The calibration model is described in Section 3.3.6. This model is more complex than the original proposal to take into account effects not corrected yet in the PSF/LSF calibration Section 3.3.8. As presented in Section 5.3 of Lindegren et al. (2018), the validation of the calibration model is strongly tied to the analysis of AGIS residuals, i.e. one looks for any dependencies in the residuals that can be compensated by some effect in the calibration model. Besides the analysis of the residuals one can also plot the calibration parameters and check that they are consistent with the instrument history.

In particular the calibration model contains chromatic corrections (see Section 3.3.4) that were critical to get a good astrometric solution. In Figure 3.19 one can see the time evolution of the chromatic corrections per CCD and per field of view as computed in AGIS for window class 0 observations. On those plots one can see how the chromatic corrections evolves with time for each CCD and field of view. The impact of the decontamination and refocusing events are clearly visible as specific break points have been set for them around revolutions 1316, 1443, 2324 and 2574. From these plots one can also infer that the contamination is more sever in AF1 to AF4 compared to AF5 to AF9. Compared to the calibration model used in Gaia DR1, we used for Gaia DR2 much larger time intervals of chromatic corrections with a default step of 252 revolutions in order to reduce a degeneracy of the chromatic calibration with the reference frame Section 3.3.2.

Besides residual analysis and plotting of the fitted calibration values, one can also compare the geometric calibration model to the BAM metrology. In particular, as we have introduced break points in the calibration model corresponding to the times of some of the largest BAM jumps, one can compare the jumps derived from the calibration model with the ones introduced in the basic angle corrector. From the current state of this analysis, it seems that a relatively large fraction of the BAM jumps are not astrometric jumps as they are fully compensated by a jump of opposite amplitude in the calibration model.