From a stack of approved PDFs to a register you can defend, tag by tag.
Reads cached PDFs with Azure Document Intelligence and GPT-4o, recovers the tags the model misses, and normalises every one to a single canonical form.
Measures everything against the Master Tag Register, the yardstick. Coverage is counted, not assumed: exact match, punctuation-blind match, or flagged as possibly OCR-garbled.
A read-only toolset over the cached documents, showing where they agree, differ only in wording, or genuinely contradict one another about the same tag.
One row per register tag: values resolved from the rows and from document title blocks, a queue for genuine conflicts, and an evidence trail behind every value.
Documents fill it in. They do not define it.
Genuine disagreements stay blank and go to a human queue. It never invents a winner.
Every count states its caveats.
Each resolved value carries a trail back to the document it came from.