Why adding dielectric plates above filters and gaps decrease losses?

For example, some 0.5mm ceramic piece can give nice signal improvement if glued above DC decoupling gaps. I have doubts about how it works. Because sometimes covering part of gap gives better result, so maybe it alters impedancea a bit and works as matching? For example, if we currently somewhere on a Smith chart, where small changes in line lengths gives big impedance change. Or dielectric plate reflects radiated signal back to microstrips and nothing to do with impedance matching?

It's just for tuning, to add some capacitance (electric field in high epsilon instead of air).

if you take a fully tuned filter, measure its passband insersion loss,....then add these dielectric plates on one side AND retune the filter.....the filter SHOULD be lossier with the dielectric plates. the reason is that some of the electic field will now be in the dielectric, and the dielecric has a loss (Tan del), so if it is near a resonator, the resonator Q will drop.

If you are talking about only tiny dielectric chips, then as said, they are probably minor tuning elements, and their added loss is swamped by the improved return loss they provide.