Three copies of every doc, a draft named “final,” nothing marked current — so your AI picks one with total confidence and picks wrong. The fix is the one software solved decades ago: version your knowledge like code. Assert what's true, mark what's history, make it machine-readable.
Drop in your real files — Word, Excel, PowerPoint, PDF, Markdown — mark each one current, draft, or replaced, and download them back with that assertion stamped into each format's metadata layer: invisible in the document, readable by machines. Not a line of it on your cover page. You assert what's true; we make it machine-readable — and your files never leave your browser.
Drop a document and see what its metadata already says
Then stamp the truth in. Word · Excel · PowerPoint · PDF · Markdown — up to 5files per batch (one document's versions at a time), 25MB each
Files never leave your browser — stamping happens on your machine.
Where the metadata goes:Word, Excel, and PowerPoint get custom document properties (File → Info — invisible in the document body). PDFs get namespaced keywords in the document metadata. Markdown gets a frontmatter block — that's its native channel. Plain .txt has no invisible layer, so there the header sits visibly at the top and we say so. Any system that indexes your files — search, document management, RAG pipelines, your intelligent OS — can read all of it. A chatbot you paste a file into usually won't: that gap is exactly the difference between pasting files at an assistant and running one that knows you.
Two versions was easy. Ten files, three drafts named “final”, and no idea which is true?
That's not a formatting problem anymore — that's an audit. Upload the set and get back the map, the contradictions, and a reconciliation you can apply in one paste.
Upload 2–10 documents — versions of the same thing, or a junk drawer. The Context Auditor maps what they collectively know, shows exactly where they contradict each other (with the receipts), ranks which version is likely current with its reasoning and confidence, and proposes the reconciliation — which you apply with Context Stamp above.
The whole ladder is three verbs: stamp it(assert what's true), audit it (find where truth has drifted), architect it (keep it true automatically). And the honest ceiling: both free tools are a snapshot of a few documents. Keeping truth from drifting across a whole knowledge base — and enforcing it automatically, every day — is a standing system, not a one-time cleanup. When you hit that wall, that's the work we do. Meanwhile, the blog series walks the whole method.