What can go wrong
A custom evaluator can accumulate code, cases, and process without changing the actual decision. If a simpler tool produces the same outcome and preserves the same evidence, the custom layer is overhead—not a research contribution.
What we know
The frozen VCE regression cases remain the reference. Their expected accept and reject decisions have been reproduced inside established evaluation infrastructure. That proves the surrounding framework can be reused. It does not prove that VCE itself adds value.
What we are testing now
Run the same cases, scoring rules, evidence requirements, and modest adversarial extensions through both implementations. Any claimed difference must be reproducible and important enough to change a decision—not merely a difference in presentation or code structure.
Decision rule
If the simpler implementation produces the same decisions and evidence, the bespoke runner will be retired. The frozen contract, oracle, and regression cases will remain. If VCE produces a meaningful difference, only the smallest component responsible for that difference will be kept.
What this does not prove
VCE has not established comparative superiority, production readiness, or a general safety result. Correct decisions on frozen synthetic cases validate those cases and rules—not every real-world continuity or governance problem.