Address feedback from PR; move integration test, drop note to performance in the repo itself instead of just PR

Author: nonprofittechy

README.md

@@ -48,6 +48,9 @@ promptfoo eval -c promptfooconfig.yaml
 
 The eval uses `formfyxer/tests/passive_voice_test_dataset.csv` to ensure sentences labeled as passive produce a non-empty `fragments` array. If you change the prompt inside `formfyxer/passive_voice_detection.py`, copy the updated text (the `system_prompt` and numbering format) into `promptfooconfig.yaml` so the evaluation mirrors runtime behavior.
 
+Raw percentage correct performance using gpt-5 nano on the benchmark dataset is **95.56%.** (Did not
+calculate more specific F1 score).
+
 ## Functions
 
 Functions from `pdf_wrangling` are found on [our documentation site](https://suffolklitlab.org/docassemble-AssemblyLine-documentation/docs/reference/formfyxer/pdf_wrangling).