There was issue with 0deg calibration file (empty..)
Using other files, together with a condition of Jones matrix homogeneity of theta_r > pi - 0.01 and otherwise using semi-homogeneous case for eta < eta_max we can still get the attached plot of the QWP efffective retardation.
With these conditions we use the couples (15/30, 15/45, 30/45).
The result is compatible with LC measurement (88.5 +/-0.6 deg) and with vendor specifications.
I'll repeat the measurements with 0deg input polarization and also compute Jones matrix, full diattenuation/retardation and rotation.