Semantic Modernization of VistA: A Deterministic, Audited, Multi‑Phase Pipeline with FileMan Metadata Integration and MCP Deployment

Semantic Modernization of VistA presents a deterministic, audited, multi‑phase semantic front end for the Veterans Health Information Systems and Technology Architecture (VistA). The system combines a clean‑room tree‑sitter grammar, a parser‑independent intermediate representation, a corpus‑wide dependency graph, a FileMan metadata layer, and a standalone Model Context Protocol (MCP) server. The pipeline parses the full WorldVistA/VistA‑M corpus (33,951 routines, 3.64M lines), resolves 659,973 static call sites, classifies 4,353 dynamic calls without fabricating targets, and explains 7,665 of 8,066 XECUTE‑of‑global sites using typed metadata. The metadata layer catalogs 150,828 stored‑code nodes with complete two‑parser agreement on all nodes that parse cleanly. The final graph contains 661,507 nodes and 1,673,086 edges, each carrying one of five epistemic derivations (parser fact, metadata join, static inference, runtime‑required, or reserved LLM inference). Reproducibility is guaranteed through closed JSON Schemas, deterministic rebuilds, and independent verification. Two full runs produce byte‑identical metadata and graph streams, and a complete rebuild on a second machine reproduces every stream hash. A blind audit performed 172 sampled checks, recomputed 640 RPC/option findings, and exhaustively re‑evaluated 6,959 literal‑reference XECUTE patterns, identifying four systematic issues — all corrected before the reference release. The release includes the full dependency graph, metadata, MCP corpora, schemas, run records, evidence directory, and a one‑command reproduction script. This is a deterministic, audited, reproducible research artifact intended for modernization research, analysis, and tooling development. It is not an operational platform.