[lbarrosby]

When an amount has cents, _money in normalization/english.py builds the spoken whole part from int(math.floor(number)) and the bill word from plural(bill, count=number):

input	before	after
-$1.50	minus two dollars and fifty cents	minus one dollar and fifty cents
-£3.75	minus four pounds and seventy-five pence	minus three pounds and seventy-five pence
$1.05	one dollars and five cents	one dollar and five cents
€1.20	one euros and twenty cents	one euro and twenty cents

floor rounds a negative amount down, so every negative amount with cents was read one unit too high. inflect pluralizes for any count other than 1, so 1.05 always got the plural. Whole amounts (-$5, $1.00) take the other branch and were already right.

The cents branch now reads the sign on its own (int(-0.25) would drop it), uses int(abs(number)) for the words and the plural, and takes the cents from the absolute value. math is no longer used in the module, so its import is gone.

Testing

• uv run pytest passes locally (API changes)
• npm run test:web passes locally (web player changes)
• New or changed behavior has a test
• Ran in Docker (npm run cpu:up / gpu:up), for inference or docker changes
• CHANGELOG.md entry, if users will notice the change

Five new CASES in api/tests/test_normalizer.py (the four rows above plus -$0.25, which keeps "minus zero dollars and twenty-five cents"). All five fail on master and pass with the fix. I ran test_normalizer.py, test_normalizer_contract.py and test_normalizer_properties.py (159 passed) in a small venv with the locked inflect 7.5.0, without torch or model weights, so I did not tick the full uv run pytest box. ruff check and ruff format --check are clean.

AI disclosure: written with the help of Claude Code, reviewed and tested by me.

[remsky]

Thanks @lbarrosby, good fix and coverage.

Small nit: think this brings in duplicate bills handling at english.py:318, but it won't have any perf impact. Will give it a test and tidy in a follow-up if needed.