The market has already been warned — correctly — that fully automated AI migration produces output nobody wants to maintain. We agree. AI accelerates this work; it does not perform it. What we sell is evidence.
Static analysis parses artifacts and draws graphs. Useful — and everyone has it. What nobody else can do is execute the legacy side: run your batch on a compatible environment, produce the file, push it across the seam, and byte-compare what comes out the other end. That upgrade — from "the mapping looks right" to "we ran it, and here is the diff" — is the whole product line.
No one on our team holds a licence to any product we provide compatibility with. Our specifications come from published material; our environments are independent clean-room implementations; the demo estate is synthetic and says so. This is why we can put "no vendor licence consumed, no licensed instance touched" in a proposal — and why competitors structured as vendor partners cannot copy the claim.
A clone validating itself against itself proves nothing. Expected-output fixtures come from your staff, on your licensed systems, under a written SOW protocol — binary-safe transfer, sha256-pinned, volatile fields declared up front. We never log on. Every engagement's fixtures make the environment more faithful for the next one; by the tenth estate the conformance corpus is the moat.
Coverage percentages print on every report. Constructs we can't resolve are listed UNRESOLVED instead of guessed. Divergences without a proven cause say DIVERGENT-UNEXPLAINED. The environment is "specification-conformant," never "identical." And the capability page ships with every proposal, so the roadmap can't blur into the invoice.
Manila and Singapore, on the estates this region actually runs: the mainframe ledger, the IBM i subsidiary, the Oracle mid-office, and the seams between them. Local instructors, local time zones, and the regulator's calendar taken seriously.