Light is not a scalar wave. We only get away with treating it as such when the degree of polarization is very low. This condition often holds for seeing-limited single telescopes, but becomes less likely at spatial resolutions typical for interferometers. For the interferometric environment, optical polarimetry may need to assimilate radio-polarimetric concepts. In particular, the Stokes parameters should be defined in terms of complex correlations rather than as differences of orthogonally-polarized fluxes. Polarization effects in the Coudé train and delay lines spoil the accuracy of traditional quasi-scalar interferometers. An alternative optical architecture is proposed, using traditional (i.e. single-beam) optical polarimetry in the correlator, but 'radio-type' transfer of light from telescope foci to correlator (i.e. 2 clean, fully-polarized, signals from each telescope). Such a fundamental solution can eliminate errors due to inclined mirrors (phase shifts and added polarization). The architecture enables full-Stokes polarimetry at the resolution of the interferometer, but also a 'no-polarization-desired' mode which does not necessarily involve loss of signal-to-noise ratio and yet is free from polarization-induced errors of photometry. Existing polarization components permit a very wide instantaneous bandwidth (e.g. 0.3 to > 1 μm, matching CCD or STJ detectors).