Fastest theoretical run

[RamboBones]

I was recently re-reading this thread on Doomworld and had a question. What is currently considered the fastest possible (TAS, built or even (unlikely) human) run of Doom 2 possible?

In the aformentioned thread Andrey says he belives that sub 14 minutes would be possible, although he doesn't wish to do another run to do so.

Currently the fastest demos are TAS demos but if someone were crazy enough to spend the time doing a built run of doom 2 would that be significantly faster?

I'm asking more out of curiosity than any real desire to try and achieve the best time.

[Grazza]

I think there isn't so much to add to what has been said in that thread and others on the topic (notably this one). Note that AdamH is the authority on this stuff, so for details, see his posts. I don't recall any drastic new tricks being discovered in the time since then.

There are really several separate questions here, as these are all quite distinct from one another:
best realistically feasible non-TAS time
best possible traditional TAS time (only manual strafe-50)
best possible super-TAS time (permanent strafe-50)
best possible time for a built demo (constructed 100% via LMPC)

best realistically feasible non-TAS time
Current record is 26:09 (by Radek). It's an impressive run, but presumably a serious effort by one of the best speedrunners could bring the time down. By how much, I don't want to speculate. (Vile and Henning brought the 30nm time down by more than 4 minutes last year. Hard to imagine there is that much scope for improvement in the 30uv, but then again a year ago, a 30nm below 30 minutes would have seemed unbelievable.)

best possible traditional TAS time (only manual strafe-50)
Current record is Sedlo's old TASMBF 16:17. AdamH's estimate was that with current routes (and without MBF's bugfixes that prevent some keygrabs) this could come down to around 15 minutes. Then you can factor in that Sedlo is a bit faster now than when he recorded the 16:17.

best possible super-TAS time (permanent strafe-50)
Current record is Andrey Budko's 14:41. He seemed to be estimating that around 14:10 should be possible (skip map32, and shave the odd second off here and there).

best possible time for a built demo (constructed 100% via LMPC)
Cameron estimated less than 13 minutes ("sub-13 do-able possibly 12mins is optimal time"), but as you can see in the previous discussions, there is skepticism about the feasibility of that. He has yet to demonstrate the new tricks that would be needed.

[xit-vono]

Well yeah there are quite a few things to consider. First of all built demos can be improved just like a TAS demo, so you would have the fastest feasable built demo, as well as the mathematically optimal time. Well I don't know what the limits would be for those, but I would guess around 12 minutes or so for the absolute optimal time.