Document-level event extraction often requires the same argument candidate to fill multiple valid event-slot targets, with each assignment grounded in different evidence. Many existing methods introduce target-specific signals during event construction, matching, or role scoring, while explicit per-target argument views jointly conditioned on event-slot semantics and instance-specific information before role scoring remain less explored. This late arrival of target-specificity is especially problematic for many-to-many (M2M) assignment, where the same candidate must preserve distinguishable evidence for several valid targets before decoding. We propose Target-conditioned Gating for Inter-target Decoupling (TGID), an encoding-stage MoE framework that produces target-conditioned argument views before decoding. TGID combines event-slot semantics with instance probes to form target queries and uses independent sigmoid gates to remove direct cross-gate coupling induced by softmax normalization, while schema-structured routing initialization and a multi-positive target-view separation objective stabilize the resulting views. By incorporating instance probes, TGID can distinguish Cross-Type, Cross-Role, and Same-Type Multi-Instance targets before decoding. Across ChFinAnn, DuEE-Fin, and FNDEE, TGID achieves consistent gains over strong trigger-free DEE baselines, especially on M2M subsets. Parameter-matched softmax controls and routing diagnostics indicate that target-sensitive routing contributes beyond additional model capacity. Code is available at https://github.com/LeonaKTH/TGID.