CONTEXT:
These trajectories show a software engineering agent trying to fix a bug or implement a feature. The agent can use various tools including subagents (specialized AI assistants) to help solve the problem.

TRAJECTORIES:
{{TRAJECTORIES}}

TOOL TO ANALYZE: {{TOOL_NAME}}

Your task is to determine if the subagent "{{TOOL_NAME}}" was helpful in this set of trajectories. 
A tool is considered helpful if:
1. It was called/invoked by the main agent in the main agent trajectory
2. It provided useful information, analysis, or insights that contributed to solving the problem
3. The main agent made progress after using this tool (e.g., identified the issue, made code changes, validated results, etc.)
4. It completed its task as intended (followed proper analysis process, not just got lucky results)

Look for positive evidence such as:
- The subagent being called with appropriate parameters
- The subagent providing insights that led to code changes or problem understanding
- The main agent referencing or building upon the subagent's output
- The subagent's output being used in subsequent reasoning or actions

Look for negative evidence such as:
- The subagent not being called by the main agent, or called incorrectly
- The subagent providing irrelevant or incorrect information that was not later used
- The subagent's response was valid but did not move the main agent closer to resolving the problem.
- The subagent failed to execute properly or had many errors during the its run.
- The subagent's output appeared correct, but its trajectory did not actually achieve those results (e.g., claimed to test code but just reported all tests passed).
- The main agent had to call the subagent over and over again to get the proper results.
- The subagent trajectory was unnecessarily long or verbose, taking many steps to complete its task
- The main agent trajectory became inefficient due to excessive subagent calls or overly verbose subagent responses
- The subagent's results did not actually help the main agent make progress in resolve the issue. If a subagent did not contribute to producing the correct patch, e.g. only improved performance, style, or documentation, this is NOT helpful.

Respond with YAML format (exactly):
```yaml
helpful: true/false
reasoning: |
  Brief explanation of why the tool was or wasn't helpful, including specific evidence from the trajectories
```

- Always use the block scalar `|` for `reasoning` and indent its text by two spaces.
- Only respond with the YAML block; no additional text before or after.