A is incorrect: Suggesters provide query auto-completion and suggested search terms, enhancing the user experience, but they do not manage term equivalency within search results.

B is correct: This feature allows administrators to define sets of equivalent terms, ensuring that a search for any term in the set will return results containing any of the other terms.

C is incorrect: Skillsets are used for enriching data by integrating AI capabilities like entity recognition or image analysis, not for handling linguistic equivalencies between search terms.

D is incorrect: Scoring profiles influence how search results are ranked based on criteria like recency or specific field matches, rather than bridging semantic gaps between different search terms.

A is incorrect: LLM nodes are used for interactions with large language models, such as generating text or performing classification, but they don't inherently provide the infrastructure for arbitrary conditional branching within the flow.

B is incorrect: Index nodes typically interact with search indexes or vector stores to retrieve relevant information, rather than implementing conditional logic for flow control.

C is incorrect: Tool nodes are designed for integrating external functions or APIs into the flow, not for direct conditional routing based on internal logic.

D is correct: This node type allows for custom code execution, enabling developers to write conditional statements that control the flow's path based on input data or prior processing results.

A is incorrect: While RAG with a vector store is excellent for retrieving relevant information, it's primarily for answering specific questions from documents, not for summarizing an entire large document efficiently in terms of token usage for the summary itself.

B is incorrect: Asking a model to guess or infer information without providing sufficient context from the entire document would lead to inaccurate or incomplete summaries and is not a viable or efficient strategy for this task.

C is correct: This strategy involves breaking down the large document into smaller segments, summarizing each segment independently, and then consolidating these intermediate summaries into a final, comprehensive summary. This method effectively manages token limits by processing information in manageable chunks.

D is incorrect: Submitting a document exceeding 1,000 pages in a single prompt would far surpass the token limits of most language models, leading to an error or truncation, and is therefore highly inefficient and impractical.

A is correct: Since the Azure AI Agent Service does not natively provide a `last_accessed` timestamp, the recommended practice is for developers to store a custom timestamp (e.g., `last_interaction`) in the thread's metadata, updating it with each user or agent message.

B is incorrect: The `list_messages` count indicates the total number of messages in a thread, which does not directly reveal the time of the last interaction.

C is incorrect: The `created_at` timestamp indicates when the thread was initiated, not when it was last interacted with, so it cannot be used to determine inactivity.

D is incorrect: The `purge` API is used to delete threads, not to identify them based on inactivity criteria.

A is incorrect: Hierarchical communication structures involve agents organized in a tree-like fashion, with communication flowing up and down defined levels, rather than through an open shared space.

B is correct: This pattern utilizes a shared data repository where agents can read and write information, allowing for indirect communication, collaborative problem-solving, and multiple agents responding to contributions.

C is incorrect: A linear communication pattern implies a sequential flow where agents communicate one after another, not a shared broadcast model.

D is incorrect: Parallel communication involves agents acting concurrently, often independently or with direct peer-to-peer exchanges, but not necessarily through a central shared message board for collective response.

A is incorrect: The web UI hosting is handled by web application services or infrastructure components, distinct from the role of a Service Connection.

B is correct: Service Connections are specifically designed to centralize and securely manage access tokens, API keys, and connection details for external platforms or services used by AI projects.

C is incorrect: Indexing capabilities for documents like PDFs are provided by specific search services or data processing pipelines, not directly by Service Connections.

D is incorrect: Storing LLM weights is not the primary function of a Service Connection; these are typically part of model deployment or fine-tuning assets.

A is incorrect: Text Moderation analyzes content for harmfulness or inappropriate language, which is distinct from protecting the integrity of system prompts.

B is incorrect: Severity levels are used for classifying the impact or risk of detected content, not for actively preventing prompt exfiltration.

C is incorrect: RBAC manages user permissions to resources and actions, but it does not directly prevent a model from revealing its internal prompt instructions during interaction.

D is correct: Prompt Shields are specifically implemented to counteract 'System Prompt Injection' and 'Exfiltration' (leaking) efforts, safeguarding the underlying instructions given to the model.

A is incorrect: A system prompt provides instructions and context to the LLM but cannot supply dynamic, real-time data such as the current date and time.

B is incorrect: While a code interpreter can execute code, directly connecting to a reliable external time source via a dedicated function tool is more robust and secure for this specific need.

C is correct: Large Language Models lack an inherent clock; therefore, a Function Tool that makes external calls to a system clock or Network Time Protocol (NTP) API is the most accurate method for fetching current, real-time date and time information.

D is incorrect: A vector store is used for semantic search and retrieval of information, not for obtaining real-time temporal data.

A is correct: A Router agent's specific function is to analyze the incoming request or query and subsequently direct it to the most suitable tool, sub-agent, or processing path based on the inferred user intent.

B is incorrect: A Summarizer agent focuses on condensing information or conversation history, which is unrelated to tool selection based on user intent.

C is incorrect: An Executor agent is responsible for carrying out the actions or using the tools that have already been selected or planned, not for the selection process itself.

D is incorrect: A Planner agent typically breaks down complex tasks into a sequence of sub-tasks and determines the overall execution flow, rather than directly selecting individual tools for an initial intent.

A is incorrect: DALL-E 2 can generate images, but it struggles significantly with rendering coherent or readable text within those images compared to its successor, DALL-E 3.

B is incorrect: GPT-4o is a multimodal large language model, excelling in text and vision understanding, but its primary function is not high-quality image generation with embedded text like DALL-E 3.

C is incorrect: Midjourney is an image generation model, but it is not an Azure OpenAI model, and DALL-E 3 generally offers superior text rendering capabilities within images.

D is correct: DALL-E 3 exhibits a significantly enhanced ability to adhere to intricate instructions and render highly legible text within the images it generates, surpassing its predecessors and other models in this capability.