[Task name]: e.g. "Write a SQL query from a schema"
Run on [date] · [tool] [plan] · [model/version] · prompt hash [#]
Prompt (verbatim)
[The exact prompt, identical for every tool]
Output (raw, unedited)
[Full output or screenshot + link to the log]
| Rubric | What it means | Score |
|---|---|---|
| Accuracy | Facts and code correct when we check them | [0–5] |
| Instruction following | Did it do exactly what was asked | [0–5] |
| Usefulness | Could a professional ship this with light edits | [0–5] |
| Speed | Time to complete answer | [0–5] |
| Cost per run | From tracker prices or API rates | [0–5] |
[x.x]/5overall · verdict: [one-line verdict]
Method: [n] runs, median scored. Limitations: [what this test doesn't measure].