Excel conversion script check

Method and scope

On October 18, 2026 (America/Los_Angeles), we ran the pinned standalone Python converter on one synthetic XLSX with two sheets. This checks that file and dependency combination, rather than an Agent session or arbitrary workbook accuracy.

Windows; Python 3.12.14; MarkItDown 0.1.8; pandas 3.0.1; openpyxl 3.1.5. Pinned upstream script at d7e4ad98. The verifier checks its SHA-256 before execution.

Download the synthetic input · Raw Markdown output · Checksums and machine-readable results

What the input contained

The Work sheet has five columns. Design uses the formula =B2*C2 with inputs 5 and 25 and a saved value of 125. Review has a saved formula result of 40; the total is 165. Other cells contain a real zero, missing inputs, the note “Concept | prototype,” and a multiline note. The Stock sheet distinguishes a zero count from a missing count and includes Chinese text. The fixture was authored with Artifact Tool and its two sheet previews were visually checked.

Before conversion, openpyxl confirmed both the formula text and saved values in the exported XLSX. The verifier then copied the input into a fresh temporary directory and ran the original converter without changing it.

Observed results

  • Exit code 0; input XLSX hash unchanged; both sheet headings and Chinese text retained.
  • Saved formula values appeared as 125.0 and 165.0. The =B2*C2 expression was absent. This does not establish formula recalculation.
  • The real zero became 0.0; the blank stock count became NaN. The 25.00 number format became 25.0.
  • The multiline note contained a literal backslash-n sequence.
  • The pipe in “Concept | prototype” was unescaped. Its Markdown row contained six cells beneath a five-column header.
  • No image folder was generated for this image-free input.

The extra column matters when an agent extracts values by column position. Check the raw table against the workbook before analysis. Keep the workbook as the source for formulas and current calculated values.

Reproduce the check

Use a separate Python environment. Install markitdown[xlsx]==0.1.8, pandas==3.0.1, and openpyxl==3.1.5. Download the pinned converter, verification script, and fixture into a working folder.

Run python verify.py --converter convert_excel_to_md.py --fixture conversion-fixture.xlsx --output-dir reproduced. It checks the fixture’s stored formulas and values, executes the converter on a fresh copy, and writes converted.md plus results.json. It performs no network requests.

Agent loading and installation, Excel/LibreOffice recalculation, formula caches missing or out of date, native charts, embedded pictures, hidden sheets, protected files, recursive folders, batches, and security were not tested. Other dependency versions may produce different output. This check provides no evidence of search traffic gains.