Fixed!!
TL;DR: Matt side up to pentaprism, shiny fresnel down to mirror, and even more importantly the rounded corners must be on the lens side and the angular corners on the film side. Because of the second problem (rotation) the lens focus scale was all wrong (2m for a 5m subject), and correcting the up/down makes all parts of the focusing screen be in agreement.
First off, it's impossible to put the focus screen in the wrong *location* but it may be possible to insert it upside-down. You might think (As many here said) that the matt side goes down to mirror (Possibly exposing the side that is easier to clean?) however when I had the lens inserted that way the focusing was unsatisfactory. Specifically, when the split focusing screen was aligned the image in the metering spot and outside looked not quite focused, so do you focus on the split line, or the metering spot, or the textured area, or the rest of the frame in the matt area?? Well, obviously they should all agree so I decided that matt down to mirror is wrong for this camera.
So I decided to go with the service manual, which is not surprisingly correct if badly worded ("Position the screen with its matted face coming on pentaprism side"). However I also noticed that there is a reference to "R Corner" and "B side mount", presumably meaning that the rounded corners go toward the bayonet mount.
So I put the rounded corners to the bayonet mount by rotating 180 degrees relative to where I was before, and put matt up to pentaprism. Now some exciting things happen:
1) When I focus on a subject 5 metres away the lens focus scale reads exactly 5m quite accurately.
2) The lens goes very slightly beyond infinity, which is normal, but when I focus on a lamppost at infinity the focus point is on the left edge of the infinity symbol, which is satisfactory.
3) At infinity focus, everything at infinity looks in focus a) in the split focus spot b) in the metering spot c) in the textured area around the metering spot d) in the matt area of the rest of the viewfinder.
I am so confident of the positioning that I have chosen not to waste the remainder of the test roll, I will use it tomorrow on a trip to the country.
So the correct positioning for the OM10 focusing screen is: Matt side to pentaprism, Shiny fresnel side to mirror, rounded corners to lens, angular corners to film.
The rotation of the focusing screen 180 degrees is what caused the lens to read 2 metres instead of 5 and 10 metres instead of infinity etc, not its inversion mirror to pentaprism. Clearly the fresnel lens is not symmetrical, I should not have assumed it to be so.
Thanks all for your help. It's astonishing how much disagreement exists on this, it may be different for different cameras, but I am 100% confident that it is now correct for the OM10 due in no small part to interpreting the slightly cryptic service manual, and will hopefully be 110% certain when I develop the test roll which will be all out of focus to date.