| Pending item | Stuck on | How to close |
|---|---|---|
| Actual docx differences across renderers | No LibreOffice here | Install one, export both to PDF, compare page by page |
| Read time for very large CSV (million rows) | No sample at that scale | Generate or source one, record time and memory |
| Behaviour on encrypted PDFs | No encrypted sample | Password-protect one and see whether it errors or returns empty |
| Character mapping in older Chinese PDFs | Common in 2000s scan-plus-OCR files | Find one and check whether extraction is garbled |
| Real Windows long-path ceiling | Measured only to 87 characters here | Build a path near 260 characters and test the boundary |
Why some entries have a method and no conclusion
Because conclusions must come from measurement. Every 'to be tested' item here is something we know how to do and have not done — writing the method respects the reader; inventing the result does not.
The fastest way to judge whether a claim is trustworthy is to look for a checked date and a status marker. Dates, statuses, and failure lists usually mean a real run; prose that flows with no boundaries usually means second-hand.