RuffledBricks

That's right. So if you got stair clip into the Offy First route then the difference in time would be even more pronounced.

So I tried out @IRF's suggestion on stream this evening - I did almost two hours of playing around with different routes for the MegaTree/Off Licence section and, owing to my somewhat rusty playing, I didn't manage to get a full comparison of every single scenario I wanted. I did, however, get a good comparison of the main contention discussed in this thread: whether Off Licence should come before or after the MegaTree in an optimum OL/OOAL deathwarp route.
I timed the route via RTA on 128K Spectrum (Fuse emulator), starting from entering At The Foot Of The MegaTree and ending upon entering The Drive. Both I started with eight lives. The results were as follows:
MegaTree before Off Licence - 3:51
Off Licence before MegaTree - 3:49 (and this was WITHOUT stair clip, just FYI)
There were obviously minor variations here and there in the execution of each room, but nothing that I believe would have affected the final conclusion: Off Licence before MegaTree is, for whatever reason, faster.
 
 
 

I think if @crem's gonna spend that much time and effort on it then he should at least start a Patreon account. 😄

Anyhow, onto stuff that I know definitely will save time in the run, here's a rather pleasing find from @crem's algorithm:
 
Beach rope warp can now do a full 2 seconds worth of time saving. And despite how it looks, it's actually a pretty forgiving strat - seems to have a pretty decent range of timing and positioning that will allow you to pull it off.

I'm still not convinced. And to be honest, I don't think I will be until I see it in action. And I don't have a lot of time for recording a comparison at the moment, especially not for a section of the game that long.

I understand all that, but losing one life only speeds things up by ~0.4%, and doing the Off Licence after the MegaTree means slower timings on the Saw at the junction of Foot. I'm just not convinced that it would be enough to balance all that out.

I might have a go at trying to do a recording of that section, for comparison with your run through the equivalent section - although I'm not sure whether my personal degree of aptitude is up to it! 🤔

Anyhow, onto stuff that I know definitely will save time in the run, here's a rather pleasing find from @crem's algorithm:
 
Beach rope warp can now do a full 2 seconds worth of time saving. And despite how it looks, it's actually a pretty forgiving strat - seems to have a pretty decent range of timing and positioning that will allow you to pull it off.

God, everyone really hates that walk to the Off Licence around here, don't they?
For the route you've mentioned, I'd imagine the difference would probably be quite ineffectual given that you withhold the Off License deathwarp life until later. There might be an argument to be made for doing two deathwarps in the MegaTree and then doing the Off Licence normally, but I'd rather have someone confirm that for sure with maths and science rather than having to try and do it manually to find out.
Go on then. 😛

Well, referring back to the small discussion in the other thread, I believe *this* alone proves that crem's effort is worth it 🙂 .

Anyhow, onto stuff that I know definitely will save time in the run, here's a rather pleasing find from @crem's algorithm:
 
Beach rope warp can now do a full 2 seconds worth of time saving. And despite how it looks, it's actually a pretty forgiving strat - seems to have a pretty decent range of timing and positioning that will allow you to pull it off.

God, everyone really hates that walk to the Off Licence around here, don't they?
For the route you've mentioned, I'd imagine the difference would probably be quite ineffectual given that you withhold the Off License deathwarp life until later. There might be an argument to be made for doing two deathwarps in the MegaTree and then doing the Off Licence normally, but I'd rather have someone confirm that for sure with maths and science rather than having to try and do it manually to find out.
Go on then. 😛

Possibly. The other thing to consider is how manageable it is. Although I rather arrogantly stated that the triple jump was "easy" in the other thread, on subsequent attempts I've found the timing requires some getting used to. Plus the point I made previously about the OOAL deathwarp not being that easy to pull off quickly still stands.

At any rate, I'll only be fully convinced once I actually see the two strats for the UTM-OOAL section side by side, and my JSW-playing is a bit rusty at the moment, so that's not a video I'll be able to make in a hurry!
Doing that means you have to do a slow cycle on the Saw though. The MegaTree section has, to the best of my knowledge, been routed to minimise waiting time on all enemies within the Foot and Trunk rooms.
 
 

I think if @crem's gonna spend that much time and effort on it then he should at least start a Patreon account. 😄

I think if @crem's gonna spend that much time and effort on it then he should at least start a Patreon account. 😄

A message of assorted topics about how everything is failing but there's hope. 🙂

(1) It is clear now, that the attempted brute force solution to build the route doesn't work.
It may find good enough walkthroughs and maybe even better than the currently best known routes, but it surely won't find the optimum.
My nightly attempt to re-run the same Warpless+Maxlives lives turned out to give even longer (by 23 frames), despite having much wider search (7'571'762'474 states instead of 2'439'193'796) and entire evening to fighting algorithm's memory hunger.
The route is slightly different, here it is: https://www.twitch.tv/videos/1023369202 (I found that the easiest way to record and share the video is to stream on twitch).
I won't try any complex improvements to this algorithm. Instead I'll do one small tweak (point 3 below), run the algorithm for all the categories (4), and then probably will try a different approach (6).
 
(2) Combining the spliced walkthrough to the proper one didn't work for some of room transitions.
(technical stuff hidden in spoiler)
Anyway, that's an easy thing to fix. I split such problematic entryways into several and recalculate affected routes. This is what's happening now, should take 1 day.
 
(3) The current algorithm is too greedy, I'll try to come up with better game progress metric, which should help but not by much.
The algorithm prefers to grab lots of easy items right away, and then struggles with the remaining ones.
Currently, only "number of collected items" is used as a metric to track progress. Replacing it with a better one can improve things. I'll try to do the following:
Make items valued differently (e.g. Ballroom West items are cheapest as they are easily accessible and plentiful). Penalize some sets of remaining items (e.g. if two remaining items are in different corners of the map). That will require some manual judgment though.
 
(4) I'm still planning to generate all the speedrun categories using the current algorithm.
But first I'll have to fix issues (2), (3), and for the realtime runs, do the (5).
 
(5) For In-game-time optimization, I use number of frames. For Real-time runs, I'll need number of CPU cycles.
My plan is to measure number of CPU cycles for every of ~250000 routes that were generated (including time to draw the next room) and use it as the edge cost in the algorithm. I expect this process to take ~3 hours.
Some notes regarding that:
I'm not sure how precise is the emulator library that I use. I surely don't emulate any input or memory delays, but otherwise it should be fine. What to do with variable number of lives affecting speed? I can either test a few routes and come up with a formula how number of lives affects frame time, or I can run all routes for all number of lives. Then it will take ~24 hours, still probably that's the easiest. 🙂  
(6) Second attempt will be an AlphaZero-style neural network (if there will be a second attempt).
If the results of the current algorithm won't be satisfiable, the next thing I'll try (given enough time and enthusiasm) is to try a AlphaZero-style reinforcement learning.
For approach I'm very confident that it will work when implemented, but it's not very fast to write. Probably 1-2 months worth of winter evenings, which translates to 3-4 months of summer evenings.. But depends on weather. 🙂

Re: the original discussion, there was never any doubt that deathwarping OOAL was quicker than doing the room normally. The discussion was about whether the OOAL deathwarp actually saved more time in a full run of the game than deathwarping one of the other MegaTree rooms.
In order to do the OOAL deathwarp, one would need to do both Under The Megatree and Cuckoo's Nest 'normally' so that you can preserve your last spare life. So whilst you would be saving time on OOAL, you would be adding 9 seconds of time back on whichever of the other two rooms you normally deathwarp (let's say UTM, given that it's ever so slightly more of a timesave than CN), plus a marginally slower game speed for MegaTrunk and Tree Top, given that you are doing those rooms with one life more than you would have been.
So to prove that OOAL deathwarp is more beneficial than UTM deathwarp, you would need to factor all of that in.

Re: the original discussion, there was never any doubt that deathwarping OOAL was quicker than doing the room normally. The discussion was about whether the OOAL deathwarp actually saved more time in a full run of the game than deathwarping one of the other MegaTree rooms.
In order to do the OOAL deathwarp, one would need to do both Under The Megatree and Cuckoo's Nest 'normally' so that you can preserve your last spare life. So whilst you would be saving time on OOAL, you would be adding 9 seconds of time back on whichever of the other two rooms you normally deathwarp (let's say UTM, given that it's ever so slightly more of a timesave than CN), plus a marginally slower game speed for MegaTrunk and Tree Top, given that you are doing those rooms with one life more than you would have been.
So to prove that OOAL deathwarp is more beneficial than UTM deathwarp, you would need to factor all of that in.

Have been trying some of these out today. Already posted some of my thoughts on these in the Speedtrum Specrunning Discord, but I'll repeat them here as well.
Rather pleasingly, I've been finding several of these to be very replicable. The improved Cold Store deathwarp doesn't take too much work to get the hang of, and the triple jump in OOAL is easy once you've found the right starting position - it also reopens the discussion about whether it's worth utilising a deathwarp in that room, as the second half of the room has now been made longer as a result.
Nightmare Room is my favourite of these new ones so far, although the second jump in it is pretty difficult. It's both pixel perfect AND frame perfect as far as I can tell, meaning that the failure threshold is pretty low. However, once you've nailed that jump, the rest of the room is an absolute dream compared to the current strat, so I reckon it's well worth it.
Off Licence I'm having a lot of trouble with, as you need to pass Cyan Ball at the earliest opportunity and hit your turning jump onto the conveyor belt pretty precisely to get the Green Radar jump right. Then you also need to mind your item jumps so that you don't miss any as you head right. It's pretty unforgiving.
Oh, and I cannot get my head around the Ballroom West one at all. I just can't find any good visual hooks to reproduce it yet.
Looking forward to more of these, so cheers for your hard work @crem.

Have been trying some of these out today. Already posted some of my thoughts on these in the Speedtrum Specrunning Discord, but I'll repeat them here as well.
Rather pleasingly, I've been finding several of these to be very replicable. The improved Cold Store deathwarp doesn't take too much work to get the hang of, and the triple jump in OOAL is easy once you've found the right starting position - it also reopens the discussion about whether it's worth utilising a deathwarp in that room, as the second half of the room has now been made longer as a result.
Nightmare Room is my favourite of these new ones so far, although the second jump in it is pretty difficult. It's both pixel perfect AND frame perfect as far as I can tell, meaning that the failure threshold is pretty low. However, once you've nailed that jump, the rest of the room is an absolute dream compared to the current strat, so I reckon it's well worth it.
Off Licence I'm having a lot of trouble with, as you need to pass Cyan Ball at the earliest opportunity and hit your turning jump onto the conveyor belt pretty precisely to get the Green Radar jump right. Then you also need to mind your item jumps so that you don't miss any as you head right. It's pretty unforgiving.
Oh, and I cannot get my head around the Ballroom West one at all. I just can't find any good visual hooks to reproduce it yet.
Looking forward to more of these, so cheers for your hard work @crem.

Have been trying some of these out today. Already posted some of my thoughts on these in the Speedtrum Specrunning Discord, but I'll repeat them here as well.
Rather pleasingly, I've been finding several of these to be very replicable. The improved Cold Store deathwarp doesn't take too much work to get the hang of, and the triple jump in OOAL is easy once you've found the right starting position - it also reopens the discussion about whether it's worth utilising a deathwarp in that room, as the second half of the room has now been made longer as a result.
Nightmare Room is my favourite of these new ones so far, although the second jump in it is pretty difficult. It's both pixel perfect AND frame perfect as far as I can tell, meaning that the failure threshold is pretty low. However, once you've nailed that jump, the rest of the room is an absolute dream compared to the current strat, so I reckon it's well worth it.
Off Licence I'm having a lot of trouble with, as you need to pass Cyan Ball at the earliest opportunity and hit your turning jump onto the conveyor belt pretty precisely to get the Green Radar jump right. Then you also need to mind your item jumps so that you don't miss any as you head right. It's pretty unforgiving.
Oh, and I cannot get my head around the Ballroom West one at all. I just can't find any good visual hooks to reproduce it yet.
Looking forward to more of these, so cheers for your hard work @crem.

Are you going for the fastest 'real world' completion time, or the earliest finish according to the on-screen clock (i.e. the fewest number of frames/ticks/passes through the Main Loop)?
The merits of climbing up the Bathroom ramp at the start of the game, in order to access the Top Landing item, were discussed at an earlier stage (around the time that RuffledBricks first posted his speedruning video), and I think the consensus emerged that that early decision could have a bearing on the completion time, depending on what you're trying to achieve.
If it's the earliest finish according to the on-screen clock (fewest frames), then it stands to reason that you should collect the Top Landing item as the second item (after the Bathroom tap), so that Willy isn't wasting frames walking back into the Bathroom at the end of the game to access the ramp up to the upper levels of Top Landing.
If it's the fastest real world completion time, then RB pointed out that the game runs faster at the end (assuming you've used up all your spare lives by that point 'kamikazeing' items), because the execution of the drawing of spare lives on the status bar during each frame, slows down the game to a degree which is perceptible over the course of a whole game.  So the fact that Willy has to take a longer route (more frames) to collect the Top Landing item at the end, is outweighed by the fact that the game runs quicker at that point if you haven't got any spare lives left.
****
Incidentally, if it's the fastest real world completion that you are trying to achieve via your algorithm, then you would need to get it to release and then press 'ENTER' in the first two frames of the game, in order to turn off the in-game music.  Because the playing of the in-game tune further slows down the running of the game.

The approach I've picked supports both, but as counting frames is much easier than CPU cycles, I'll start with the in-game time.
It also supports all Any%/Warpless/Max Lives categories, so I'm planning to do all that (in-game time first, then real time later).
It's not possible to process WRITETYPER run with this approach though, but I have a different idea how to do that, so unless I lose enthusiasm by then, after normal JSW runs I'll also generate the WRITETYPE route.
The most labour-intensive part of the realtime-clock-based run generation will be to gather per frame timings:
"Base" time for one frame per room (which will be different as there will be different amount moving objects) How much extra time does every extra life add. How much time it takes to switch a room. (if there's a difference of speed depending on number of displayed collectible items [which I assume there is, but hopefully negligible], I'll have to ignore that due to the way the algorithm works).

Yes, the music on/off button can be pressed simultaneously with a movement key. But I was referring to the need to 'unpress' ENTER and then repress it if you want to turn the music off at the start of a game (there is a keypress flag that means if you keep ENTER pressed at the start of a game, then it won't toggle the music off until you let go of ENTER and then repress it).
But starting the game, switching the music off, abandoning that game and restarting another (so the music is already switched off at the start of the game) is an even simpler way of achieving the same thing.

I will leave you to get on with it then.

Addendum

Meaning :- I will not bother myself any further on your behalf.