Uploaded September 2023 | Updated September 2026, 1 week ago
Last video I showed off a funky wall clip(technically it works in all directions but walls are coolest) that you can do by slipping into a float precise gap in block collision(youtu.be/wNyKwe5c_-M). The only issue is that doing that is rather hard! And also very time consuming! Both in-game and to TAS. This was mostly because I was doing it all manually though, cuz as it turns out 64 bit floats are very tiny!! And humans(at least this one) is are very not good at that!!
So why not force a computer to find inputs for me? Not only does it take negative time and effort from me(which is great, because TASing is a hobby(?) I want to spend as little time doing as possible), but also makes the setups are wayyy faster in-game. Cuz computers are apparently pretty good at math.... This lets you align yourself in like half a second in most reasonable situations, vs the like 20 seconds in my original video. The closer your position is to 0 on the axis the harder it will be since the floats are closer together though, which is why the one at 6m is a bit longer. I included it though because literally if you're off by 0.000000000000001m it just fails and that's insane. literal minecraft femtometer(surely that's a meaningful comparison) you cannot even perceive that difference. Which tbf doesn't help my case because it is impossible to comprehend the difference between that number and the 0.0000000000001m of the 270m one.
I also showed off a fun way to climb the walls faster once you clip in. I explain how this climbing works more in the original video above, but basically by moving on specific axes you can make the floors/ceilings tangible/intangible. So by moving just right you can make the ceiling disappear long enough to jump up, and then kill your movement to make the floor come back to land again. In the original video i sped this up by making the ceilings tangible just in time so I could bonk on them and shorten each jump, but turns out there's a faster way too. If you try clipping further into the block the server will tp you back and kill your momentum, which will conveniently cut the peak off your jump sooner, as well as make the floor tangible again. It's a lot easier to do rta too since it doesn't require any tick perfect inputs. Just jump and move parallel to the wall at the same time, and then hold into the wall once you get high enough. Plus if you're sneaking you can start moving a tick late.
And speaking of rta, since these setups are so much tinier, they're also much more ideal for eventually pulling off rta. I still need to work out the weird sensitivity slider stuff to get the right angles, but only needing to hit 4 angles is so much better than like 50. The inputs can also probably be simplified, but these ones can already be "pause buffered" pretty easily on singleplayer. But doing it on a server is also funny so idk give me some ideas to test it on(that aren't malicious). probably works on anything you can join in 1.14-1.17 and ig 1.18+ if you're brave
As for the actual how's and who's, here's the code I used: pastebin.com/na3YWYb7
Be warned I guess lol. Probably not the best way to do things, and it is kinda slow, but it works so idc!!! It just generates a bunch of individual tiny movements and tries to piece them together to get to your target. It's not perfect since it can't track /every/ input, and each movement has to be completely individual(usually by abusing the 0.003 velocity snapping stuff), but way better than I can do manually. It can also be a bit silly because the same movement can move you 1 float more/less depending on your exact position, but that's usually avoidable by just moving the movement orders around. i have no idea how float addition works tho so idk exactly what's going on there. Also probably doesn't work across powers of two
i don't know. enjoy the funny. perhaps i shall go turn myself into a human today
music: cat - c418
Last video I showed off a funky wall clip(technically it works in all directions but walls are coolest) that you can do by slipping into a float precise gap in block collision(youtu.be/wNyKwe5c_-M). The only issue is that doing that is rather hard! And also very time consuming! Both in-game and to TAS. This was mostly because I was doing it all manually though, cuz as it turns out 64 bit floats are very tiny!! And humans(at least this one) is are very not good at that!!
So why not force a computer to find inputs for me? Not only does it take negative time and effort from me(which is great, because TASing is a hobby(?) I want to spend as little time doing as possible), but also makes the setups are wayyy faster in-game. Cuz computers are apparently pretty good at math.... This lets you align yourself in like half a second in most reasonable situations, vs the like 20 seconds in my original video. The closer your position is to 0 on the axis the harder it will be since the floats are closer together though, which is why the one at 6m is a bit longer. I included it though because literally if you're off by 0.000000000000001m it just fails and that's insane. literal minecraft femtometer(surely that's a meaningful comparison) you cannot even perceive that difference. Which tbf doesn't help my case because it is impossible to comprehend the difference between that number and the 0.0000000000001m of the 270m one.
I also showed off a fun way to climb the walls faster once you clip in. I explain how this climbing works more in the original video above, but basically by moving on specific axes you can make the floors/ceilings tangible/intangible. So by moving just right you can make the ceiling disappear long enough to jump up, and then kill your movement to make the floor come back to land again. In the original video i sped this up by making the ceilings tangible just in time so I could bonk on them and shorten each jump, but turns out there's a faster way too. If you try clipping further into the block the server will tp you back and kill your momentum, which will conveniently cut the peak off your jump sooner, as well as make the floor tangible again. It's a lot easier to do rta too since it doesn't require any tick perfect inputs. Just jump and move parallel to the wall at the same time, and then hold into the wall once you get high enough. Plus if you're sneaking you can start moving a tick late.
And speaking of rta, since these setups are so much tinier, they're also much more ideal for eventually pulling off rta. I still need to work out the weird sensitivity slider stuff to get the right angles, but only needing to hit 4 angles is so much better than like 50. The inputs can also probably be simplified, but these ones can already be "pause buffered" pretty easily on singleplayer. But doing it on a server is also funny so idk give me some ideas to test it on(that aren't malicious). probably works on anything you can join in 1.14-1.17 and ig 1.18+ if you're brave
As for the actual how's and who's, here's the code I used: pastebin.com/na3YWYb7
Be warned I guess lol. Probably not the best way to do things, and it is kinda slow, but it works so idc!!! It just generates a bunch of individual tiny movements and tries to piece them together to get to your target. It's not perfect since it can't track /every/ input, and each movement has to be completely individual(usually by abusing the 0.003 velocity snapping stuff), but way better than I can do manually. It can also be a bit silly because the same movement can move you 1 float more/less depending on your exact position, but that's usually avoidable by just moving the movement orders around. i have no idea how float addition works tho so idk exactly what's going on there. Also probably doesn't work across powers of two
i don't know. enjoy the funny. perhaps i shall go turn myself into a human today
music: cat - c418









![Faster 5% Dragon Insta-Kill Setup 1.12 SSG
(if you want setup info skip towards the end :] Most of this is just context for my own selfish indulgence)
Recently I released a video showing off a potential setup for a 1.12 ssg insta-kill(https://youtu.be/VMHl0SOFw_0). And while it was good enough for people to hit it, it left a lot to be desired in terms of consistency and speed. Originally I estimated that ~3.8% of dragons would give you enough velocity based on 5000 random end fights that perched and charged within 30 seconds. I then eyeballed that about 50% of those would bonk on terrain and lose the necessary velocity before you could use it. This left a rough 2% estimate of hitting it, but it was also slightly inaccurate
(fwiw, almost exactly 30% of end fights met the 30 second charge restriction, and almost all were insta-perches. Which makes sense, since the 1.12 dragon is so fast that it can get to the 7/8 node faster than it can register the end crystals(which takes 5 seconds). As a result it only rolls a 1/3 to insta-perch since it doesnt think theres any crystals up, similar to how the first holding path only ever routes to the middle nodes. And factoring in 1/8 & other 1/13s, you get ~30%.)
Anyway, while confirming my physics n stuffs, I noticed that the dragon almost always spawns with a yaw of 229.78471. Im not entirely sure /why/ this is the case or if you can manipulate it, but it seems to come from the fact that it uses world random. Which conveniently gets reset right before dragon spawn if theres end island chunks that need to be generated(smthn smthn end city gen resets world random)(?). Regardless, its pretty epic cuz it makes the dragon paths a bit more predictable than I assumed. And it actually brings my original setup up to a 5% to get the velocity, ignoring bonks.
And! Speaking of bonks, I took the time to simulate the full client/server physics interaction during the insta-kill. This let me get actual numbers on how often/where the player bonks, and all in all it gave an estimated ~3% chance of that original setup hitting(or maybe a bit higher I forget the exact decimal)
But anyway, now for this setup. It again works by using the end terrain/obsidian pillars to strategically slow down different dragons and create hotspots where a tickle spot is more likely to be. So by standing in one spot when the dragon takes off you can set its target to get the right path, and then you can get the best chance of hitting the insta-kill by standing elsewhere in a good hotspot. And using blocks is just the easiest way to align yourself and prevent getting launched by the initial velocity transfer
SETUP INFO!
Pickaxe is kinda required, and you can do the setup up as soon as you want, its just hard to see. You just cant get much closer without sneaking or dealing with fireballs. All that matters tho is that you stand at (54.3, 59, 38.7)(Y important!) when the dragon takes off, and then move to (56.7, ~, 39.7) to shoot. Y isnt important there, but digging down in that corner is prolly best. This gives you a 7.00% chance of getting close to the tickle spot, and to help avoid bonking/snagging your Z server velocity you also want to place an endstone somewhere (47-55, 60, 39). Technically the lower X the better, but X 50 works fine. This gives you an overall 5.56% chance of getting enough velocity and not bonking, out of 100k end fights that perch and charge within 23 seconds of dragon spawn(vs 3.7% without the end stone). Of course this assumes you shoot tick perfectly(as shown in the vid, Keep in mind this timing is a few ticks later than the first setup cuz the dragon slowed), but again you only lose 9% of your velocity each subsequent tick. So better late than early, but it does get increasingly rarer to hit the later you are. This setup also averages 12.3 seconds from perch leave to death dance(12.2-12.55), vs 13.9(13.75-14.35) for the old setup. (oops the original times I had there were jank sorryyyy)
Anyway thats all for now. Hopefully better/faster setups coming soon, but maybe this can tide you over. Id wait until I had something better but this is More Dramatic and Also I Felt Bad. Im also considering looking at 1.9 setups(?) Im hesitant cuz hunger sucks, but on the other hand theres potential for some silly stuff. Its the only version with the bug that directly ties the dragons yaw to its head/neck hitbox heights(they literally use the dragons yaw in degrees to offset the hitbox y in meters). And since the dragon slows down partly based on those hitboxes, it might not slow down in scenarios where a 1.12 dragon would, or vice verse depending on wherever the head ends up. And since the dragon slowing down is what makes different setups good or bad, that could result in some fun stuff. Of course, Id actually have to look into it to find out, but thats like the only version difference anyway
also! these %s includes the backwards dragons, which always fail, and happen 4% of the time Faster 5% Dragon Insta-Kill Setup 1.12 SSG](https://i.ytimg.com/vi/sJOT9dejeno/mqdefault.jpg)
