Structured missingness can fail in two different ways: a model may distort the detailed profile of conditional missingness rates, or it may alter how strongly missingness depends on related variables.
To separate these effects, we keep profile-only fidelity as the canonical co-missing subitem and report relation-strength fidelity as an auxiliary diagnostic. Differences between the two indicate whether a model preserves broad dependence amplitude more easily than fine-grained missingness profiles.