RunExecutionContext

Run Execution Context

Encapsulates the complete execution information for a single Run.

Core Concepts:

• Task: Represents a user's original request (business-level unique identifier)

• Run: A specific execution of the same Task (experiment/training level)

• RunExecutionContext: Encapsulates the execution context for one Run

Relationship Example:

Task "task-001" (User asks: "What's the weather today?")
├─ Run "0" → RunExecutionContext (msgIds=[msg-1, msg-2], msgs=[...], reward=0.8)
├─ Run "1" → RunExecutionContext (msgIds=[msg-3, msg-4], msgs=[...], reward=0.9)
└─ Run "2" → RunExecutionContext (msgIds=[msg-5, msg-6], msgs=[...], reward=0.85)

Data Contents:

• Task ID: Unique identifier for user request

• Run ID: The Nth execution of this Task (incrementing from "0")

• msg_ids: All message IDs from LLM interactions in this Run (returned from Trinity API)

• msgs: All messages from Agent memory in this Run (raw data for reward calculation)

• Timing info: Start time and duration of this Run

Lifecycle:

1. TrainingRouter creates RunExecutionContext (corresponds to one Run)
2. Passed to TrinityModelAdapter (automatically collects msg_ids)
3. Shadow Agent execution completes, all data collected
4. TrainingRouter calculates reward
5. TrainingRouter submits Feedback using (taskId, runId, msg_ids, reward)
6. Registered to TaskExecutionRegistry (optional, for later queries)

Thread Safety:

• Uses CopyOnWriteArrayList to store msg_ids

• Supports concurrent additions (although currently executed synchronously)