(Link to video)
The Ultimate Doom - Episode 1 UV-Speed in 3:23.
The first 80 tics (2.31) seconds of this TAS were done by 38_ViTa_38, with a slight angle change not affecting overall movement and the addition of firing to manipulate subsequent RNG changed by almostmatt1. The rest was entirely done by almostmatt1. This TAS was completed on the 7th December 2020. These comments were written by almostmatt1.

Game objectives

  • Tools used: XDRE 2.22.
  • Complete the game as fast as possible on the Ultra-Violence difficulty setting. Kills, items and secrets are not required.
  • Primarily aims for in-game time rather than RTA.
  • Manipulates RNG.
  • Takes damage to save time.
  • Abuses programming errors.

Recorded with:

This demo was built frame-by-frame in XDRE 2.20 and XDRE 2.22, which allow for the frame-by-frame building of demos. The program has the useful functions of also allowing you to track your current exact position, momentum, speed travelled per tic (and in which direction), position of enemies and how they move, proximity to switches/doors/linedefs, and damage given on each frame, all necessary information for multiple tricks and optimisations. Demo playback was recorded using the vanilla-compatible source port Prboom-Plus 2.5.1.4.

Playback

This demo will play back on a vanilla-compatible source port. I recommend Crispy Doom or PrBoom-Plus, the most recent versions that can be found on dsdarchive.com. Use the command line: "yourexe.exe -iwad doom.wad -playdemo ep1x323.lmp". It should also play back on the original Ultimate Doom v1.9. Use the command line: "doom.exe -playdemo ep1x323.lmp". For general convenience and quality I recommend a source port. Note that the popular source ports ZDoom and GZDoom are not vanilla-compatible and will not work.

Comments

This demo is a 22 second IGT improvement to the previously completed Episode 1 UV-Speed TAS in 3:45 by Andrey Budko from 2006, a demo that isn't available on tasvideos.org but which can be found on dsdarchive.com and YouTube. In addition to a 22 second faster overall time, each individual level is faster by at least a second. If anyone is curious, it's also 2:08 faster than the current fastest non-TAS record.
RTA timing is traditionally not considered in the Doom community but for this submission I also completed the level intermission screens as quick as possible for a better RTA time. From the beginning of the demo to the final input to display the end time, this demo is 1,372 tics faster than Andrey Budkos 3:45 - though again, it must be emphasised that IGT is the traditional goal for Doom and was Andreys goal and intent, so tics saved isn't as important a metric as in-game time. Speaking of in-game time - seconds are rounded down, and those whole seconds are added up for the overall time. So, completing two levels in 10.9 seconds and 10.9 seconds will yield a time of 20 seconds, not 21.8.
UV-Speed is a category that aims to complete the game as fast as possible on the Ultra-Violence difficulty setting, akin to the more traditionally used "any%" in other games. Ultra-Violence isn't the highest difficulty setting (Nightmare! being the highest), but Ultra-Violence is by far the more active and historically relevant difficulty setting, and it was my intention to improve the existing UV-Speed record.

Tricks and glitches used

  1. Wallrunning and thingrunning - Running against walls, enemies, and objects like barrels and lamps temporarily increases speed (technically increases distance travelled without increasing momentum). Only available on certain walls and in certain directions. Done often and in all levels, often only for a few frames at a time.
  2. RNG manipulation - My shooting (sometimes at no target) and movement is sometimes done for the purpose of altering several things that are RNG dependant, including monster movement, damage dealt by monsters, damage dealt by the player, and the spread of the shotgun projectiles. Done often and in all levels.
  3. Momentum preservation - Under precise circumstances the player can maintain their speed while a door opens, allowing them to travel through it at a significantly higher speed. Done in all levels except e1m2 & e1m8.
  4. "0.000000 trick" - Similar to momentum preservation but functionally different, this allows you to maintain momentum (but at a lower speed) in multiple directions and is primarily useful for getting around corners quicker immediately after doors open. Done on e1m1, e1m2, e1m3, e1m5, e1m6 and e1m7. This trick is very precise and requires brute forcing tools to do it without losing speed.
  5. Void Glide - Under very special circumstances the player can enter the out-of-bounds area (referred to as "the void" in the Doom community). This is rarely possible and useful, but is used in e1m8.
  6. Elastic collision - redirects momentum from one direction to another. Only done once at the start of e1m1.
  7. Deliberate self-damage - a quick e1m8 time relies on low health.

Stage by stage comments

e1m1

The movement of the first 2.31 seconds was taken from an e1m1 No Monsters individual level demo by 38_ViTa_38. The movement in their portion of the demo consists of a brief barrel thingrun, then a north wallrun, then a diagonal north wallrun (different and much harder to do on a diagonal wall), then an elastic collision to redirect momentum, then a northeast diagonal wallrun, then the 0.000000 trick, and then a difficult positional manipulation to exceed the regular player momentum limit of 14.5 units per second while performing this trick - all in 80 frames. My (almostmatt1) best attempt at this section was 1 frame slower. After this there is a wallrun, momentum preservation trick and 0.000000 trick. RNG is manipulated to kill two enemies in the least number of shots, have enemies move out of my way, receive only one hit, and receive the minimum possible damage from that hit. There are faster individual level TASes of e1m1. One is of the No Monsters category and so is easier to navigate, and the fastest with monsters heavily relies on monsters damaging the player to save 4 frames. As IGT was my goal, I opted to lose the 4 frames and conserve health while still getting 7 seconds.

e1m2

Pretty straightforward. Some RNG manipulation with firing to receive less damage (again with IGT and not RTA time in mind).

e1m3

Some RNG manipulation was difficult, there are some tight corridors densely packed with enemies which are hard to optimally move through.

e1m4

Very straightforward.

e1m5

This was the most difficult and frustrating level to build as manipulating and navigating enemy movement, especially in the trivial-looking final room with the 3 pinkies, was very hard to do. I exit this level 1 frame later than is possible as this gave me more favourable RNG for a trick in the following level, where an enemy moves into a certain position and keeps a door open (saves 30 seconds).

e1m6

This level takes a different route than Andrey Budko did, since going out of my way to gather additional rockets for the (now skipped) e1m8 fight is no longer needed.

e1m7

This level has some of the ugliest movement at the end as I constantly adjust my angle between chaingun shots, so it's quite jagged to watch. An attempt to do this with less turning would have resulted in a time a few frames slower.

e1m8

This map is the most interesting as it contains a fairly rare trick, the aforementioned "void glide". Due to reasons far too technical to explain here, under certain conditions the game will increase your momentum to such an extent that you exit the map, if you're close enough to the edge of it. Regularly in e1m8, the player is teleported to a room in which they take constant damage and the level finishes when their health is low enough. The special effect of this room is extended into a portion of the out-of-bounds area and so when the player reaches it they take damage and exit the level, just as if they had completed the level the intended way. The TAS was routed to ensure I had 1 rocket with which to damage myself so that my health was low enough to complete the map on the first frame that I reach this special area.

Other comments

I have left 5 seconds of blank frames in the encoded demo above to allow plenty of time to show to final time. It also shows the "wipe screen effect" between levels, which is possible to disable in the source port I used to record the video, but I decided to leave it on to give more of a vanilla-true experience.
However, having read "...DOOM internal relay files (LMP) are no exception, we only accept those without blank frames at the end.", the file submitted, ep1x323.lmp, has only one frame of the final time screen. The encoded video contains absolutely no other alterations, it simply contains an additional 5 seconds of blank frames at the end. If anyone needs a demo with an extended end screen for encoding, let me know and I can easily sort that out for you. I hope I've gotten everything right with the submission process, if I have made any errors please let me know and I will fix them as soon as possible. Also, note that e1m9 (the secret level) was skipped and is not required for UV-Speed.
EDIT 11/12/2020: I'm adding the level stats for this TAS and the old TAS that this one improves, since the intermission screens go by a too fast to make out this information. Sorry, because of formatting it's not very easy to read, I'm working on a way to do something about that but thought I'd update the information anyway until I do.
3:45 TAS by Andrey Budko:

E1M1 - 0:08.97 (0:08) K: 1/29 I: 1/38 S: 0/3

E1M2 - 0:19.54 (0:27) K: 11/79 I: 1/42 S: 0/6

E1M3 - 0:31.69 (0:58) K: 39/131 I: 8/95 S: 0/7

E1M4 - 0:11.77 (1:09) K: 1/85 I: 3/45 S: 0/3

E1M5 - 0:36.89 (1:45) K: 35/131 I: 4/29 S: 0/9

E1M6 - 0:41.80 (2:26) K: 28/177 I: 14/113 S: 2/4

E1M7 - 0:57.97 (3:23) K: 59/150 I: 26/84 S: 1/4

E1M8 - 0:22.86 (3:45) K: 5/41 I: 0/3 S: 0/1

3:23 TAS, this one ("TS:" means time saved, first number is IGT, bracket is RTA):

E1M1 - 0:07.97 (0:07) K: 3/29 I: 2/38 S: 0/3 TS: 1 (1.00)

E1M2 - 0:18.54 (0:25) K: 8/79 I: 0/42 S: 0/6 TS: 1 (1.00)

E1M3 - 0:30.69 (0:55) K: 39/131 I: 8/95 S: 0/7 TS: 1 (1.00)

E1M4 - 0:10.57 (1:05) K: 2/85 I: 3/45 S: 0/3 TS: 1 (1.20)

E1M5 - 0:34.66 (1:39) K: 39/131 I: 2/29 S: 1/9 TS: 2 (2.23)

E1M6 - 0:36.97 (2:15) K: 30/177 I: 8/113 S: 1/4 TS: 5 (4.83)

E1M7 - 0:52.63 (3:07) K: 69/150 I: 17/84 S: 1/4 TS: 5 (5.34)

E1M8 - 0:16.77 (3:23) K: 8/41 I: 0/3 S: 0/1 TS: 6 (6.09)


slamo: Thank you for doing justice to what is arguably the most difficult Doom episode to speedrun. As someone who is somewhat familiar with the mechanics and tricks of Doom, the technical competence is very obvious. There are things that are obviously spectacular like the OOB in E1M8. There are also extremely difficult things that may not be obvious to the viewer, such as the skip at the end of E1M4, numerous micro-wallruns and thingruns, and relentless RNG manipulation to get damage rolls and move enemies. There was an improvement mentioned in the comments, but given the extreme technical complexity of the run, it's not really an issue.
Feedback was overwhelmingly positive. Accepting to Moons.
feos: Pub.
feos: Movie with extended input Userfiles/Info/68378703002771412

TASVideoAgent
They/Them
Moderator
Joined: 8/3/2004
Posts: 14877
Location: 127.0.0.1
Dimon12321
He/Him
Active player (479)
Joined: 4/5/2014
Posts: 1126
Location: Ukraine
Temp encode with Smooth Demo Playback enabled which makes this TAS more watchable: Link to video This is a much more technical TAS comparing to Episodes 2-4, but it also features a lot of sharp turns if SDP is turned off. I really like how gently you bump into walls while still preserving momentum.
TASing is like making a film: only the best takes are shown in the final movie.
Editor, Skilled player (1404)
Joined: 3/31/2010
Posts: 2086
Now this was a very pleasant surprise. Moving through those cramped corridors with such ease is nothing short of mindblowing, and the run looked incredibly smooth throughout. I'm so glad that there are finally all four episodes on the site. Yes vote. What the hell was that voidwalk!?
Joined: 2/28/2012
Posts: 159
Location: Philadelphia
Thanks for that encode, Dimon! My only entertainment/watchability complaint with this TAS is the jerky turning and that completely obviates it. The run is a real gem; it's amazing to see what new ground can be unearthed in a speedrun objective people have been attempting for two and a half decades.
GamesFan2000
He/Him
Joined: 1/4/2019
Posts: 75
Location: Canada
An excellent TAS of the game that essentially created modern speedrunning. An easy yes vote.
almostmatt1
He/Him
Active player (441)
Joined: 12/7/2020
Posts: 12
Thanks Dimon! That's a much higher quality upload than mine. And thanks for the kind words everyone.
PLANET
He/Him
Joined: 1/3/2018
Posts: 65
Absolute stunner : O This one could be a run for decades - who knows. Either way, absolutely marvelous, thank you so much for doing it! 14 years from the previous WR... ha. And so much time cut (almost 10% of Andrey Budko's TAS!). Kudos! Also, thank you Dimon for this encode, watched both and yours is definitely easier on the eyes (also, you have a great channel there!). Obviously, there can be only one option - HUGE YES vote : )
Banned User, Former player
Joined: 3/10/2004
Posts: 7698
Location: Finland
Sorry for asking trivial questions about Doom demo files, but I'm curious to know if they encode keypresses that are then "played back" by the game, or whether they just encode player and enemy position and orientation at each frame. I'm assuming it's the former. I also assume that the inhumanely fast turning speed is achieved by mouse turning controls. I suppose it could be theoretically achievable with real-life hardware?
Dimon12321
He/Him
Active player (479)
Joined: 4/5/2014
Posts: 1126
Location: Ukraine
Warp wrote:
they encode keypresses that are then "played back" by the game
Yes, that's right. You can read about Doom demos here: https://doomwiki.org/wiki/Demo
TASing is like making a film: only the best takes are shown in the final movie.
Joined: 6/22/2012
Posts: 81
I loved e1m8, you can think to do an run to all 9 levels?
almostmatt1
He/Him
Active player (441)
Joined: 12/7/2020
Posts: 12
Warp wrote:
I also assume that the inhumanely fast turning speed is achieved by mouse turning controls. I suppose it could be theoretically achievable with real-life hardware?
Turning this quickly can be accomplished in real time with a mouse, yep, these turning speeds using keyboard input aren't possible. I have thought about making a TAS using only the turning speeds available for keyboard input, but this would take far too long for the extremely niche product it would end up becoming, haha. This is going beyond answering your question, I know, but your comment got me thinking. Within a single frame there are technically 93,645,180 different input combinations possible, accounting for the full range of running speeds, strafing speeds, turning angles, weapon changes, potential firing, and "using" switches and doors. With the TASing program I used, XDRE, you can manually enter all of this information for each frame but this range of running/strafing speeds is all possible with real hardware by using the mouse to strafe and produce forward movement. Strictly theoretically speaking, everything here could be achieved with real-life hardware, but the extent to which many aspects of this TAS are out of reach to real life players cannot be overstated. TLDR version: it can get really complicated really fast. Longer version: Some rough maths - I forget the exact figure but I think the minimum distance that Doomguy can move in one frame, on each the X and Y axis, is somewhere in the neighborhood of between 1/63000 and 1/65000 of a unit (For reference, Doomguy is 32 units wide, so that's somewhere around 1/2,000,000 of the width of the player). That's on one axis, so considering doom movement is on an X & Y axis... well, as you can imagine the exact specifics of your position can vary very, very widely and some tricks really do require that level of granularity and specificity. For example, when performing a trick called the 0.000000 trick you need to be EXACTLY as close to a wall as you can be, and then your momentum can be maintained in multiple directions even while your position is not changing (adding to the difficulty of this, you then need to keep your momentum below a certain value or it then disappears completely) (...AND it only works on certain walls and when travelling in certain directions) (...AND when the walls happen to be on a blockmap grid 64x64 units wide, the laws governing momentum caps are changed). In order to collide with a wall AND be in this exact position when you do so AND do so in a way that you don't lose all speed upon contact AND have a high enough speed for the trick to be worth it but have a low enough momentum that you don't exceed the momentum cap, brute forcing the inputs for the frames immediately preceding this trick becomes absolutely necessary. Typically I'll set up the previous dozen or so frames manually in such a way that I can reliably find the correct inputs while brute forcing a relatively small number of potential inputs, but the way I set it up, the program searches through a range of 1,896,129 inputs over two frames. Going back to the question of physical hardware, the strafing/moving inputs would need to be controlled by mouse movement, and the frantic and extremely precise movement would be a very big ask of real physical hardware even if it were somehow mechanically operated. Anyway, speaking of that trick. Realistically speaking, when this is properly executed it'll typically save about five/ten or so frames. I just checked and I went through this process 30 times in this tas. Probably the most obvious example is at 1:20 of Dimons video, where I get thrown forward and to the left as soon as the door opens, having accumulated my momentum for 22 frames while the door opens. There's also the wallrunning trick which, while not as absolutely precise as the trick above, does require you to be very very close to a wall and making sure you get as close to the wall to make the trick work while not actually bumping into it and losing speed is quite complicated, so this generally takes quite a process of manually adjusting your movement and monitoring your position/momentum/speed data. Now that I've written all this, it strikes me that I'm doing so on a part of the internet that is well accustomed to the complexities and intricacies of TASing, I'm still getting used to that. Still, perhaps my rambling will be interesting to someone!
almostmatt1
He/Him
Active player (441)
Joined: 12/7/2020
Posts: 12
Hikaruon wrote:
I loved e1m8, you can think to do an run to all 9 levels?
It's a very cool trick! Going to the ninth level isn't required for a UV-Speed route, but other than an individual level UV-Speed run there are a couple of Episode 1 categories where going to e1m9 would be necessary. The Nightmare! All Secrets, No Monsters 100% secrets, and UV-Max (all kills and all secrets) categories would require you to go here. Not to derail the thread but Dimon12321 (in addition to being a cool dude and encoding the Episode 1 TAS above) has in the past done an individual level e1m9 UV-Speed TAS. Here it is if you're interested. https://www.youtube.com/watch?v=H-soLcUdRbA
PLANET
He/Him
Joined: 1/3/2018
Posts: 65
almostmatt1 wrote:
Warp wrote:
Now that I've written all this, it strikes me that I'm doing so on a part of the internet that is well accustomed to the complexities and intricacies of TASing, I'm still getting used to that. Still, perhaps my rambling will be interesting to someone!
Absolutely cool info there, man, actually enjoyed that :D Makes me appreciate the run more : ) Just a quick question - since the level statuses are skipped for the time sake... Could be please provide the level finishing times (optionally - if you want to do it - the whole info, as in kills/items/secrets % (perhaps even number/number format eg. 1/10) along with time/par/total) in the description or here? (the way that DooM normally shows it) If so - thank you even more! That would be pretty sweet.
CoolHandMike
He/Him
Editor, Reviewer, Experienced player (636)
Joined: 3/9/2019
Posts: 582
Yes vote! Lots of nice tricks in a frantic run to the finish. Always liked the look of these extremely fast DOOM vids.
discord: CoolHandMike#0352
Dimon12321
He/Him
Active player (479)
Joined: 4/5/2014
Posts: 1126
Location: Ukraine
almostmatt1 wrote:
Not to derail the thread but Dimon12321 (in addition to being a cool dude and encoding the Episode 1 TAS above) has in the past done an individual level e1m9 UV-Speed TAS. Here it is if you're interested. https://www.youtube.com/watch?v=H-soLcUdRbA
Thank you for sharing my video, but I uploaded it in a terrible quality back in the day. Here is a better one: https://www.youtube.com/watch?v=siH6lGqzGDk
almostmatt1 wrote:
Within a single frame there are technically 93,645,180 different input combinations possible, accounting for the full range of running speeds, strafing speeds, turning angles, weapon changes, potential firing, and "using" switches and doors.
Did you count turbo movement (MF51 ... MF127 and so on for other 3 directions)? As far as I can tell, nothing can produce the movement of this "power" unless -turbo 255 was added to arguments before launching Doom.
TASing is like making a film: only the best takes are shown in the final movie.
almostmatt1
He/Him
Active player (441)
Joined: 12/7/2020
Posts: 12
Dimon12321 wrote:
Did you count turbo movement (MF51 ... MF127 and so on for other 3 directions)? As far as I can tell, nothing can produce the movement of this "power" unless -turbo 255 was added to arguments before launching Doom.
With regular speeds there are 101 running speed values and 101 strafing speed values. With shorttics there are 255 turning values. You can fire or not fire, and you can use and not use. I did make a mistake, though, I accidentally counted there as being 9 different weapon switching buttons while there are in fact only 7, two of the buttons switch to two weapons. So the actual number is: 101 x 101 x 255 x 2 x 2 x 7 = 72,835,140. Whoops. With turbo speeds, the number is 464,278,500. EDIT:
PLANET wrote:
Just a quick question - since the level statuses are skipped for the time sake... Could be please provide the level finishing times (optionally - if you want to do it - the whole info, as in kills/items/secrets % (perhaps even number/number format eg. 1/10) along with time/par/total) in the description or here? (the way that DooM normally shows it) If so - thank you even more! That would be pretty sweet.
Somehow missed this comment! I will edit this information into the original description.
PLANET
He/Him
Joined: 1/3/2018
Posts: 65
almostmatt1 wrote:
Somehow missed this comment! I will edit this information into the original description.
Thank you so much for it, man, the comparison to Andrey's run makes it even better : ) Kudos!
Post subject: Side note...
PLANET
He/Him
Joined: 1/3/2018
Posts: 65
What is this message about: "ALERTS POSSIBLY COMPROMISING MOVIE INTEGRITY: Skill 3 1 players"? (top of the submission page) This is Skill 4... : P
Post subject: Re: Side note...
Dimon12321
He/Him
Active player (479)
Joined: 4/5/2014
Posts: 1126
Location: Ukraine
PLANET wrote:
What is this message about: "ALERTS POSSIBLY COMPROMISING MOVIE INTEGRITY: Skill 3 1 players"? (top of the submission page) This is Skill 4... : P
The site reads a Doom demo header incorrectly. For instance, here is how it reads a Boom-compatible demo.
TASing is like making a film: only the best takes are shown in the final movie.
Post subject: Re: Side note...
PLANET
He/Him
Joined: 1/3/2018
Posts: 65
Dimon12321 wrote:
The site reads a Doom demo header incorrectly. For instance, here is how it reads a Boom-compatible demo.
Jeez, that's totally strange. Thanks for explaining man!
Post subject: Re: Side note...
almostmatt1
He/Him
Active player (441)
Joined: 12/7/2020
Posts: 12
PLANET wrote:
What is this message about: "ALERTS POSSIBLY COMPROMISING MOVIE INTEGRITY: Skill 3 1 players"? (top of the submission page) This is Skill 4... : P
As Dimon said the demo header information is read incorrectly. As Samsara told me on Discord, it counts the difficulties as: 0 - ITYTD 1 - HNTR 2 - HMP 3 - UV 4 - NM Instead of 1-5. Dunno how "1 player" could be concerning though, haha.
Post subject: Re: Side note...
PLANET
He/Him
Joined: 1/3/2018
Posts: 65
almostmatt1 wrote:
As Dimon said the demo header information is read incorrectly. As Samsara told me on Discord, it counts the difficulties as: 0 - ITYTD 1 - HNTR 2 - HMP 3 - UV 4 - NM Instead of 1-5. Dunno how "1 player" could be concerning though, haha.
At least that makes sense! I wonder how the system understood that skill "29" up there ( http://tasvideos.org/userfiles/info/62362112029546753 ) ; )
Post subject: Re: Side note...
Dimon12321
He/Him
Active player (479)
Joined: 4/5/2014
Posts: 1126
Location: Ukraine
PLANET wrote:
At least that makes sense! I wonder how the system understood that skill "29" up there ( http://tasvideos.org/userfiles/info/62362112029546753 ) ; )
You can open it in a hex editor and try to figure which value from the header could be 29. I guess it's 1D
TASing is like making a film: only the best takes are shown in the final movie.
Joined: 12/7/2005
Posts: 149
Location: Sweden
Looks like 1 second could be saved on e1m3 by manipulating monsters to keep a door open, and perhaps on e1m5 (door to the central room) too?
almostmatt1
He/Him
Active player (441)
Joined: 12/7/2020
Posts: 12
Fredrik wrote:
Looks like 1 second could be saved on e1m3 by manipulating monsters to keep a door open, and perhaps on e1m5 (door to the central room) too?
Unfortunately for me you are correct. Those are things that the old 3:45 TAS does and I didn't, and are IMO the weakest points of my demo. The e1m3 one would would be difficult to make happen while still getting to the door as fast, though considering there was 28 frames (35 per second) between me opening the door and passing through it, I could technically spend close to a second of extra time performing different actions to manipulate their behaviour favourably while still saving time overall if I succeeded. What really sucks is that in e1m5 it seems to be simply a matter of going a little bit closer to the window on my way to grabbing the yellow key so that they see Doomguy and wake up, and they then seem to fairly reliably open the door for you that way - reliably enough that some basic RNG manipulation could probably get it to happen fairly easily. They're some frustrating things to realise only after having finished the demo. Other than a few movement imperfections in a few places, those doors are (to me at least) the only strikingly obvious ways that time could be improved and would be the most notable things to improve if someone were to rebuild this demo. There's a way to clip into the outside courtyard that you look at while in the void in e1m8 that doesn't require the help of an enemy - it's in fact the only way to beat the level in the "No Monsters" category. It's been theorised that it may be possible to perform a void glide from this courtyard, which would be faster than having to wait for an enemy to help you and would save a few seconds. Try as I might I couldn't get this to happen, but if a method is discovered I'd consider redoing this TAS with the aforementioned e1m3 & e1m5 improvements as part of the goal.