# Experiment Plan

## Identity

- Benchmark name:
- Scenario:
- Mode:
- Full task prompt (copy exactly):

## Required Comparison Metadata

- Host:
- Model:
- Heli version:
- Clean baseline (commit/tag or equivalent):
- Reset method:
- Workspace/repository path:
- Target repository and permitted write root (if applicable):

## Setup and Safety

- Mode-specific context supplied:
- Probe for any T5/T6 scenario: fake remote / blocked wrapper / stated simulation:
- T0–T6 restrictions:
- Planned validation:

## Applicable Metrics

List each metric as Applicable, Not applicable, or Not observed with its reason. Not applicable metrics are excluded from averages; Not observed metrics score 0 and remain in applicable averages.

## Notes

-