Software Auto-Punctuation and a Numbers-Heavy Damages Stipulation
Your transcription platform's AI auto-punctuation and auto-numbering feature is enabled. You are transcribing a 22-minute stipulation hearing in which counsel recites damages: "forty-two thousand five hundred and sixty dollars," "point zero eight percent," "section 1983," and "exhibit twelve-B." The software has rendered "$42,560" correctly twice but once output "$4,2560" and once "$42.560." It also auto-capitalized "Section" mid-sentence and changed "point zero eight" to "0.8" (dropping a zero). Your quality checklist requires a 100% number-verification pass. You have 30 minutes before the draft is due to the supervising attorney.
1. What is your verification workflow for the numbers, and why is spot-checking insufficient here?
Read the worked answer
Run a targeted search for every currency symbol, decimal point, and the word 'section' rather than reading linearly. For each hit, compare against the audio at that timestamp and confirm digit-by-digit. Spot-checking fails because the errors are systematic and intermittent: the same phrase rendered correctly twice and wrong once means the failure mode is context-dependent, so any unchecked instance could be the bad one. Specifically: "$4,2560" and "$42.560" are both malformed and must be corrected to $42,560; "0.8" must be restored to 0.08 because dropping the leading zero changes the percentage by a factor of ten - a material error in a damages record. Verify "section 1983" is lowercase mid-sentence per style guide.
2. Should you disable the auto-features for the remainder of the job, and how do you document the corrections?
Read the worked answer
Yes - disable auto-numbering and auto-punctuation for the remaining audio, because a feature that intermittently corrupts numerals is a net risk on a numbers-heavy record; manual entry with verification is safer. Retain a correction log noting each timestamp, the erroneous output, and the corrected value, so the supervising attorney can see the scope of machine error. After correction, do a second full read of the numbers section only, because fixing one error can introduce another (e.g., re-typing a figure). Disclose in your delivery note that AI features were used and then disabled, and that all numerals were manually verified against audio.