structure of event data produced by the data source 602. For example, the source type can indicate which fields to expect in events generated at the data source 602 and the keys for the values in the fields, and possibly other information such as sizes of fields, an order of the fields, a field separator, and so on. The source type of the data source 602 can be specified when the data source 602 is configured as a source of event data. Alternatively, the parsing module 634 can determine the source type from the event data, for example from an event field in the event data or using machine learning techniques applied to the event data.23103175Docket No. 102454.9082PCT -34- C / P / 1064223 / WO / SEC / 1

[0109] Other information that the parsing module 634 can determine includes timestamps. In some cases, an event includes a timestamp as a field, and the timestamp indicates a point in time when the action represented by the event occurred or was recorded by the data source 602 as event data. In these cases, the parsing module 634 may be able to determine from the source type associated with the event data that the timestamps can be extracted from the events themselves. In some cases, an event does not include a timestamp and the parsing module 634 determines a timestamp for the event, for example from a name associated with the event data from the data source 602 (e.g., a file name when the event data is in the form of a file) or a time associated with the event data (e.g., a file modification time). As another example, when the parsing module 634 is not able to determine a timestamp from the event data, the parsing module 634 may use the time at which it is indexing the event data. As another example, the parsing module 634 can use a user-configured rule to determine the timestamps to associate with events.

[0110] The parsing module 634 can further determine event boundaries. In some cases, a single line (e.g., a sequence of characters ending with a line termination) in event data represents one event while in other cases, a single line represents multiple events. In yet other cases, one event may span multiple lines within the event data. The parsing module 634 may be able to determine event boundaries from the source type associated with the event data, for example from a data structure indicated by the source type. In some implementations, a user can configure rules the parsing module 634 can use to identify event boundaries.[OHl] The parsing module 634 can further extract data from events and possibly also perform transformations on the events. For example, the parsing module 634 can extract a set of fields (key -value pairs) for each event, such as a host or hostname, source or source name, and / or source type. The parsing module 634 may extract certain fields by default or based on a user configuration. Alternatively or additionally, the parsing module 634 may add fields to events, such as a source type or a user-configured field. As another example of a transformation, the parsing module 634 can anonymize fields in events to mask sensitive information, such as social security numbers or account numbers. Anonymizing fields can include changing or replacing values of specific fields. The parsing module 634 can further perform user-configured transformations.23103175Docket No. 102454.9082PCT .35.C / P / 1064223 / WO / SEC / 1

[0112] The parsing module 634 outputs the results of processing incoming event data to the indexing module 636, which performs event segmentation and builds index data structures.

[0113] Event segmentation identifies searchable segments, which may alternatively be referred to as searchable terms or keywords, which can be used by the search system of the data intake and query system to search the event data. A searchable segment may be a part of a field in an event or an entire field. The indexer 632 can be configured to identify searchable segments that are parts of fields, searchable segments that are entire fields, or both. The parsing module 634 organizes the searchable segments into a lexicon or dictionary for the event data, with the lexicon including each searchable segment (e.g., the field “src=10.10.1.1”) and a reference to the location of each occurrence of the searchable segment within the event data (e.g., the location within the event data of each occurrence of “src=10.10.1.1”) . As discussed further below, the search system can use the lexicon, which is stored in an index file 646, to find event data that matches a search query. In some implementations, segmentation can alternatively be performed by the forwarder 626. Segmentation can also be disabled, in which case the indexer 632 will not build a lexicon for the event data. When segmentation is disabled, the search system searches the event data directly.

[0114] Building index data structures generates the index 638. The index 638 is a storage data structure on a storage device (e.g., a disk drive or other physical device for storing digital data). The storage device may be a component of the computing device on which the indexer 632 is operating (referred to herein as local storage) or may be a component of a different computing device (referred to herein as remote storage) that the indexer 632 has access to over a network. The indexer 632 can manage more than one index and can manage indexes of different types. For example, the indexer 632 can manage event indexes, which impose minimal structure on stored data and can accommodate any type of data. As another example, the indexer 632 can manage metrics indexes, which use a highly structured format to handle the higher volume and lower latency demands associated with metrics data.

[0115] The indexing module 636 organizes files in the index 638 in directories referred to as buckets. The files in a bucket 644 can include raw data files, index files, and possibly also other metadata files. As used herein, “raw data” means data as when the data was produced by the data source 602, without alteration to the format or content. As noted23103175Docket No. 102454.9082PCT -36- C / P / 1064223 / WO / SEC / 1previously, the parsing module 634 may add fields to event data and / or perform transformations on fields in the event data. Event data that has been altered in this way is referred to herein as enriched data. A raw data file 648 can include enriched data, in addition to or instead of raw data. The raw data file 648 may be compressed to reduce disk usage. An index file 646, which may also be referred to herein as a “time-series index” or tsidx file, contains metadata that the indexer 632 can use to search a corresponding raw data file 648. As noted above, the metadata in the index file 646 includes a lexicon of the event data, which associates each unique keyword in the event data with a reference to the location of event data within the raw data file 648. The keyword data in the index file 646 may also be referred to as an inverted index. In various implementations, the data intake and query system can use index files for other purposes, such as to store data summarizations that can be used to accelerate searches.

[0116] A bucket 644 includes event data for a particular range of time. The indexing module 636 arranges buckets in the index 638 according to the age of the buckets, such that buckets for more recent ranges of time are stored in short-term storage 640 and buckets for less recent ranges of time are stored in long-term storage 642. Short-term storage 640 may be faster to access while long-term storage 642 may be slower to access. Buckets may be moves from short-term storage 640 to long-term storage 642 according to a configurable data retention policy, which can indicate at what point in time a bucket is old enough to be moved.

[0117] A bucket’s location in short-term storage 640 or long-term storage 642 can also be indicated by the bucket’s status. As an example, a bucket’s status can be “hot,” “warm,” “cold,” “frozen,” or “thawed.” In this example, hot bucket is one to which the indexer 632 is writing data and the bucket becomes a warm bucket when the index 632 stops writing data to it. In this example, both hot and warm buckets reside in short-term storage 640. Continuing this example, when a warm bucket is moved to long-term storage 642, the bucket becomes a cold bucket. A cold bucket can become a frozen bucket after a period of time, at which point the bucket may be deleted or archived. An archived bucket cannot be searched. When an archived bucket is retrieved for searching, the bucket becomes thawed and can then be searched.

[0118] The indexing system 620 can include more than one indexer, where a group of indexers is referred to as an index cluster. The indexers in an index cluster may also be23103175Docket No. 102454.9082PCT .37.C / P / 1064223 / WO / SEC / 1referred to as peer nodes. In an index cluster, the indexers are configured to replicate each other’s data by copying buckets from one indexer to another. The number of copies of a bucket can be configured (e.g., three copies of each buckets must exist within the cluster), and indexers to which buckets are copied may be selected to optimize distribution of data across the cluster.

[0119] A user can view the performance of the indexing system 620 through the monitoring console 616 provided by the user interface system 614. Using the monitoring console 616, the user can configure and monitor an index cluster, and see information such as disk usage by an index, volume usage by an indexer, index and volume size over time, data age, statistics for bucket types, and bucket settings, among other information.

[0120] FIG. 7 is a block diagram illustrating in greater detail an example of the search system 760 of a data intake and query system, such as the data intake and query system 510 of FIG. 5. The search system 760 of FIG. 7 issues a query 766 to a search head 762, which sends the query 766 to a search peer 764. Using a map process 770, the search peer 764 searches the appropriate index 738 for events identified by the query 766 and sends events 778 so identified back to the search head 762. Using a reduce process 782, the search head 762 processes the events 778 and produces results 768 to respond to the query 766. The results 768 can provide useful insights about the data stored in the index 738. These insights can aid in the administration of information technology systems, in security analysis of information technology systems, and / or in analysis of the development environment provided by information technology systems.

[0121] The query 766 that initiates a search is produced by a search and reporting app 716 that is available through the user interface system 714 of the data intake and query system. Using a network access application 706 executing on a computing device 704, a user can input the query 766 into a search field provided by the search and reporting app 716. Alternatively or additionally, the search and reporting app 716 can include pre-configured queries or stored queries that can be activated by the user. In some cases, the search and reporting app 716 initiates the query 766 when the user enters the query 766. In these cases, the query 766 maybe referred to as an “ad-hoc” query. In some cases, the search and reporting app 716 initiates the query 766 based on a schedule. For example, the search and reporting app 716 can be configured to execute the query 766 once per hour, once per day,23103175Docket No. 102454.9082PCT -38- C / P / 1064223 / WO / SEC / 1at a specific time, on a specific date, or at some other time that can be specified by a date, time, and / or frequency. These types of queries maybe referred to as scheduled queries.

[0122] The query 766 is specified using a search processing language. The search processing language includes commands or search terms that the search peer 764 will use to identify events to return in the search results 768. The search processing language can further include commands for filtering events, extracting more information from events, evaluating fields in events, aggregating events, calculating statistics over events, organizing the results, and / or generating charts, graphs, or other visualizations, among other examples. Some search commands may have functions and arguments associated with them, which can, for example, specify how the commands operate on results and which fields to act upon. The search processing language may further include constructs that enable the query 766 to include sequential commands, where a subsequent command may operate on the results of a prior command. As an example, sequential commands may be separated in the query 766 by a vertical line (“|” or “pipe”) symbol.

[0123] In addition to one or more search commands, the query 766 includes a time indicator. The time indicator limits searching to events that have timestamps described by the indicator. For example, the time indicator can indicate a specific point in time (e.g., 10:00:00 am today), in which case only events that have the point in time for their timestamp will be searched. As another example, the time indicator can indicate a range of time (e.g., the last 24 hours), in which case only events whose timestamps fall within the range of time will be searched. The time indicator can alternatively indicate all of time, in which case all events will be searched.

[0124] Processing of the search query 766 occurs in two broad phases: a map phase 750 and a reduce phase 752. The map phase 750 takes place across one or more search peers. In the map phase 750, the search peers locate event data that matches the search terms in the search query 766 and sorts the event data into field-value pairs. When the map phase 750 is complete, the search peers send events that they have found to one or more search heads for the reduce phase 752. During the reduce phase 752, the search heads process the events through commands in the search query 766 and aggregate the events to produce the final search results 768.

[0125] A search head, such as the search head 762 illustrated in FIG. 7, is a component of the search system 760 that manages searches. The search head 762, which may also be 23103175Docket No. 102454.9082PCT -39- C / P / 1064223 / WO / SEC / 1referred to herein as a search management component, can be implemented using program code that can be executed on a computing device. The program code for the search head 762 can be stored on a non-transitory computer-readable medium and from this medium can be loaded or copied to the memory of a computing device. One or more hardware processors of the computing device can read the program code from the memory and execute the program code in order to implement the operations of the search head 762.

[0126] Upon receiving the search query 766, the search head 762 directs the query 766 to one or more search peers, such as the search peer 764 illustrated in FIG. 7. “Search peer” is an alternate name for “indexer” and a search peer may be largely similar to the indexer described previously. The search peer 764 may be referred to as a “peer node” when the search peer 764 is part of an indexer cluster. The search peer 764, which may also be referred to as a search execution component, can be implemented using program code that can be executed on a computing device. In some implementations, one set of program code implements both the search head 762 and the search peer 764 such that the search head 762 and the search peer 764 form one component. In some implementations, the search head 762 is an independent piece of code that performs searching and no indexing functionality. In these implementations, the search head 762 may be referred to as a dedicated search head.

[0127] The search head 762 may consider multiple criteria when determining whether to send the query 766 to the particular search peer 764. For example, the search system 760 may be configured to include multiple search peers that each have duplicative copies of at least some of the event data and are implanted using different hardware resources q. In this example, the sending the search query 766 to more than one search peer allows the search system 760 to distribute the search workload across different hardware resources. As another example, search system 760 may include different search peers for different purposes (e.g., one has an index storing a first type of data or from a first data source while a second has an index storing a second type of data or from a second data source). In this example, the search query 766 may specify which indexes to search, and the search head 762 will send the query 766 to the search peers that have those indexes.

[0128] To identify events 778 to send back to the search head 762, the search peer 764 performs a map process 770 to obtain event data 774 from the index 738 that is maintained by the search peer 764. During a first phase of the map process 770, the search peer 76423103175Docket No. 102454.9082PCT -40- C / P / 1064223 / WO / SEC / 1identifies buckets that have events that are described by the time indicator in the search query 766. As noted above, a bucket contains events whose timestamps fall within a particular range of time. For each bucket 744 whose events can be described by the time indicator, during a second phase of the map process 770, the search peer 764 performs a keyword search 772 using search terms specified in the search query 766. The search terms can be one or more of keywords, phrases, fields, Boolean expressions, and / or comparison expressions that in combination describe events being searched for. When segmentation is enabled at index time, the search peer 764 performs the keyword search 772 on the bucket’ s index file 746. As noted previously, the index file 746 includes a lexicon of the searchable terms in the events stored in the bucket’s raw data 748 file. The keyword search 772 searches the lexicon for searchable terms that correspond to one or more of the search terms in the query 766. As also noted above, the lexicon incudes, for each searchable term, a reference to each location in the raw data 748 file where the searchable term can be found. Thus, when the keyword search identifies a searchable term in the index file 746 that matches a search term in the query 766, the search peer 764 can use the location references to extract from the raw data 748 file the event data 774 for each event that include the searchable term.

[0129] In cases where segmentation was disabled at index time, the search peer 764 performs the keyword search 772 directly on the raw data 748 file. To search the raw data 748, the search peer 764 may identify searchable segments in events in a similar manner as when the data was indexed. Thus, depending on how the search peer 764 is configured, the search peer 764 may look at event fields and / or parts of event fields to determine whether an event matches the query 766. Any matching events can be added to the event data 774 read from the raw data 748 file. The search peer 764 can further be configured to enable segmentation at search time, so that searching of the index 738 causes the search peer 764 to build a lexicon in the index file 746.

[0130] The event data 774 obtained from the raw data 748 file includes the full text of each event found by the keyword search 772. During a third phase of the map process 770, the search peer 764 performs event processing 776 on the event data 774, with the steps performed being determined by the configuration of the search peer 764 and / or commands in the search query 766. For example, the search peer 764 can be configured to perform field discovery and field extraction. Field discovery is a process by which the search peer23103175Docket No. 102454.9082PCT -41- C / P / 1064223 / WO / SEC / 1764 identifies and extracts key -value pairs from the events in the event data 774. The search peer 764 can, for example, be configured to automatically extract the first 100 fields (or another number of fields) in the event data 774 that can be identified as key-value pairs. As another example, the search peer 764 can extract any fields explicitly mentioned in the search query 766. The search peer 764 can, alternatively or additionally, be configured with particular field extractions to perform.

[0131] Other examples of steps that can be performed during event processing 776 include: field aliasing (assigning an alternate name to a field); addition of fields from lookups (adding fields from an external source to events based on existing field values in the events); associating event types with events; source type renaming (changing the name of the source type associated with particular events); and tagging (adding one or more strings of text, or a “tags” to particular events), among other examples.

[0132] The search peer 764 sends processed events 778 to the search head 762, which performs a reduce process 780. The reduce process 780 potentially receives events from multiple search peers and performs various results processing 782 steps on the received events. The results processing 782 steps can include, for example, aggregating the events received from different search peers into a single set of events, deduplicating and aggregating fields discovered by different search peers, counting the number of events found, and sorting the events by timestamp (e.g., newest first or oldest first), among other examples. Results processing 782 can further include applying commands from the search query 766 to the events. The query 766 can include, for example, commands for evaluating and / or manipulating fields (e.g., to generate new fields from existing fields or parse fields that have more than one value). As another example, the query 766 can include commands for calculating statistics over the events, such as counts of the occurrences of fields, or sums, averages, ranges, and so on, of field values. As another example, the query 766 can include commands for generating statistical values for purposes of generating charts of graphs of the events.

[0133] The reduce process 780 outputs the events found by the search query 766, as well as information about the events. The search head 762 transmits the events and the information about the events as search results 768, which are received by the search and reporting app 716. The search and reporting app 716 can generate visual interfaces for viewing the search results 768. The search and reporting app 716 can, for example, output23103175Docket No. 102454.9082PCT -42- C / P / 1064223 / WO / SEC / 1visual interfaces for the network access application 706 running on a computing device 704 to generate.

[0134] The visual interfaces can include various visualizations of the search results 768, such as tables, line or area charts, Choropleth maps, or single values. The search and reporting app 716 can organize the visualizations into a dashboard, where the dashboard includes a panel for each visualization. A dashboard can thus include, for example, a panel listing the raw event data for the events in the search results 768, a panel listing fields extracted at index time and / or found through field discovery along with statistics for those fields, and / or a timeline chart indicating how many events occurred at specific points in time (as indicated by the timestamps associated with each event). In various implementations, the search and reporting app 716 can provide one or more default dashboards. Alternatively or additionally, the search and reporting app 716 can include functionality that enables a user to configure custom dashboards.

[0135] The search and reporting app 716 can also enable further investigation into the events in the search results 768. The process of further investigation may be referred to as drilldown. For example, a visualization in a dashboard can include interactive elements, which, when selected, provide options for finding out more about the data being displayed by the interactive elements. To find out more, an interactive element can, for example, generate a new search that includes some of the data being displayed by the interactive element, and thus may be more focused than the initial search query 766. As another example, an interactive element can launch a different dashboard whose panels include more detailed information about the data that is displayed by the interactive element. Other examples of actions that can be performed by interactive elements in a dashboard include opening a link, playing an audio or video file, or launching another application, among other examples.

[0136] FIG. 8 illustrates an example of a self-managed network 800 that includes a data intake and query system. “Self-managed” in this instance means that the entity that is operating the self-managed network 800 configures, administers, maintains, and / or operates the data intake and query system using its own compute resources and people. Further, the self-managed network 800 of this example is part of the entity’s on-premise network and comprises a set of compute, memory, and networking resources that are located, for example, within the confines of an entity’s data center. These resources can23103175Docket No. 102454.9082PCT .43.C / P / 1064223 / WO / SEC / 1include software and hardware resources. The entity can, for example, be a company or enterprise, a school, government entity, or other entity. Since the self-managed network 800 is located within the customer’s on-prem environment, such as in the entity’s data center, the operation and management of the self-managed network 800, including of the resources in the self-managed network 800, is under the control of the entity. For example, administrative personnel of the entity have complete access to and control over the configuration, management, and security of the self-managed network 800 and its resources.

[0137] The self-managed network 800 can execute one or more instances of the data intake and query system. An instance of the data intake and query system may be executed by one or more computing devices that are part of the self-managed network 800. A data intake and query system instance can comprise an indexing system and a search system, where the indexing system includes one or more indexers 820 and the search system includes one or more search heads 860.

[0138] As depicted in FIG. 8, the self-managed network 800 can include one or more data sources 802. Data received from these data sources may be processed by an instance of the data intake and query system within self-managed network 800. The data sources 802 and the data intake and query system instance can be communicatively coupled to each other via a private network 810.

[0139] Users associated with the entity can interact with and avail themselves of the functions performed by a data intake and query system instance using computing devices. As depicted in FIG. 8, a computing device 804 can execute a network access application 806 (e.g., a web browser), that can communicate with the data intake and query system instance and with data sources 802 via the private network 810. Using the computing device 804, a user can perform various operations with respect to the data intake and query system, such as management and administration of the data intake and query system, generation of knowledge objects, and other functions. Results generated from processing performed by the data intake and query system instance may be communicated to the computing device 804 and output to the user via an output system (e.g., a screen) of the computing device 804.

[0140] The self-managed network 800 can also be connected to other networks that are outside the entity’s on-premise environment / network, such as networks outside the entity’s 23103175Docket No. 102454.9082PCT .44.C / P / 1064223 / WO / SEC / 1data center. Connectivity to these other external networks is controlled and regulated through one or more layers of security provided by the self-managed network 800. One or more of these security layers can be implemented using firewalls 812. The firewalls 812 form a layer of security around the self-managed network 800 and regulate the transmission of traffic from the self-managed network 800 to the other networks and from these other networks to the self-managed network 800.

[0141] Networks external to the self-managed network can include various types of networks including public networks 890, other private networks, and / or cloud networks provided by one or more cloud service providers. An example of a public network 890 is the Internet. In the example depicted in FIG. 8, the self-managed network 800 is connected to a service provider network 892 provided by a cloud service provider via the public network 890.

[0142] In some implementations, resources provided by a cloud service provider may be used to facilitate the configuration and management of resources within the self-managed network 800. For example, configuration and management of a data intake and query system instance in the self-managed network 800 may be facilitated by a software management system 894 operating in the service provider network 892. There are various ways in which the software management system 894 can facilitate the configuration and management of a data intake and query system instance within the self-managed network 800. As one example, the software management system 894 may facilitate the download of software including software updates for the data intake and query system. In this example, the software management system 894 may store information indicative of the versions of the various data intake and query system instances present in the self-managed network 800. When a software patch or upgrade is available for an instance, the software management system 894 may inform the self-managed network 800 of the patch or upgrade. This can be done via messages communicated from the software management system 894 to the self-managed network 800.

[0143] The software management system 894 may also provide simplified ways for the patches and / or upgrades to be downloaded and applied to the self-managed network 800. For example, a message communicated from the software management system 894 to the self-managed network 800 regarding a software upgrade may include a Uniform Resource Identifier (URI) that can be used by a system administrator of the self-managed network23103175Docket No. 102454.9082PCT .45.C / P / 1064223 / WO / SEC / 1800 to download the upgrade to the self-managed network 800. In this manner, management resources provided by a cloud service provider using the service provider network 892 and which are located outside the self-managed network 800 can be used to facilitate the configuration and management of one or more resources within the entity’s on-prem environment. In some implementations, the download of the upgrades and patches may be automated, whereby the software management system 894 is authorized to, upon determining that a patch is applicable to a data intake and query system instance inside the self-managed network 800, automatically communicate the upgrade or patch to selfmanaged network 800 and cause it to be installed within self-managed network 800. IV. ADDITIONAL CONSIDERATIONS

[0144] Although the present disclosure has been described in certain specific aspects, many additional modifications and variations would be apparent to those skilled in the art. In particular, any of the various processes described above can be performed in alternative sequences and / or in parallel (on the same or on different computing devices) in order to achieve similar results in a manner that is more appropriate to the requirements of a specific application. It is therefore to be understood that the present disclosure can be practiced other than specifically described without departing from the scope and spirit of the present disclosure. Thus, embodiments of the present disclosure should be considered in all respects as illustrative and not restrictive. It will be evident to the person skilled in the art to freely combine several or all of the embodiments discussed here as deemed suitable for a specific application of the disclosure. Throughout this disclosure, terms like “advantageous”, “exemplary” or “example” indicate elements or dimensions which are particularly suitable (but not essential) to the disclosure or an embodiment thereof and may be modified wherever deemed suitable by the skilled person, except where expressly required. Accordingly, the scope of the disclosure should be determined not by the embodiments illustrated, but by the appended claims and their equivalents.

[0145] Any reference to an element being made in the singular is not intended to mean “one and only one” unless explicitly so stated, but rather “one or more.” All structural and functional equivalents to the elements of the above-described preferred embodiment and additional embodiments as regarded by those of ordinary skill in the art are hereby expressly incorporated by reference and are intended to be encompassed by the present claims.23103175Docket No. 102454.9082PCT -46- C / P / 1064223 / WO / SEC / 1

[0146] Moreover, no requirement exists for a system or method to address each and every problem sought to be resolved by the present disclosure, for solutions to such problems to be encompassed by the present claims. Furthermore, no element, component, or method step in the present disclosure is intended to be dedicated to the public regardless of whether the element, component, or method step is explicitly recited in the claims. Various changes and modifications in form, material, workpiece, and fabrication material detail can be made, without departing from the spirit and scope of the present disclosure, as set forth in the appended claims, as might be apparent to those of ordinary skill in the art, are also encompassed by the present disclosure.23103175

Claims

Docket No. 102454.9082PCT .47.C / P / 1064223 / WO / SEC / 1CLAIMSWhat is claimed is:

1. A computer-implemented method, comprising:generating a dataset of questions that mimic user queries by a multi-agent framework, wherein the multi-agent framework includes deployment of a plurality of artificial intelligence (Al) agents;running a subset of the dataset of questions through a first agent of the plurality of Al agents where each running of a question produces a trajectory which is a sequence of tools that the first agent calls;selecting a most common trajectory (MCT) for each question across all runnings, wherein the MCT for each question is annotated to indicate a correctness of the MCT; reverse engineering alternate questions from the responses for trajectory verification;extracting features from the MCT for each question that has been annotated and the alternate questions; andtraining a discriminative machine learning model on the features.

2. The computer-implemented method of claim 1, wherein the plurality of Al agents utilizes an API-based LLM.

3. The computer-implemented method of claim 1, wherein the plurality of Al agents includes an investigator agent configured for query generation, an assistant agent configured for answering queries, wherein the assistant agent is the first Al agent, and a reverse engineer agent configured for question generation based on a given response.

4. The computer-implemented method of claim 1, wherein running the subset of the dataset of questions through the first agent includes performing in-context learning (ICL), wherein a first ICL sample includes a trajectory Tl, a question that was answered using that trajectory QI, and a second trajectory T2 randomly sampled from a seed dataset, and wherein the investigator agent executes T2 and is tasked with writing a plurality of questions similar in style to QI by studying the tool responses, which results in a set of synthetic questions, wherein set of synthetic questions is filtered down to a smaller sample size resulting in the subset of the questions.23103175Docket No. 102454.9082PCT -48- C / P / 1064223 / WO / SEC / 15. The computer-implemented method of claim 1, wherein training a discriminative machine learning model utilizes stratified ten-fold cross-validation thereby accounting for label imbalance.

6. A computing device, comprising:a processor; anda non-transitory computer-readable medium having stored thereon instructions that, when executed by the processor, cause the processor to perform operations including:generating a dataset of questions that mimic user queries by a multi-agent framework, wherein the multi-agent framework includes deployment of a plurality of artificial intelligence (Al) agents,running a subset of the dataset of questions through a first agent of the plurality of Al agents where each running of a question produces a trajectory which is a sequence of tools that the first agent calls,selecting a most common trajectory (MCT) for each question across all runnings, wherein the MCT for each question is annotated to indicate a correctness of the MCT,reverse engineering alternate questions from the responses for trajectory verification;extracting features from the MCT for each question that has been annotated and the alternate questions, andtraining a discriminative machine learning model on the features.

7. The computing device of claim 6, wherein the plurality of Al agents utilizes an API-based LLM.

8. The computing device of claim 6, wherein the plurality of Al agents includes an investigator agent configured for query generation, an assistant agent configured for answering queries, wherein the assistant agent is the first Al agent, and a reverse engineer agent configured for question generation based on a given response.

9. The computing device of claim 6, wherein running the subset of the dataset of questions through the first agent includes performing in-context learning (ICL), wherein a first ICL sample includes a trajectory Tl, a question that was answered using that trajectory QI, and a second trajectory T2 randomly sampled from a seed dataset, and 23103175Docket No. 102454.9082PCT .49.C / P / 1064223 / WO / SEC / 1wherein the investigator agent executes T2 and is tasked with writing a plurality of questions similar in style to QI by studying the tool responses, which results in a set of synthetic questions, wherein set of synthetic questions is filtered down to a smaller sample size resulting in the subset of the questions.

10. The computing device of claim 6, wherein training the discriminative machine learning model utilizes stratified ten-fold cross-validation thereby accounting for label imbalance.

11. A non-transitory computer-readable medium having stored thereon instructions that, when executed by one or more processors, cause the one or more processors to perform operations including:generating a dataset of questions that mimic user queries by a multi-agent framework, wherein the multi-agent framework includes deployment of a plurality of artificial intelligence (Al) agents;running a subset of the dataset of questions through a first agent of the plurality of Al agents where each running of a question produces a trajectory which is a sequence of tools that the first agent calls; andselecting a most common trajectory (MCT) for each question across all runnings, wherein the MCT for each question is annotated to indicate a correctness of the MCT;reverse engineering alternate questions from the responses for trajectory verification;extracting features from the MCT for each question that has been annotated and the alternate questions; andtraining a discriminative machine learning model on the features.

12. The non-transitory computer-readable medium of claim 11, wherein the plurality of Al agents utilizes an API-based LLM.

13. The non-transitory computer-readable medium of claim 11, wherein the plurality of Al agents includes an investigator agent configured for query generation, an assistant agent configured for answering queries, wherein the assistant agent is the first Al agent, and a reverse engineer agent configured for question generation based on a given response.23103175Docket No. 102454.9082PCT -SO- C / P / 1064223 / WO / SEC / 114. The non-transitory computer-readable medium of claim 11, wherein running the subset of the dataset of questions through the first agent includes performing in-context learning (ICL), wherein a first ICL sample includes a trajectory Tl, a question that was answered using that trajectory QI, and a second trajectory T2 randomly sampled from a seed dataset, and wherein the investigator agent executes T2 and is tasked with writing a plurality of questions similar in style to QI by studying the tool responses, which results in a set of synthetic questions, wherein set of synthetic questions is filtered down to a smaller sample size resulting in the subset of the questions.

15. The non-transitory computer-readable medium of claim 11 , wherein training the discriminative machine learning model utilizes stratified ten-fold cross-validation thereby accounting for label imbalance.23103175

Images