Measurement-Based Loss Tolerance in Graph-GKP Codes through Syndrome-Resolved Pauli-Frame Decoding

Graph codes offer multiple physical representatives of logical observables, while Gottesman-Kitaev-Preskill (GKP) codes retain analog information about bosonic displacement noise. We develop a causal framework that unifies these mechanisms for measurement-based loss tolerance under pure loss followed by quantum-limited amplification. In this framework, each local GKP recovery produces a refreshed logical block, a continuous syndrome record, and a confidence score for the inferred Pauli class. Low-confidence outcomes are deliberately converted into located erasures, so the availability pattern is generated directly from the bosonic data rather than sampled independently. Both accepted and rejected syndromes contribute to a syndrome-resolved posterior over the graph branch, which determines accessible logical representatives and the outgoing logical Pauli frame. We derive decoder-conditioned branch restriction, signed-outcome reconstruction, Pauli-frame updating, recursive concatenation of graph-GKP modules, and syndrome-resolved logical fusion. Numerical simulations across several squeezing levels identify task-dependent loss-tolerance behavior and finite-depth pseudothresholds for square and hexagonal GKP lattices. The resulting graph-GKP interface provides a unified causal control layer for fault-tolerant MBQC, fusion-based computation, and all-photonic repeaters.