As of November 15, 2026, the Charrette AI Room has published nine verified guides for common office workflows, distinguishing itself by refusing to rank AI models and instead focusing on reproducible, file-based tests. Written by JH-A, an architect rather than a developer, the hub targets professionals who need practical, checkable advice for tasks like cleaning messy spreadsheets or pasting AI-generated tables into Word and Google Docs. Each guide adheres to a strict editorial policy: it must name the specific file used, provide a measurable result, and explicitly state what the test did not cover.
The Spreadsheet Cleanup Test
The most detailed guide addresses a frequent pain point: cleaning vendor expense exports from other systems. The test file, a 60-row CSV with mixed date formats and currency notations, revealed a critical failure mode in AI-assisted cleanup. A currency-blind AI reported a total of 21,748,113.34, overstating the actual USD total of 51,464.59 by over 21.6 million. The guide demonstrates that separating currencies into distinct columns before conversion, using explicit date format lists, and de-duplicating on invoice numbers are essential steps. The full test file and answer key are publicly available, allowing readers to validate their own AI tools against known values.
The Paste and Print Problem
Another guide tackles the discrepancy between how AI-generated tables appear in chat interfaces and how they render in final documents. By pasting the same AI response four ways, the author showed that while Google Docs might correctly interpret a 9x4 table from the clipboard, a plain-text paste results in 50 pipe characters and no table structure. Even converting Markdown to .docx via Pandoc 3.9 can produce a table with correct cell counts but zero visible borders, a failure that only appears when printing. The guide emphasizes that 'pasting as a table' is not the same as 'printing as a table,' offering a reference-document fix to preserve table styles.
Honest Gaps and Future Work
The hub openly acknowledges its limitations, particularly regarding PDF workflows. Two planned guidesβone for converting styled PDFs to slide decks and another for extracting scanned tablesβremain unpublished because the tests could not produce definitive, reproducible numbers. This transparency contrasts with typical AI content, which often glosses over unresolved issues. The site also covers scheduling complexities, such as the New York to Seoul time zone shift after Daylight Saving Time, and offers methods for learning with AI and managing worklogs between tools like Claude Code and Codex.
Key Takeaways
- Verified guides must include a specific file, a reproducible number, and a statement of what was not tested.
- A currency-blind AI cleanup overstated a USD total by 21,696,648.75 in a controlled test.
- Pasting AI tables into documents requires checking for print-ready formatting, not just on-screen structure.
- The PDF workflow section is currently empty due to failed reproducibility tests, highlighting the hub's commitment to evidence over volume.
- All nine guides are available for free, with the spreadsheet test file publicly downloadable for user verification.
The Bottom Line
Charrette AI Room offers a refreshing antidote to AI hype by demanding proof, not just prompts. Its commitment to publishing only what can be measured and reproduced makes it a trustworthy resource for professionals tired of vague, unverified tutorials.