Result of an evaluation.

A custom exception class for CaMeL, wrapping a standard exception with additional context. Attributes: exception: The original exception

Represents an integer value in CaMeL.

Represents the None value in CaMeL.

Sub-agent: Manages rule_base.json read/write operations.

Represents a floating point number in CaMeL.

Adaptive Learning Framework Engine - Phase 1 (Expert-Led Correction). The engine loads a structured rule base and evaluates rules against

Orchestrator for the benchmark agent.

A single result yielded by the VTO pipeline.

Mock Vision API response for testing.

Rule compliance validation for a single phase

Encapsulates the agent and runner for the Presentation Expert. Initializes all clients, services, and the main orchestrating agent.

Defines the content and structure for a single presentation slide. This is the single source of truth for both planning and final output.

A single result yielded by the glasses VTO pipeline.

Allowed access to the tool.

Represents a product in a customer's purchase history.

Structured output from RuleWriterAgent.

Represents a set in CaMeL.

Validation of data source usage per Section 0

Denied access to the tool.

Represents the input for a chat request with optional configuration.

Represents a list in CaMeL.

Represents a tuple in CaMeL.

Configuration settings for the customer service agent.

The full specification for a presentation deck.

Structured output from ImpactAssessorAgent.

Service for managing process states in Firestore.

Represents a single correction rule in the Structured Rule Base. A rule has: - id: Unique identifier (e.g., "ALF-001") - na

Agent that uses ADK to interact with LLMs and tools.

Detailed evaluation result for a single asset.

Audit Logging Agent

Represents an iterator value in CaMeL.

Represents a string in CaMeL.

Manages Google Calendar API operations for the business calendar.

Sub-agent: Loads all artifacts for a case and builds context. This replicates the context-building logic from ALF/alf_engine.py Acti

A record for storing a question, its ground truth answer, and justification.

Result of running an evaluation test case.

Class for the Gemini model.

The verdict for a specific guideline.

Represents a customer's purchase.

A wrapper around VertexAiCodeExecutor that implements __deepcopy__ and __reduce__ to fix RecursionError and SerializationError during agent d

Configuration for the User Story Refiner Agent.

Detects when agent intentionally bypassed a validation step. v2.0.0: Bypass indicators are now loaded from RuleDiscoveryEngine (Layer 2)

Represents an instance of a class in CaMeL.

Represents a dictionary in CaMeL.

Represents a function in CaMeL.

One case's match result.

A criterion rubric extended with the model's verdict.

Validates agent used correct data sources per Section 0 of reconstructed_rules_book.md. Critical Rule: Use EXTRACTION for invoice content

Evaluation service.

Subclass of Gemini to ensure location is passed to the internal Client.

Sub-agent: Evaluates proposed conditions against all cases. Deterministic - no LLM calls. Uses ALF ConditionEvaluator directly.

LLM-based rule validation using reconstructed_rules_book.md. Key differences from original: - Uses detailed sections from reconstruc

Typed accessor for master_data.yaml contents.

Final Output Generation Agent

Validates a case against discovered rule groups using LLM with ultra-conservative checks. Layer 3 of the 3-layer validation architecture:

Structured output from RuleDiscovererAgent.

Extracts sections from reconstructed_rules_book.md format. The reconstructed rules book uses format like: - ## 0. Data Source Priori

Analysis of a specific rejection pattern

Discovers validation rules from reconstructed_rules_book.md using LLM at startup. Layer 2 of the 3-layer validation architecture: -

Event wrapper that hides final response status. This is used to prevent parent agents from breaking their execution loop prematurely when a s

Extracts tolerance values from rules book dynamically. Supports extracting: - Numerical tolerances (e.g., "0.5 hours", "3%", "$1.00"

Line Item Transformation Agent

Raised when Veo returns an empty result.

Represents a customer's address.

Long running tool that escalates and stops invocation for ADK Agent

Complete investigation of agent processing for one case

Configuration for the Technical Designer Agent.

Configuration for the Agent.

Configuration for the Task Planner Agent.

Agent Engine application wrapping the ADK app.

Evaluates agents against test cases.

Aggregate analysis across all cases

Agent model settings.

Agent model settings.