my 2 cents: the engine could use some defaults for lightmapping. like 5 cm per luxel for average gi smear. the world space bounding box scale could make automatic adjustments to the lightmap scale of any prop or blockout mesh. that’s unless you need more shadow detail and tweaks. also… light (GI) and shadow could be decoupled and deliver different light map and shadow map resolutions. or hybrid light and shadow code. baked lighting with realtime shadows. you know?1?
in terms of performance for archviz: it will almost always be mostly static lighting. there’s no point computing dynamic lighting. just bake it offline. it will run on a toaster to present it to a customer.
if you modify it on the fly you gonna need a decent laptop to run lumen, but simple lighting is very much achievable without using path tracing.
it depends, really. if you render a previz for a customer? you bake or render a pathtraced video. if you’re on site with them, deal with the computation. archviz be archviz. hmm