[lbarrosby]

_money in normalization/english.py treats every fractional part as cents: it takes the digits after the dot as an integer and reads them as the coin. That only works for one or two decimals. Prices with more decimals are common (US fuel prices, per-token or per-unit rates, crypto):

input	before	after
Gas is $3.499 a gallon	three dollars and four hundred and ninety-nine cents	three point four nine nine dollars
$0.0012 per token	zero dollars and twelve cents	zero point zero zero one two dollars
-$1.005	minus one dollar and five cents	minus one point zero zero five dollars

The $0.0012 case is the worst one, because the amount gets 10,000 times bigger when read aloud.

When the matched number has more than two decimals and is not a whole amount, the amount is now read as a decimal number followed by the plural bill, the same way $1.5 million is already read. The words come from the matched digits rather than from the float, so small values never go through scientific notation. Whole amounts with trailing zeros (£2.000) still take the existing branch and stay "two pounds". One and two decimals ($5.5, $5.50, $1.05) are unchanged.

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

Four new CASES in api/tests/test_normalizer.py (the three rows above plus £2.000). The three rows fail on master and pass with the fix. I ran test_normalizer.py, test_normalizer_contract.py and test_normalizer_properties.py in a small venv with the locked inflect 7.5.0, without torch or model weights (178 passed; the two test_normalizer_redos.py cases that import the full text processor need torch and fail the same way on master), 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]

Looks good, thanks @lbarrosby