Keep a copy before removing structure
Whitespace can encode indentation, paragraph breaks or table alignment. Start by trimming only what is unnecessary. Deduplicate complete lines only when that matches your purpose; CSV records need table-aware tools. Compare results with the original text.
Check characters, not just appearance
Visually identical words may use different Unicode representations. Normalization is not translation; compatibility normalization and accent removal can alter spelling or meaning. Keep correctly spelled display text even if you create simplified matching keys.
Review timing and supported markup
Subtitle conversion supports plain cues, not every styling feature. A fixed time shift moves all cues equally and does not fix drift. Markdown conversion also supports only its stated subset. Preview subtitles in the target player and HTML in its intended context.