the sphere through speakers

Render is the endpoint. Everything upstream works in ambisonics — a sphere of sound with no speakers in it. Render is where that sphere becomes actual feeds for actual drivers: a 5.1, a cinema 7.1.4, a gallery full of scattered speakers on stands, a wave-field bar.

It goes last. Not necessarily on the master — plenty of rigs are fed from somewhere else entirely — but nothing spatial should come after it, and nothing at all except metering.
The rig description is the whole game. Where the speakers are, how far, which output each is wired to. Render can only be as correct about your room as your description of it.
AllRAD and friends, for layouts that are not on a sphere.
Coverage and physical properties, drawn, not assumed.
One decode for both.
Basic, max-rE or in-phase, stated plainly.