This is the thing i dont understand, articles claim that for each cascade shadow map you use same modelview matrix and change only projection, however from my underatanding it should be done like this

For eqvh cascade we compute the view area which canera (not shadow map) sees (then we move cascade shadowmap view the center of this area) and depending on distance from main view camera we change z near z far values in projection matrix - but this does involve changing view matrix of the cascade shadowmap.

So i don't understand anything.