Excel's gene-symbol problem and other auto-coercions
Lesson 5 of 6 · Spreadsheets and instrument exports
1 · Learn the move · Examples + few-shot
Open a gene list in Excel and MARCH1 becomes 1-Mar, SEPT7 becomes 7-Sep, and DEC1 becomes 1-Dec - silently, on open, before you touch anything. The damage is old and documented - it forced actual gene renaming - and it generalizes: identifiers with leading zeros lose them, long numeric IDs collapse to scientific notation, and anything date-shaped gets eaten. Detection is a pattern-recognition job, which makes it a few-shot job: show the model example pairs of original and corrupted values, and it generalizes the scan across your columns far better than from a description alone. One hard rule governs the fix: corruption is often not reversible from the file - 1-Mar could have been MARCH1 or MARC1, and 2.30E+11 has lost real digits. The model flags; the repair comes from the upstream source, re-exported with the column typed as text. Reconstruction by plausibility is just fabrication with a spreadsheet accent.
Scan these columns for spreadsheet auto-coercion damage. Examples of the corruption patterns: MARCH1 -> 1-Mar, SEPT7 -> 7-Sep, DEC1 -> 1-Dec (gene symbols to dates); 007234 -> 7234 (leading zeros stripped); 230000001234 -> 2.30E+11 (long IDs to scientific notation). Generalize from these examples: flag every value matching a corruption pattern, with row numbers and the pattern matched. Do NOT propose reconstructed originals - corruption is lossy; the fix is re-export from the source with text-typed columns. Columns: [paste]
2 · Your turn. You write the prompt
A collaborator's differential expression results arrive as an .xlsx, and scrolling the gene column you pass 2-Sep sitting where a gene symbol belongs. It has been through Excel at least once, and you do not know what else got eaten. Write the prompt that scans every column for the whole corruption family, using examples to teach the pattern.
Remember: the AI sees only your prompt, not this page. If the situation isn't in your prompt, it doesn't exist.
Optional. These shape the output when you run your prompt below, not your score.