# My Arena experiment worksheet

This is a blank worksheet, not a completed experiment or benchmark. Keep it where you choose. Share only material you have permission to share; do not include secrets or private data.

## Scope and controls

- Date(s) and task:
- Input or dataset and version:
- Comparison type (model, harness, instruction, skill, integration or complete stack):
- What I kept the same, including tool permissions and settings:
- What changed, plus any other differences I could not control:
- Independent checks and expected results, including field-specific quality requirements:
- Repetition plan, including how failures and unassessed runs will be recorded:

## Configurations and observations

For each setup, record the harness, model, versions, instructions, skills, integrations and permissions actually used. Write “not shown” for details you cannot verify. Keep output artifacts and dates only where you are allowed to share them.

| Setup | Run | Date | Checks passed / attempted or not assessed | Actual cost and currency, or unknown | Actual seconds, or unknown | Output, failure or uncertainty |
| --- | --- | --- | --- | --- | --- | --- |
| A | 1 | | | | | |
| B | 1 | | | | | |

Add a row for every run, including failed or unassessed runs. Blank cost or timing is unknown, not zero. A zero is appropriate only when you actually know the value is zero. Use the same checks for a fair comparison; equal check counts alone do not prove checks were equivalent.

## Interpretation and limits

- Results and how the checks were verified:
- Costs, time and measurements that were unavailable:
- Failed or unassessed runs and possible confounders:
- What this experiment does not establish, and what to repeat or review next:

This worksheet does not establish professional approval, safety, or a universal ranking. Field-specific work may require qualified independent review.
