Best practise question: modular assets with or without back faces?

Something I saw different approaches on over the time, now.
If you create tileable assets for levels/environments - do you keep the rear faces of walls and roofs that nobody will ever see or not?

So far I simply deleted the rear faces of walls the player/camera will never be able to view anyways. Unity’s lightmapper apparently does not seem to have a problem with correct shadows, either. So - at least so far I never saw a necessity to keep even a simplyfied rear geometry of a wall.
Then again I’ve seen a few people do this, anyways. Granted that was for UDK and I don’t know UDK’s lightmapper very good, at all. I would imagine that even simply for editing reasons - deleting any backfaces would be a good idea as it gives a better view on the “inside” of an environment.

What are you artists out there doing in this regard?
Is there a reason to keep backsides of models in Unity?

Depends on the pieces, and the platform. General rule of thumb is if it wont be seen, it should not exist, delete it, and you wont waste any uv space on it. However, it depends on the needs of the game, because if you set something up well, and then find that it may be seen from that other angle, you want something there. I do a lot of modular stuff, and my preference is optimized less faces means more pieces can be used at once.

It depends. I leave back faces in on certain objects, so that when you’re building a level you have more options. A wall section for a building can probably also be a ledge, pillar or corner piece, but if you’ve deleted the backfaces you wouldn’t have those options, and would end up having to make another almost identical object.

You may also decide that you want to add some physics to certain objects, in which case you’ll need bottoms. But if you’re 100% certain that you’ll never need the backfaces, get rid of them.

That’s what I was thinking, too.
Then again - could the extra geometry reduce light map bleeding on the edges in the final result?

I’ve created the same topic on polycount and got mixed answers. Apparently Unreal’s lightmapper does not catch shadows when there’s no back faces. At least that’s what someone posted. I haven’t worked in UDK, lately. So I can’t test such in-depth problems there, ATM.
Beäst seems to have no problem in catching shadows even if there’s no rear faces on a moudlar tile piece. Last week I did notice that the edges sometimes bleed through a little, even when increasing the edge padding by one pixel. I still need tro test this theory more thoroughly. Does anyone else have any experience with that?