MUMPS and M systems run healthcare and finance, they are vast, and they are famously undocumented, which is exactly where a structural map earns its keep before anyone touches the logic. An AI rewrite that builds and runs can still violate a convention the system holds everywhere and that no living maintainer ever wrote down.
Step one · the map
doloop reads your M routine call structure, the DO calls, the extrinsic functions, the routine references, and makes the system navigable. You get the load-bearing routines a change ripples out from, the globals and the data layer, a triage list of routines the static call graph never reaches, and the convention strata that show which parts were written when. On a vast undocumented M system, that map is the first thing your team needs. It shows where the system is structured, not what it computes. The scan is free, and it reads none of your business logic.
M uses dynamic dispatch and late binding, so a routine the static call graph does not reach is a lead to review, not a verdict that the code is dead.
Step two · the gate
Once the system is navigable, doloop can gate the changes. Where your code keeps a convention, it infers that convention from your own code and flags an AI rewrite that breaks it, deterministically, with the rule, the line, and the rate behind every flag.
M code is terse and dense, so the first scan measures, on your own code, exactly where your conventions run tight enough for the gate to enforce. The map pays off on every M system. The gate runs where your code holds a convention, and the scan shows you precisely where that is.
The map is the strong fit. Large undocumented M systems are exactly where structural navigation pays off first, and the first scan reads your M system and hands you the map.
The method reads a system's own conventions in any language. It works on code as different as modern Python and forty-year-old COBOL, where 525 of 525 held-out programs passed leave-one-out, verified three independent ways. Your M system is next. Your source never leaves your machine. Bring your own model. Request your first scan →
Routes you to a real doloop page, asks when your question is ambiguous, or tells you when there is no answer. No model runs on the answer path, so it cannot invent one.