Listening field note
What columns should a transcript CSV contain?
A delimited transcript needs a text column and may include start, end, and speaker columns; row widths, quoting, duplicate headers, time shapes, byte limits, and end-after-start rules should be validated atomically.
Accept a bounded vocabulary
Recognize documented header aliases rather than guessing from every possible spreadsheet. Write the decision in plain language and retain the original source separately so the index can be audited against something outside the browser project.
Fail the whole import on structural errors
Partial rows can shift speaker, time, and text into the wrong fields. Reject ragged or malformed input instead of salvaging silently. Review the result as a retrieval aid rather than a conclusion. Human-authored fields must remain visibly distinct from exact source-derived text.
Neutralize exported spreadsheet formulas
When exporting user-authored text to CSV, prefix cells whose first meaningful character could trigger formula execution. Carry the limitation into the frozen edition and passive handoff so a recipient can understand what was checked, what was not checked, and what remains their responsibility.