This paper addresses sequential synchronization for first-order affine systems in the presence of bounded disturbances. We develop a reinforcement learning (RL) based controller that combines an anisotropic weighted power-sign baseline with a projected residual actor-critic. The baseline uses a diagonal weighting to shape the direction field of a transformed error and thereby encode the desired convergence order across state channels. A lightweight projection comprising a norm cap and a half-space alignment with the descent direction filters the residual policy so as to preserve the baseline Lyapunov decrease. We establish fixed-time stability and sequential convergence for the baseline via a composite Lyapunov argument, and prove that the RL residual maintains these properties under the projection. A spacecraft translational tracking case study with bounded sinusoidal disturbances validates the design: the position states converge in the prescribed order, the residual actions remain consistently bounded with respect to the baseline control magnitude and direction, and learning signals stabilize. Quantitatively, axis-wise first-hit times, control energy, and peak control corroborate the theoretical guarantees while illustrating performance benefits of the residual layer.