<p>Metaphotonics uses nanoengineered materials to manipulate the electromagnetic fields and is of use in multidimensional optoelectronic applications such as Stokes detection. Machine learning algorithms are often used in the device design and post-signal processing of these systems. During post-signal processing, such algorithms can be used to reconstruct the physical quantities from multiparameter optical responses, typically via inversion of the system’s response function. However, when response channels are coupled or signal amplitudes vary greatly, post-detection decoding becomes difficult due to potential information loss. Here we report a metaphotonic photodetector capable of direct Stokes quantification. We use channel-level decoupling to design independent photovoltage channels for each Stokes parameter with minimal crosstalk. The device responsivity matrix achieves a near-unity condition number, reducing reliance on complex algorithmic post-processing. Our approach illustrates how device-level optimization can enhance detection capabilities in parallel with algorithmic techniques.</p>