Paper #63: Süssmayr’s Heirs: A Composition-Invariant View of Generative Stylistic Completion

Stylistic completion, extending an unfinished work in the style of its creator, is among the oldest practical problems in Western music, recurring from Süssmayr's 1791 completion of Mozart's Requiem through to the symbolic infilling models, lyrics-to-song systems, and voice cloning tools released in 2025-2026. These systems are studied as separate technical domains despite a shared task structure. We propose a unified framework in which stylistic completers, human or learned, are operators on a musical action space, and a feature is composition-invariant under a composer's stylistic prior iff it is preserved by every continuation in the prior's support. The kernel of composition-invariants identifies what makes a continuation belong to a composer's voice: the features distinguishing stylistic identity from interpretive choice. We connect the framework to existing musicological theories (Schenker, Lerdahl-Jackendoff, Lewin, Tymoczko, Kramer); propose a diagnostic battery computable on contemporary symbolic and audio outputs; apply two diagnostics in a three-way comparison on Mozart's Lacrimosa (Mozart prompt, Süssmayr, Suno V5), calibrated against within-Mozart baselines on the K.626 Kyrie; and draw implications for evaluation, licensing, and co-creative tool design at a moment when these systems face simultaneous research, commercial, and legal scrutiny.