This paper proposes a trust-adaptive nonlinear observer architecture for oscillatory systems subject to drifting and intermittently corrupted diagnostics. The framework jointly estimates phase, frequency drift, and frequency slew while performing residual-driven trust-weighted fusion across measurement channels. A boundedness result is established under local observability and bounded disturbance assumptions. A comparison theorem demonstrates structural model mismatch in frequency-only observers under nonzero slew. Analytical stress tests quantify bias-induced error growth and formalize impulse suppression under residual-driven trust adaptation. Magnetically confined plasma mode estimation is presented as a motivating application. No claim is made regarding stochastic optimality; the contribution is structural robustness under bounded disturbances.