Free direct download · template
Model Comparison Test Sheet
A comparable record of inputs, criteria, evidence, and repair effort.
Before you open it
What it helps with
A matched-condition scoring sheet for bounded model comparisons.
- Best for
- People comparing model workflows for one bounded task.
- Problem it addresses
- Unmatched prompts and vague criteria turn model comparisons into preference claims.
Set up a fair comparison
Use one row per run and keep the prompt, account tier, enabled tools, visible model, and test date explicit. Compare bounded conditions, not vague impressions from unrelated tasks.
Record evidence with every score
For each criterion, record a 1–5 score, the evidence that supports it, repair minutes, and reviewer notes. The sheet is designed to make score disagreements inspectable rather than to create a winner by default.
Read the result cautiously
Compare the resulting rows against the task’s decision and constraints. Preserve the runs, source material, and manual interventions so another reviewer can repeat or challenge the finding.
Keep the limits visible
The sheet does not establish general model quality or predict performance on another task, account, version, or date. A human reviewer still owns the conclusion.
Compatibility
- Any text editor
- Human review required
License:CC BY 4.0. Review the license before reuse.
Changelog
1.0.0 · 2026-08-06
- Initial human-reviewed launch release.
