Sourcetrace is the document-structure engine inside every Lumen & Lever build. It reads contracts, leases and project records as structure, not as flattened text, and preserves the layout, tables, annotations and page evidence before anything reaches a model. It is not sold on its own. It is how the systems are built.
Local-first. Runs inside your environment or private Australian hosting.
A document is pulled in, turned into a run of text, chunked, embedded, and treated as if its structure did not carry meaning. Often the structure was the meaning. Sourcetrace keeps it.
Columns, headers, footnotes and the order a person would actually read them, kept rather than collapsed into one stream of text.
Rows keep their relationship to columns. A figure stays attached to the period and the line item it belongs to, instead of becoming a loose number in a paragraph.
Comments, mark-ups and notes stay attached to the clause or cell they refer to, not detached into a separate list with no anchor.
Every extracted element carries a reference back to the document and page it came from. That reference is what makes traceability possible downstream.
Where a read is uncertain, Sourcetrace exposes the uncertainty and the diagnostics behind it, rather than presenting an inferred result as a fact.
If the page reference is thrown away when the document is read, no amount of work later puts it back. The system can cite a source only if the source was preserved on the way in.
That is why Sourcetrace sits at the front of every build. The five commitments on the engineering standard, traceable above all, depend on what the engine kept.
Sourcetrace runs local-first, inside your environment or private Australian hosting. Your documents are not sent to a third-party service to be read.
Contracts, leases, correspondence, project records, PDF and RTF bundles.
Structure preserved: layout, tables, annotations, page evidence, diagnostics.
A traceable document the system can reason over and still cite by page.
The structure-preserving approach behind Sourcetrace started as rtfstruct, an open-source parser that reads Rich Text Format as a structured document tree rather than as plain text. It is public, so the engineering thinking is open to inspection.
rtfstruct is published under Apache-2.0. See the source at github.com/keny369/rtfstruct, and the background in why we built rtfstruct.
Thirty minutes. You describe the documents and where you suspect leakage. We tell you which lane fits and whether a Proof Week is worth your money. Sometimes it is not, and we say so.
We sign your NDA, take a sample of documents, and build. Five days later you watch the system run on your own record.
Production build, or a findings register and a handshake. No retainer, no lock-in, no deck.
Or write direct: hello@lumenandlever.com · Melbourne, Australia