The report

Scoring one file tells you about one file. The report tells you where the whole repo stands: which prompts have ever been measured, what they scored, and when.

/hyperpower:eval --report

It writes an HTML page listing every prompt file grouped by folder, each marked recorded with its weighted score and date, or none. The ones marked none are the useful part: a change to any of them cannot be scored against anything yet.

It lands in .hyperpower/evals/report.html, which git ignores because it is generated. To keep it in the repo, write it somewhere tracked:

../scripts/run_evals.py report --out docs/evals.html

The page says this on itself, and it is worth repeating: a baseline records how a prompt scored when it was measured. It does not say the prompt is correct.