The reporting problem
A reporter receives three PDFs about one decision. Two are copies of the same record under different filenames. The third gives a different amount. Which passages can an editor actually cite?
What Source Ledger does
The reporter selects two to six text PDFs on a UGREEN NAS and enters one precise claim. Source Ledger is designed to return the relevant passages alongside their filenames and locations, group byte-identical copies, and show conflicting figures side by side. If a passage cannot be located or a file changes during review, the output flags the gap. The editor decides what the evidence supports.
How it is built
The pre-test package uses Markdown instructions for the UGOS file and document agents. The file agent records metadata and SHA-256 digests, the document agent retrieves passages, and the file agent rechecks the digests before an excerpt is used. Processing stays local and requires only read access to the selected files. I used OpenAI Codex and UGREEN’s UGOS Skill Creation Tool 0.7.3.
Validation and limits
UGREEN’s static checker reported zero errors and zero warnings. The package has yet to run on a UGREEN NAS. The finalist test plan covers duplicate PDFs, conflicting quantities and replacement of a file during extraction. Different file hashes alone cannot establish independent authorship. Scanned PDFs and incomplete extraction can leave a claim unresolved.
