Review The Evidence Before The Conclusion
A comparison can look complete even if the candidates did different work, a required test never ran, or a passing result belongs to an earlier setup. Check the records behind each claim before repeating it.
This exercise uses invented tools and test records. It is not a product ranking or a benchmark.
Get The Pack
Give the AI only the test input. Keep the review key and worksheet separate, and retain the first answer before requesting corrections.
Download review worksheet (.txt)
Check Three Comparison Claims
| Case | Claim to check | What the records support |
|---|---|---|
| CMP-01 | Cedar is faster than Slate. | Both times were observed, but for different input sizes. Rerun the same task with the same input before comparing speed. |
| CMP-02 | Slate failed the content check. | The input could not be submitted. There is no Slate output to evaluate. |
| CMP-03 | Cedar passes with the current preset. | The pass was recorded with the earlier preset only. Test the current preset before claiming a current pass. |
Keep the measured times, the access blocker, and the historical pass. Correcting the comparison does not require throwing out the underlying evidence or declaring a winner.
Compare A Later Review
Use the same input on a later run. Compare which records the AI preserved, which claims it limited, and which next tests it proposed. Without a comparable earlier answer, start a baseline.
Return to the comparison practice pack or choose your next check.
For a real evaluation, keep original outputs and test conditions in an approved location. See the two-tool comparison workflow for a repeatable way to investigate different outputs.