Office Metadata Regression Pack
Last updated: 8 October 2026
Word, Excel and PowerPoint files with hidden data planted in known places, and a catalogue of what a cleaning tool has to take out and what it has to leave alone. It is the corpus ScrubDrop's own engine is tested against on every change: 40 synthetic files (.docx, .xlsx and .pptx), each with its expected result.
Who it is for
People who build or test tools that clean documents before they are shared: metadata removers, data-loss prevention, e-discovery and redaction pipelines, and the QA teams that have to sign them off. If your tool claims to remove comments or tracked changes, the pack shows whether it also catches the author in Word's people list, the text a tracked deletion still keeps, or a change made inside a footnote.
What is planted
- Document properties: author, last editor, company, manager, custom properties, the saved-in path, the user a workbook is shared by
- Comments and their authors, including Word's people list
- Tracked changes, including deleted text, and editing-session ids (rsid)
- Hidden sheets, rows, columns, slides, shapes and text
- Pivot tables and their data caches, defined names, links to other files
- Speaker notes and objects outside the slide
- Embedded objects with their preview pictures, and pictures linked to other files
- Custom XML, document variables, smart tags and presentation tags
- Image metadata: Exif in JPEG pictures, text chunks in PNG
- Parts Office does not need: unrecognised extensions, PowerPoint's editing record, add-ins
- To be pointed out but not removed: white or tiny text, cells outside the print area
What the catalogue says
For each file, manifest.json lists the kinds of hidden data planted in it, the strings that must not remain anywhere in the package (every part, decompressed) once those kinds are removed, the visible strings that must remain, and the kinds a tool should only point out. Tracked changes are expected to be accepted: inserted text stays and deleted text goes. The same expectations are asserted on every file by ScrubDrop's engine regression; the catalogue is generated from that test's own manifest, so the two cannot drift apart. The README explains how to run a tool against it.
Samples
Three files from the pack, with their entries in the catalogue. They are the same made-up files the scrub page offers to try.
- demo-letter.docx: a letter with a reviewer's comment, a deleted salary figure, a hidden note and editing-session ids
- demo-budget.xlsx: a budget whose hidden sheet holds staff salaries, with a comment and a hidden row
- demo-deck.pptx: a pitch deck with speaker notes, a hidden slide of internal pricing and a comment
Synthetic only
Every file is written by a script. No real documents were used and none are included; the names, companies and figures are placeholders made up for the purpose. The licence is provisional: for evaluating and testing software, not for redistribution.
The full pack
The full pack is sent on request. The button opens an email to us in your own email program; this page collects nothing, and no price is set yet.
If Google Analytics is switched on, pressing the button is counted as one event that says only which pack and which page (Privacy). It is not switched on yet.