Theme — simple text formats win
Plain-text source formats maximize parsing accuracy and enable precise line-by-line editing and version control.
Because language models operate directly on tokenized text sequences, plain-text formats offer the highest processing fidelity and reliability across computational pipelines:
| Preferred text format | Formats to avoid during analysis | Key analytical advantages |
|---|---|---|
| CSV / TSV | Excel (.xlsx, .xls) |
Transparent tabular structure without hidden workbook formulas, macros, or cell styles |
| Markdown | Word (.docx) |
Human-readable markup; diffable in Git and converted cleanly via Pandoc |
| LaTeX | Inspectable mathematical markup that can be revised and re-compiled programmatically | |
| HTML / SVG | PowerPoint (.pptx) |
Standalone, interactive visualizations viewable in any browser without binary slide engines |
Adopt a boundary conversion pattern: Execute data transformations and authoring in structured plain-text formats, converting to presentation deliverables (DOCX, PDF, PPTX) at the boundary of your research pipeline (Inputs, Outputs, What code makes possible).