Enterprise guideline documents mix narrative text, complex tables, and embedded images, and converting them into structured work artifacts typically requires two to three days of manual effort. Existing language and vision‑language models can extract information but provide no governed workflow: there is no validation, consistency checking, or traceable artifact generation. We introduce SAGE, a governed multi‑stage LLM pipeline organized around a shared versioned rule store with stable identifiers. The pipeline enforces schema‑validated inter‑stage contracts and end‑to‑end provenance tracking.
Extracted rules first undergo deterministic structural validation, followed by LLM‑based semantic scoring. A consistency module then removes duplicates, flags contradictions, and surfaces specification gaps; only uncertain or flagged items are sent to reviewers, while high‑confidence outputs are auto‑approved.
On a benchmark of 120 documents, SAGE reduces turnaround time from days to 20‑100 minutes, achieving a 96% document‑level success rate with a hallucination rate of 3.2% (without governance the hallucination rate rises to 15.7%). The system extracted 3,896 rules and produced 812 artifacts ready for human review.
Review: SAGE demonstrates that versioned rule stores, structural validation, and consistency checks can turn unstructured enterprise guidelines into review‑ready artifacts efficiently and traceably, dramatically cutting manual effort while suppressing model hallucinations.