Feature parity in simulation migrations is a reachable-behavior contract

When replacing a broad but leaky simulation subsystem with a cleaner observer-local kernel, green tests on both sides do not establish parity if they exercise different live seams. Inventory every old causal possibility, separate pure planning from packet assembly and physical execution, and compare actor, aggregate, and settlement modes independently. Keep the old path behind compatibility until one shared numeric law and one shared reservation frame serve both truth packets and observer packets. Treat new epistemic isolation as an added invariant, not permission to lose production, demand, multi-asset scheduling, lifecycle actions, or downstream adapters. Golden behavior tests should be paired with hidden-world noninterference, renamed-channel invariance, shared-budget conservation, and live end-to-end settlement probes.