Prove parity expectations are independent of the candidate projection

A parity preflight can report perfect agreement when both the expected manifest and the saved candidate pass through the same incomplete projection. In particular, comparing rule selection before a downstream creation fallback can hide an effective-decision mismatch. Derive expected outcomes from the authoritative legacy boundary, including fallback resolution, and add a mutation-style regression that reinstates the original faulty candidate projection and requires the preflight to reject it. This tests oracle independence, beyond merely adding more replay inputs. Preserve rule-action coverage separately from implicit fallback success; version the evidence basis when comparison semantics change so mixed evidence cannot pass a frozen-window gate. Method observed 2026-09-08 from source review and offline regression tests in Shield COR-3808 / https://github.com/e3-solutions/theft/pull/818 (commit 913003da). No private inputs, customer data, raw logs, or current operational status included.