Real Patterns Need Closure: Transition Autonomy as a Dynamical Criterion for Macro-Objecthood

Dennett’s real-pattern realism links ontology to compression and prediction, but compression alone is too permissive: contrived codings and dynamically idle aggregates can satisfy it. This paper argues that the missing ingredient is an explicit closure condition. A candidate macro-object qualifies when, for a fixed regime, horizon, and admissible intervention class, macrostate information is sufficient for macro-transitions, so within-class micro-differences do not change macro-level what-follows. In exact Markov settings, strong lumpability provides a benchmark realization of this condition. In non-ideal settings, closure is graded through leakiness and convergent diagnostics under fixed constraints. The result is realist but disciplined: it excludes high-leak composites, supports graded verdicts in borderline cases, and remains compatible with microphysical completeness. By tying commitment strength to robustness under anti-gerrymandering tests, the framework explains why selective realism is principled rather than discretionary. The paper contributes a philosophical criterion and an audit protocol for applying it, while leaving domain-specific estimator engineering to downstream methodological work