[TAS] Scale the Mountain 0xA - 2377 fifdspence 2022-03-04 | This star was quite a journey. The previous WR by lemon (no YouTube video, but her channel is here: youtube.com/channel/UCoZ8JbiXlq_Zh0DGvaGRTDQ) uses the breeze to get to the second level, uses the log to get to the steep slope, and gets the star using an unusual kind of rollout discovered by Tyler Kehne ( youtube.com/watch?v=O6Wvz7XJIuw ). Lemon had also demonstrated that it was possible to use the fly guy to skip the log and get straight to the really steep slope, so there was a known improvement over the previous WR.Lord Snek wondered whether it was fast to use the TTM water upwarp to get to the second level more quickly. Here is an example non-ABC TAS by FramePerfection that does this: youtube.com/watch?v=TtCXT9fKy94I tried it, and initially thought it wasn't possible 0xA. Mario needs to get airborn to clip through the steep floor. Mario can dive recover to the bottom of the slope where the clip happens, but he loses speed rapidly as he runs, and he doesn't have enough speed to dive by the time he gets to the sufficiently steep triangles. There's shallow water available, though, so Mario can use that to dive again and DR again, keeping 48ish speed much higher and enabling him to do the clip.After that, Mario can enter the high water. I had a hard time avoiding the low water near the starting area, but I was eventually able to dive directly to the high water and upwarp to the waterfall area. However, I couldn't see a fast way to leave the water. It's way too slow for Mario to swim to a slope he can walk up. I was able to swim out of the Western side of the water into air and dive onto a slope, but it was a slope that led toward the bottom, and Mario inevitably slid back down.Fortunately, there was also the possibility of groundpounding. Mario can exit the water and groundpound to gain height. The floor triangle above where Mario upwarps is a little bit tilted, and it happens that he can get high enough to groundpound in some areas. Those areas were far South of where Mario upwarped, though, and the lower-down water section was blocking Mario's direct path there. So after the upwarp to the high water, I had Mario turn around, swim to the South (and also swim upward, as his water plunge with very negative vertical speed took him well below the top of the water) and then exit the water and groundpound. This was close to being competitive, but still a little too slow compared to the breeze strat.So I set out to get Mario to upwarp further South. I decided to land in the lower water after all, because when Mario exits water into air, he gets 1f of turning up to 11.25 degrees. So Mario can enter water, exit water, turn right, and continue with a better angle. This costs some speed, as Mario can only get back up to 32ish instead of 48ish due to drag barriers, but it's worth it: with enough straining, Mario could upwarp much further south. Then he just had to turn around and go upward.Turning around and going upward still took a long time. Mario was at terminal velocity before the upwarp, so he plunged very far down. I was able to save some time by having Mario upwarp to the high water, turn, exit the high water, strain back in to upwarp again (with only a little negative vertical speed this time), and exit again. This was now extremely close time-wise to the breeze strat.The final improvement was entering the lower water not once, but four times, changing Mario's angle each time. There's a small strip of space between the Southernmost part of the high water and the Northernmost part of the low water. (Note: this is not a float-perfect "water seam" as discovered by ds; it's 14 units thick.) This can be exploited: in the final TAS, Mario turns to face roughly East via four water upwarps on the North of the low water, does an upwarp to the high water on the South side, exits, and then upwarps again with very little negative speed. Mario no longer has to spend time turning around in water or going way upward, and he is facing in a good direction to swim out quickly and groundpound. As a cherry on top, Mario's groundpound spot is within reasonable slidekicking range of an edge, and it's possible to get to the fly guy pretty quickly to bounce to the slope. All this considered, the water upwarp is definitively faster.That was quite a mouthful! I might try to put together an explanation video if I have the time (though realistically I'll probably just want to continue TASing rather than working on a video).
[TAS] Elevator Tour in the Volcano 0xA - 4543 fifdspence 2024-07-28 | This TAS is based on an amazing strat found by youtube.com/@supermarble94 and youtube.com/@ds27315 . The original implementation (which was more of a proof of concept than an attempt at a highly optimized TAS) can be found here: youtube.com/watch?v=i_2K_2UOcNM Thanks to Marbler for re-TASing the bully battery segment for this new version. Also thanks to youtube.com/@Crackhex for a 1f improvement at the beginning, which also applied to Hot Foot It.The first of the two biggest improvements over that version is an improved bully escort to the crusher. This section is tricky to optimize for two reasons: 1. Mario has to keep the bully with him, and there are many different options for doing so. 2. Mario has to regain enough health that he can survive the lava boost section, which can either be done by collecting 5 coins or using the heart.I tested a variety of different routes by hand and ran a broad Scattershot run and settled on the heart route as fastest, though Scattershot didn't find it. I tried to Scattershot the full section in tunnel mode after that, but getting the bully over to the heart platform is surprisingly precise, and my Scattershot configurations weren't able to replicate it, let alone improve it. (I included bully state in Scattershot, but it had to be pretty coarse or the state space would really blow up). So I ended up running one Scattershot tunnel mode up to roughly when Mario passes the first fire spitter and one Scattershot tunnel mode from the bully reaching the heart platform to the crusher getting activated.I want to emphasize that I couldn't get Scattershot to find the heart route even in tunnel mode with the existing TAS as a hint, let alone having Scattershot discover it before humans in a broader search. So, though Scattershot is a really powerful tool, there's still room for humans in TASing even in meso-scale routing like this. Of course, the macro-scale strat finding Marbler and ds did here would be well beyond anything Scattershot can do.The second of two biggest improvements is the lava boost route to get the star. This was all Scattershot with an extremely coarse state space partition until Mario enters a certain radius of the star, at which point the partition becomes finer to be able to find dive grinds and such. It found a little slice of lavafall that bounces Mario to the right on his first lava bounce instead of straight back, which is funny to see.By the way, sorry for the weird star select textures. Trying out a switch from Jabo to gln and getting a couple of little oddities.
[Scattershot testing] Snowmans Land 100c moneybag collection fifdspence 2024-03-11 | I've been testing whether Scattershot can handle a certain segment of Snowman's Land 100 coins. I tested non-ABC, but this will be needed for the 70 star ABC TAS as well. I branched off of the previous WR with inputs by Alexpalix1 youtube.com/@PaLiX , Marbler youtube.com/@supermarble94, tjk youtube.com/@tjk113, and ideas by Eru youtube.com/@AruaErueru .After duplicating the moneybags (I think that segment was TASed by Marbler), we have to kill a bunch of moneybags and collect enough of their coins before continuing to the end of the star. The coin angles and velocities are determined by RNG, so I let Scattershot press camera buttons, since the camera affects RNG in snow levels.Scattershot was pretty effective at this. I think this is gets to the red roughly 110 logical frames sooner than the previous WR youtube.com/watch?v=Cxs6WsM67Xg . I didn't finish the star, as there are known improvements in the earlier part of the star. Note that the previous WR wasn't intended to be a super optimized version (I think it was mostly a proof of concept to get a console verifiable TAS using this strat), so humans may still be able to do similar to or better than this.This TAS hasn't been console-verified. Some SL 100 TASes using this strat have had the potential to crash due to a buffer overflow. The overflow isn't guaranteed to cause a crash, but it makes it possible. I checked in this TAS and I believe the buffer does not overflow, but a console verification would be nice to have. I'll update this description if a console verification is completed.I like how the HUD coin display lags behind the actual coin count. Mario can collect a coin per frame, but the HUD only updates every other frame and just increments by 1. This is why Mario appears to continue amassing coins until he's almost at the red.
[TAS] Scale the Mountain 0xA - 2333 fifdspence 2023-12-20 | This save was based on an idea ( youtube.com/watch?v=UL-PvJV1gVg ) by pannenkoek2012, who found two floors in TTM which aren't quite aligned properly and which have small gaps between them. These gaps can be used to grind up the last part of the mountain and make a straighter path than the previous 0xA WR.I had a lot of trouble optimizing the seam grind by hand, so I passed the task to Scattershot. Thanks to Krithalith ( youtube.com/@Krithalith ) for inventing Scattershot. I did a very long Scattershot run, since a lot of precision is required here: fractions of a unit or very small turns can make a big difference.
[TAS] Bob-omb Battlefield 100 coins + Reds 0xA - 1:4580 fifdspence 2023-08-08 | It's time for another BoB 100+reds version! This one was sparked by Marbler youtube.com/@supermarble94 , who1. Found a great triangle near Chain Chomp for speed-building2. Came up with a new route for the main speed-building section, most notably cutting through the center of the level to that great triangle instead of around the edge3. Demonstrated a fast dismount from the shell by sliding off the edge of the island.The TAS is an improvement of over 15 seconds since the last TAS: youtu.be/lUHtKsDiYkE Most of this is because of Marbler's findings. There are a few other tweaks, some of which were enabled by much better RNG manipulation tools since the last time I TASed this star. Here are some tweaks:1. I changed the first speed-building section to use Marbler's triangle as well. It requires Mario to wait quite a while between killing some of the enemies and collecting their coins, so there's a lot of RNG dependence there.2. Around 1:14, you see an idea Tyler had years ago about using water bombs to explode bob-ombs from far away. I think it ends up looking pretty funny. The water bombs spawn directly ahead of Mario at a distance proportional to his speed, so first we snipe a faraway bob-omb, then land on a sign and walk at low speeds to hit a nearby one as well. Mario still has to lure the last two bob-ombs to each other. A water bomb tried to get revenge on Mario while he grabbed the cork box around 1:21; I had to brake a couple of frames before diving to manipulate the spawn location.3. Different order and more aggressive timing for collecting the wooden post coins, made possible by the better RNG manipulation tools.4. Chain Chomp gets in the way of its post a lot more with this faster route. Mario's new movement at 0:54 doesn't lose as much time as it appears to - about 1-2f compared to the ideal Chomp, if I remember right.
[Scattershot TAS] - Shoot to the Island in the Sky 0xA - 2837 fifdspence 2023-05-17 | This is outdated by 1f. ChouxZi3 implemented a save originally demonstrated by Marbler in this video, which I had forgotten about: youtube.com/@supermarble94 In my TASes of this star, I always spent 1f on the box shedding speed, but Marbler had showed it was possible to avoid this and still stay close enough to the box to break it. Thanks ChouxZi3 for catching and fixing that.The story here is pretty much the same as the story of Behind Chain Chomp's Gate: youtube.com/watch?v=rqGUtrdW3KA Marbler youtube.com/@supermarble94 spotted a floor triangle that is really effective for building shell speed, and Scattershot seemed like a great tool to find a good shell speed route and optimize it. Shoutout to Krithalith youtube.com/@Krithalith for making Scattershot.From the shell grab to landing on the box is mostly just one big Scattershot segment, with an extra round of classic optimization tools applied to the air segment to get Mario in a good position for falling off + groundpounding the box as quickly as he can.Next goal for 70ABC (BitFS aside) is BoB 100+reds optimization. Marbler has already demonstrated some significant improvements.
[Scattershot TAS] Behind Chain Chomps Gate 0xA - 2400 DSG timing fifdspence 2023-05-02 | Long time no see! I've been in and out of TAS for the last year, and trying to contribute to N64 BitFS 0x has been my main focus lately when I work on TAS. Honestly, 70ABC is in a state where I'm pretty happy with the non-BitFS parts: Marbler youtube.com/user/supermarble94 and lemon youtube.com/@-lemon have done amazing work on optimizing the upstairs stars, so now every star in the run has had a solid TAS since the project started. The least optimized ones are probably the ones early in the run, from when I was a much weaker TASer and optimization tools weren't yet developed.As for the video, Marbler was looking at potential improvements to BOB 100+reds, one of which was going through the center instead of all the way around the edge of the arena during a speed building section. While trying it out, he found a very slippery triangle near Chain Chomp's Gate that was steep and built up speed really quickly. He saved several seconds using it, and then realized that it might also save time on Shoot to the Island in the Sky, which he confirmed soon after.I decided to try using his triangle on Chain Chomp's Gate. This seemed like a great application for Scattershot, so I put it to work, and it proved that Marbler's slope is faster for Chain Chomp as well! This is -46f from the previous TAS. No need to slowly build up speed on the slope near the Koopa. Shoutout to Krithalith youtube.com/@Krithalith for making Scattershot and helping me learn how to use it. I used the classic optimization tools for the final bomb grab+bloat section.We're planning on updating Shoot to the Island In the Sky and 100+reds with Scattershot as well. Could be a long time, though, as BitFS is my main TAS focus
[TAS] TTM 100 + Reds 0xA - 1:5805 fifdspence 2022-06-01 | The timesaves over the previous WR by lemon ( youtube.com/watch?v=PfgZ6rjrq_Y ) are kind of scattered throughout the star. The flyguy twirl to the steep slope, which lemon discovered, is probably the biggest single save. Some others in no particular order:Canceling the crazy box while going up instead of downGetting the yellow coin line more efficiently (I used an easy method that just looped around much further in a previous LOTAD, which lemon also used. Iwer Sonsch youtube.com/channel/UCnJKlalNQWIco1DS0Uw_XXw demonstrated that it was possible to get the fifth yellow much more directly.)Better RNGLess time spent rolling the logRun off after the 5th red instead of slidekickBetter slide sectionThere are some quarterframes throughout the TAS that I really struggled to save with the nature of some of these ledges and slopes. There's also a pause buffer in the log rolling section that I tried to remove with grinding, but I couldn't manage it. Finally, I didn't use the optimizer on the slide section - it would have been very tedious to run it on each part, and I think the slide is kind of boring personally. So there is probably more room for improvement here than in most of the recent 70ABC stars.
Monkey Cage 0xA ending strat comparison fifdspence 2022-05-05 | This is a time comparison of a "regular" end to Mystery of the Monkey Cage to the DSG ending. More info about the DSG and the development of this star is in the description here: youtube.com/watch?v=T5kLlm9L2VI Thanks again to pannenkoek2012, tjk, Tyler Kehne, and nim for their contributions to this star (mentioned in the other video). Thanks also to jongyon ( youtube.com/c/jongyon7192p ) for helping optimize the DSG reentry.By the way, you might notice that Mario talks to Ukiki only about half a second later for the DSG strat, but the star cutscene starts about a second and a half later. This is because Ukiki spends extra time turning toward the star from the angle where Mario talks to him. The second delay is hard to get rid of because Mario needs every frame he can get to grab the bob-omb fake. Both delays are also hard to get rid of because the object slots for the cloning don't work out unless a bowling ball breaks slightly before the star spawns, and bowling balls only spawn every 64f (and are RNG-independent, so they can also only despawn every 64f barring interference).
[TAS] Mystery of the Monkey Cage 0xA - 5400? fifdspence 2022-05-01 | Thanks to pannenkoek2012 ( youtube.com/user/pannenkoek2012 ) for doing the star cloning, tjk ( youtube.com/c/SuperMario142MC ) for work on the initial Ukiki chase, Tyler Kehne ( youtube.com/c/TylerKehne ) for the idea of cloning the star for DSG, and nim ( youtube.com/c/nim64 ) for the idea of manipulating object slots using a bowling ball.This strat saves time using a Death Star Glitch, in which Mario dies in one of a couple of ways (in this case, using the death barrier) and then collects a star. Mario doesn't have to do the star dance animation or decide whether to save after exiting the painting, so DSG saves a lot of animation time and can save time overall if it can be done without adding too much time inside the level.Tyler Kehne originally proposed this idea in May 2021. In this case, the Death Star Glitch uses a clone of the star: Mario can bring the clone down with him to the death barrier and then groundpound to collect it. My initial instinct was that the bob-omb would be too far from Ukiki to talk to Ukiki and then grab the bob-omb fake before the star spawned (note that the fake star in the cage is a different object than the actual star that gets spawned, so we can clone the actual star). I realized that it was possible to conserve speed using Ukiki's text, though, and Mario is able to grab the bob-omb fake with a few frames to spare.At this point, I was feeling pretty optimistic about the strategy. What I didn't realize, though, is that the cage creates a bunch of objects when it breaks. Mario needs to be holding a very high VS before the cage breaks to be able to get the star. After trying to optimize things to squeeze out more frames to make dust, I still couldn't do it, so I gave up for the time being.nim had the idea of using a bowling ball to manipulate object slots when the strat was originally discussed. The bowling ball timing would change as the TTM climb got optimized, and I didn't look into it at that time. Now that we're optimizing TTM stars, I started investigating the idea in earnest. It turns out that a bowling ball creates a bunch of objects when it breaks. Bowling balls only spawn when Mario is lower than the bowling ball spawner, though, so I took Mario down to the goomba area while waiting for Ukiki to run through his path. This allowed a bowling ball to spawn in at a pretty good time.Bowling balls can come on any multiple of 64 frames from level load. To get the bowling ball objects to help, they have to despawn at the right time, so I talked to Ukiki 16f later than the earliest possible frame. This let the bowling ball break relative to the star spawn time in such a way that Mario gets a really high VS, which we need to clone the star. I was having trouble getting the cloning working, so pannenkoek2012 did the cloning. I then imitated what pannenkoek2012 did while getting Mario as close to the edge as possible so that he can drop off the ledge in 1f.Note: This star uses DSG timing ("?" has become the notation for this), which is from level entry to the first frame of the death fadeout. In this star, DSG saves 1.15s over a normal non-cloning finish assuming Mupen lag emulation is correct and assuming the DSG doesn't affect the reentry time (both false assumptions, I'm afraid, but I think it saves at least 1s.) At some point, I'll upload a comparison of how the end of the star looks with each approach.
[TAS] Breathtaking View from the Bridge 0xA - 2060 fifdspence 2022-03-24 | Thanks to Fraims for the idea of the end of this TAS (channel: youtube.com/channel/UCf2cB6aFEEqj7HB7EWZrE3Q ).Previous TASes of this star (such as the previous WR by lemon; no video but her channel is here youtube.com/channel/UCoZ8JbiXlq_Zh0DGvaGRTDQ) climbed to the top of the mountain and dove/fell onto the star from above. Fraims brought up the idea of using the fly guy to twirl directly to the star. It would require some manipulation, as the fly guy has to climb quite a bit. Also, Mario would have to use the log to get to the steep slope instead of the fly guy, which is a bit slower. However, I did some tests and determined that it was at least possible to beat the traditional approach by a few seconds by hacking the fly guy to the right spot on the right frame. So it was still plausible that the fly guy manipulation would be quick enough for the strategy to be worth it.I initially messed with some ideas that had the fly guy follow Mario up from behind while he climbs the steep slope, but these were all too slow. I eventually found the approach in the video, though, which I describe below. The key step is step 5: I'm able to make the fly guy climb for a while, stop spiraling, and then climb again very soon after, giving me good control over his angle before a final raise.1. Make fly guy lunge north (toward the waterfall) and climb a bit2. Make fly guy spit fire once to give the log enough time to turn3. Make fly guy lunge toward Mario (southward) right as Mario DRs from the log with rng such that fly guy turns clockwise4. Climb the slope a bit to enter the high zone, making fly guy climb and spiral5. Go back down to the medium zone so fly guy stops spiraling and his y speed returns to 06. Go back into the high zone again with rng such that fly guy goes straight this time. Step 5 should be timed so that the fly guy is angled toward the spot where Mario needs to hit him.7. Go up the slope as fast as possible to hit the fly guy as low as possible while still making it to the starThat results in the strategy you see here, which is actually about 2-2.5s faster than the traditional approach. I was pretty happy that this is faster, not only because it gives more variety between stars, but also because it gives an opportunity to manipulate twirl yaw. Mario's angle when landing from a twirl depends on his twirl yaw, which is conserved between levels, and his angle when landing on the steep slope is very important for Mystery of the Monkey Cage (which still needs to be TASed) and Scale the Mountain.
[TAS] Mysterious Mountainside 0xA - 907 fifdspence 2022-02-23 | The first part of this TAS is by lemon (up to when Mario enters the wind). I branched off of her TAS here: youtube.com/watch?v=ngkKjKfu2METhis new TAS is 2f faster. One of the frames is saved by landing on the ledge 1f sooner. It's very difficult to do this and still slide off on the next frame; I relied heavily on an optimization tool while doing this. The other frame is saved by only spending 1f in freefall after sliding off instead of 2f.I tried to roll out 1f sooner and still get the star, but I couldn't do it even with the optimization tools. The closest I could get was 4 units away (4 centimeters). I also tried to use the misalignment to land from the wind sooner, but I don't think it's possible to cancel that without rolling out or waiting for Mario's slide to finish, both of which would be way too slow.Recently, I've been spending some of what would normally be TAS time building some RNG manipulation tools. I'm thinking it makes sense to keep working on upstairs while work on BitFS 0xA progresses separately, and then use the manipulation tools to try to resync once BitFS is done. This will hopefully make it possible to work on upstairs without any effort being wasted due to different RNG. SL 100 won't be possible to resync unless we get really lucky, but other than that I'm guessing everything can be resynced.
[Outdated TAS] Whirl from the Freezing Pond 0xA - 1557 fifdspence 2021-12-15 | This TAS has been improved by Marbler: youtu.be/WDcbyk1IOyUThis is a 6 frame improvement over the previous WR by lemon: youtube.com/watch?v=mXb8zBAabtsLife has been a bit busy recently, but I've still been picking away at the upstairs stars from time to time. I did this one a while ago and thought I might still be able to improve it by 1f more, but I haven't managed it yet and decided to upload it as-is.If I remember correctly, there are actually two tied strategies for the final part, though I don't have the file for the other approach. lemon does a dive recover into the spindrift instead of the freefall grind I do here. The dive recover has less speed, but it doesn't get the bad quarterframes I get during the standing grind. This combined with some tradeoffs with Mario's angle make these two strategies tied, as far as I can tell. The ending box break is actually pretty difficult, as Mario has to lose a lot of speed to be able to break the box and then stay within the star's future hitbox while the box breaks, so I wouldn't be surprised if I've missed a frame at the end by not landing on the spindrift 1f sooner in one of the strategies.Typically, it wouldn't be difficult to get perfect quarterframes in a grind like this where Mario spends multiple consecutive frames in freefall. However, the ledge Mario is grinding on isn't aligned with the floor next to it, so there are a bunch of out-of-bounds unit squares peeking through. If Mario hits any of these unfortunate squares, he either bonks (if he's in an air state) or his movement gets cancelled for that quarterframe (if he's in a ground state), so instead I avoid them at the cost of some quarterframes.
[OUTDADATED] [TAS] Metal Head Mario can Move 0xA - 4857 fifdspence 2021-11-06 | This TAS has been improved by ds273: youtube.com/watch?v=dI_vMSjagAY&feature=youtu.beThis is an improvement of 1s over the previous TAS ( youtube.com/watch?v=sD270bDQpAc ) by Adeal, Iwer Sonsch ( youtube.com/channel/UCnJKlalNQWIco1DS0Uw_XXw ), ds273 ( youtube.com/user/Th3DekuScrub ), and ChouxZi3 ( youtube.com/channel/UC2BRYSlctEGCRT6iKdk9kEA ).The improvement of 1s comes from a few sources. I optimized the beginning a little more, the most notable way being losing a lot of speed before the c^ slide. Mario breaks out of the slide once his speed reaches a certain negative value, so starting with less speed is better. The platform gets several wall pushes, an idea suggested independently by jongyon ( youtube.com/user/jongyon7192p ) and Fraims. Finally, I spend 1f less pushing the elevator to the cavern wall, since it turns out Mario can still clip with it a little bit further.This star used to have a really different strat, but due to a clip found by Jeaze ( youtube.com/channel/UCujLwF9ObZJguRUdDedqmqg ) and brought up in this context by MartinTheGamer ( youtube.com/channel/UCkPhIIURI5CweD4FO6datWg ), it was found to be possible much more quickly this way. Adeal came up with a bunch of different strategies, Iwer got it really close to working, and I set up a program to optimize the slidekick the last little bit more. jongyon had already demonstrated that Jeaze's clip was possible 0x, so then it became a matter of time optimization. ds273 brought up the idea to use a wall overlap to clip through at the edge of the cavern instead of the corner, so it's possible that this star could get another big update in the future if we're able to solve that.
[TAS] Blast Away the Wall 0xA - 1410 fifdspence 2021-10-22 | Not too much going on in this one. The only change relative to the previous WR ( youtube.com/watch?v=F-Vcl_CEd-w ) is that Mario grinds along the ledge just before the stairs instead of running around by the lower Thwomp.
[TAS] Fall Onto the Caged Island 0xA - 3167 fifdspence 2021-10-22 | This star has been improved in three ways compared to the previous WR ( youtube.com/watch?v=psNg9MFr5eo ):The start of the misalignment climb is a bit faster now. Instead of running in a loop around the lower Thwomp, Mario lands along the ledge and then dive grinds upward.A strat discovered by ds273 ( youtube.com/user/Th3DekuScrub ) allows Mario to climb the steep floor triangle to the King's platform more quickly. By remaining cardinal to the ledge, Mario doesn't get pushed out by the wall next to the steep floor triangle, and this lets him reach the King more quickly. Additionally, he is no longer reliant on the elevator cycle.Previously, I ran up to speed before doing some divegrinds on the King. I had to wait for the King to turn pretty far counterclockwise, since Mario divegrinds along the edge of the King, and the King starts turning back toward Mario once Mario crosses the line straight through the center of him (putting Mario on his right rather than on his left). This time, I Eru-drop off of the edge of the King, letting Mario build up some more speed while in the air, which lets Mario begin the dive grinding with more speed. In fact, Mario is even able to spend some air frames on the final dive grind and skip the pause buffer.
[TAS] WF 100+To the Top of the Fortress 0xA - 13228 fifdspence 2021-10-10 | The need for an update to this star became clear after ds273 discovered a new tech, commonly called BCA (for Behind Camera Anywhere, though I like the name See No Evil). You can see ds273's video about it here: youtube.com/watch?v=C38He-OUAYk Thanks to pannenkoek2012 ( youtube.com/user/pannenkoek2012 ) for teaching me how to do it in practice.BCA is a way of avoiding HOLP updates without having to use one of the older techniques like pause buffered hitstun or hat-in-hand. In the previous version of this star by me and Iwer Sonsch ( youtube.com/watch?v=nHx7r6T9ROA ), we didn't yet know about BCA, so we had to use pause buffered hitstun to throw the corkbox to the HOLP. That requires diving from the platforms way up high and taking big fall damage, which constrained the routes to go to the high platforms late. This new TAS uses a different route that goes to the high platforms earlier; I believe this is somewhat faster.The star is cycle-based (with period 256f=8.53s) due to the sliding platform that lets Mario get on the tower, so the question was whether we could save a cycle. The new route helps, and so does another discovery by ds273. ds273 found a way to climb up the steep triangle more quickly and without using the elevators. This not only removes the cycle dependence on the elevators, but it also saves time. This makes a cycle save pretty easily possible; I did a low-optimization TAS of the final climb, and it turned out to already be more than good enough. I don't think saving another cycle is possible. You might notice that the save is more than 8.53s. This is because the tower platforms don't move while the game is paused, and the old TAS spends longer pause buffering than the new TAS spends switching camera modes.The yellow coin spawned by the cork box requires pretty precise RNG to be near the bullet bill blaster at the right moment. Since the TAS is shorter than the old one, I thought it would advance the RNG index by less, and I would have to make extra RNG calls to get to the appropriate index. To my surprise, my first attempt advanced it much more. It turns out Mario was entering the radius of the triple butterfly spawner on the way to performing BCA, and then the butterflies called RNG 6 times per frame for several seconds while Mario did BCA. Luckily, the radius of the spawner is small enough that Mario could avoid it by swinging wide.I was originally planning to redo the tower climb by hand, but given the likely need to resync a number of upcoming stars, I thought it would be nice to have a more automated solution. I was able to set up a tool that tried to match some key values (position, angle, speed) on each frame. I ran it overnight and it was able to redo the ending, matching what was done by Iwer Sonsch in the previous version.
[TAS] Chill with the Bully 0xA - 1260 fifdspence 2021-08-18 | Thanks to Manama ( youtube.com/channel/UCqweAVPadUxg4OLRNjLZt2g/videos ) for pointing out the possibility of backward-sliding to the star and Xander ( youtube.com/c/EpicRandomGamingChannel/videos ) for pointing out that holding neutral while landing on the ice platform keeps more speed than matching yaw. Thanks also to RSW ( youtube.com/channel/UCQl5iBqmGpB7e90QmsLhIWQ ), jongyon/Dan Park ( youtube.com/user/jongyon7192p ), and Crackhex ( youtube.com/channel/UC_9O61Cu6DpvjDVMLWdUGCQ ) for other suggestions and ideas. This star is relatively similar to the non-ABC world record: do a c-up slide to get to the bully quickly. The implementation is pretty different, though. The non-ABC record enters a bellyslide after going up the slope, but its way of doing so requires A. The best option here is to punch up the slope and then slidekick instead of doing a rollout.To collect the star, the classic 0xA approach is to lava boost. This turns out to be significantly slower than just sliding backward into the star. Mario gets -16 speed from hitting the bully, he's able to get about -18.7 using some sliding speed shenanigans, and due to the slippery floor, he can keep -18ish speed most of the way to the star. During the backward rollout, it's best to strain using what's known as the -15.99 technique, giving Mario some extra distance on each frame.
[TAS] Shocking Arrow Lifts 0xA - 1377 fifdspence 2021-07-29 | I've decided to start looking at some hexable upstairs stars. I've had my eye on this one for a while, as I've always hoped to see the 70ABC project improve every star, and this was one of the most daunting prospects. The previous WR by Iwer Sonsch ( youtube.com/watch?v=D-PSVwUpdME ) is very good, and I think it's optimal for that strat.I went with a different approach at the beginning of the video, since Mario starts off closer to this edge of the starting platform. He is able to get extra distance via a good wall push from the slat, which makes up for the fact that he runs in a worse direction at the beginning. This strat barely saves 1f; I ended up using an automated optimizer to get the straining just right. Mario does not land within the warp's hitbox, but the walking frame before the crouchwarp is enough to enter the hitbox.
[TAS] Board Bowsers Sub 0xA - 12127 fifdspence 2021-07-28 | Epilepsy warning: Rapid camera presses to reduce lag.This star is made much faster by B swimming speed conservation. When Mario swims using B, he can't gain speed past 7, but he can conserve up to 16 speed. Therefore, for some stars, it's worthwhile to build up speed. pannenkoek2012 demonstrated that this could save time ( youtube.com/watch?v=LTmGnIcEZvw ). Iwer Sonsch found good triangles to use for this in the previous ABC WR for this star ( youtube.com/watch?v=oBCn4mjkTvA ), hyperspeed walking along one slope and hyperspeed punching along the next. Iwer conserved 15.44 B swimming speed, and I was able to get it to 15.52. There's actually another pair of triangles that can be used to get more speed ( youtube.com/watch?v=-7UmwBQmczo gets 15.58, and more is likely possible) but it's too far away to be worthwhile.The swimming speed buildup also happens to give a sliding speed vector that lets Mario reactivate it onto land after the swimming is over. Then he uses misalignments to climb the boxes and reach the deck to grab the star.A tricky aspect of this star is lag reduction. (Note: The 1'21"27 is according to Mupen lag.) Mupen's lag emulation is far from accurate to console lag, but since we can't TAS on console, I tried to reduce Mupen lag (though not at the cost of logical frames). During the swimming part, Mario's joystick inputs have the same effect regardless of the camera, so it wasn't too hard to play around with different camera movements. I just TASed all of the swimming (with the help of an automated optimization tool) and then changed the camera presses without changing the joystick. For the air movement parts, though, I had to redo the joystick inputs a couple of times for each section. I'm sure the lag reduction is improvable with respect to Mupen lag, let alone console lag.To make matters more confusing, RNG seems to affect lag, at least according to Mupen. I suspect it's because of bubbles that spawn on an RNG basis.
[TAS] The Manta Rays Reward 0xA - 3568 fifdspence 2021-07-28 | You may know that the manta ray rings have a bug that causes their actual hitboxes to differ from their graphical appearances. This video uses a hack that displays the manta ray rings as Mario actually experiences them. Thanks to nim ( youtube.com/channel/UCG_b-lywDked3HfbZQXU8Gw ) for the patch.This star is different from other DDD stars in that it's not fastest to build up B swimming speed. Mario can get the rings and star quickly enough that the extra time spent to get speed isn't worthwhile. This star was also optimized using automated optimization tools.
[TAS] Chests in the Current 0xA - 10370 fifdspence 2021-07-28 | This star is made much faster by B swimming speed conservation. (Edit: As pointed out by Tyler Kehne, it's actually impossible without B swimming speed conservation.) When Mario swims using B, he can't gain speed past 7, but he can conserve up to 16 speed. Therefore, for some stars, it's worthwhile to build up speed. pannenkoek2012 demonstrated that this could save time ( youtube.com/watch?v=LTmGnIcEZvw ). Iwer Sonsch found good triangles to use for this in the previous ABC WR for Board Bowser's Sub ( youtube.com/watch?v=oBCn4mjkTvA ), hyperspeed walking along one slope and hyperspeed punching along the next. Iwer conserved 15.44 B swimming speed, and I was able to get it to 15.52. There's actually another pair of triangles that can be used to get more speed ( youtube.com/watch?v=-7UmwBQmczo gets 15.58, and more is likely possible) but it's too far away to be worthwhile.I repeatedly TASed a segment of this star, fed it to an automated optimization tool, TASed a segment, etc. It's a bit of a tedious process, but I'd rather not try to optimize swimming by hand. I'm glad to be done with the swimming-heavy stars.This star descends very quickly using a ceiling, similarly to a trick which is used in Board Bowser's Sub to a lesser extent. It's nothing precise at all; when Mario swims into ceilings, he gets pushed down, and what appears to be a wall is actually a very steep ceiling. This lets Mario descend extremely quickly.
[TAS] WMotR cannonless 0xB fifdspence 2021-07-17 | sm64expert ( youtube.com/channel/UC6uLsNH0YepNCECAHLAz1VA ) demonstrated in an unlisted video that WMotR can be completed cannonless. Their technique, which is also the TAS WR, uses triple jumps off of the wing cap box to build speed. However, it requires speedkicking, so this brings up the question of whether it can be done 0xB as well.It turns out to be possible via a combination of a few ideas. Initially, I had an idea that alternated between the two clouds I use at the beginning of this video, grinding into a double jump and then triple jumping either off of the wing cap box (on the starting cloud) or the edge of the cloud (on the other cloud). However, I couldn't get enough speed this way. Speedkicking on these slopes loses no speed, whereas sliding off does lose some speed. Additionally, Mario can only slide off of a few of the slopes while having a good angle to reach the wing cap box, and those slopes aren't spaced conveniently.Fortunately, there is a technique found by bad_boot ( youtube.com/watch?v=826gWUnF-cM ) that turns out to be really helpful. To my limited understanding, Mario's speed increases/decreases proportional to his downward/upward pitch, whereas sin(pitch) would be correct in aviation theory. This means it's beneficial to point far down for a bit and then slightly up for a long time. Mario experiences a constant drag per frame (independent of his speed) but the pitch effect is enough to overcome this. The technique allows Mario to increase his flying energy, though it takes quite a while, which is tricky with the limited wing cap time. It helps that Mario can change his pitch very quickly while facing out of bounds (OOB). I started my own TAS up to reaching OOB, tried to sync up to the speed, pitch, and pitch velocity bad_boot had in the best demonstration of the technique, and then threw it into an automated optimization tool I have been using (also derived from work by bad_boot, and also Krithalith youtube.com/user/Kiefertje ).I wasn't quite able to reach the cloud with enough height, as Mario lost some speed while turning. Fortunately, jongyon ( youtube.com/user/jongyon7192p ) pointed out that Mario can turn a tiny bit without losing any speed. This is why Mario is extremely jittery while facing OOB. I alternate between 0 joystick x and the minimum joystick x, which lets Mario turn a bit every other frame. This was finally enough to barely reach the cloud. If you listen, you can hear how soon Mario's wing cap expires after reaching the cloud.For those of you wondering about the 70ABC TAS, it's kind of paused for the moment. This challenge and the next TAS competition task have kept me busy, and social life has been returning to normal post-vaccine. Also, it may be necessary to change the start of the TAS to switch to mono (!) for BitFS, so working on extending the 70ABC WIP isn't a great use of time at the moment (though working on non-RNG non-global-timer stars is still a good use of time). I still feel dedicated to the TAS, but it may be a while before I have more to upload on that front.Thanks to Xander ( youtube.com/channel/UCL8JNpJkmcX43bxcWhsXfAA ) for the encoding lua used in this video. I don't think I've uploaded videos with related lua scripts before - as informative as they are, I used to find them hard on the eyes. Now that this nice-looking one is available, though, I'll try to get in the habit of using it.
[TAS] [Outdated] Frosty Slide for 8 Red Coins - 3423 fifdspence 2021-05-26 | This star has been improved by CeeSZee and Frame ( youtu.be/OjMG6odTNNQ for next WR and then subsequent videos).0:00-0:12 and 0:24-0:29 are by Plush.Took a break from ABC to look a bit at 120, since JoshDuMan ( youtube.com/user/joshduman ) has posted a bounty for improvements on the 120 star WIP. I've been wondering how the automated optimization tools I've worked on for ABC swimming (based on a tool by bad_boot youtube.com/channel/UC280b7IZCFnzCR0RCwi6JYg and on similar efforts by Krithalith youtube.com/user/Kiefertje) could be applied to more general movement. It takes a lot of manual tuning, but it turns out to still be useful for other tricky movements like air turn and sliding. Compared to the previous WR by Plush ( youtube.com/watch?v=Uax1Cw8vhrA ), this is -5f from the third red to the warp and -1f from the seventh red to the finish, for -6f in total.The tool perturbs a time interval of a TAS's joystick inputs and tries to minimize some quantity. It's not set up to handle different button presses, as this would be a lot more difficult to manage well (probably frequently having to check different actions), so I usually had it optimize a segment, manually redo the next part of the TAS, optimize the next segment, etc... This is a very slow process, and I only expect to do it when I really want something extremely optimized or when I can't optimize it reasonably by hand.More specifically, I did the following steps for the segment before the warp:1. Note the position on the frame before the cabin glitchy wall kick. Minimize distance to this position. I had it try to reach this position 1f sooner than the previous WR did, and it was able to do so, but it couldn't 2f sooner.2. Redo the glitchy wall kick and then minimize -speed after the gwk with an infinite penalty if Mario didn't gwk (detected by checking yaw).3. Re-TAS the next part by hand, trying some different numbers of jumping frames to find a good quarterframe on the landing. I experimented with a couple of subsequent movements and settled on the one shown in the video. I'm not sure whether the new sliding frames are better than what was done in the previous WR; it might depend on Mario's speed after the gwk.4. Note the position on the frame before long jumping into the next wallkick. Minimize distance to that position along with a small penalty for bad yaws. The TAS reaches the wallkick 4f ahead, so 3f were gained between the glitchy wall kick and that wallkick.5. Re-TAS an attempt at getting the downclip after the wallkick (did not actually downclip by hand, though). Minimize a binary objective function that just checks whether Mario failed the downclip.6. Minimize -speed after the downclip with a small penalty for more clockwise angles and an infinite penalty for no longer having the downclip.7. Re-TAS up to the warp by hand.The part after the warp was much simpler; I just jumped 1f earlier and had the optimizer minimize distance to the star on the crucial frame.By the way, due to a recent discovery by ds273 ( youtube.com/watch?v=C38He-OUAYk ), the route for the ABC TAS is up in the air. We might end up redoing some segments of the TAS, so it will probably be several months at least before the TAS continues to progress. There's also a chance we will need to redo some parts for BitFS-related reasons as well (I realize this is vague, but I don't want to spoil too much).
[TAS] Go on a Ghost Hunt 0xA - 5473 fifdspence 2021-04-14 | brqm ( youtube.com/channel/UCTzULS5sXbp04fuGMpGlOpg ) and lemon ( youtube.com/channel/UCoZ8JbiXlq_Zh0DGvaGRTDQ ) are also authors of this TAS. brqm made a previous version of this star here ( youtube.com/watch?v=dY3m_HE-gJo ) which pioneered some strategies, like the dive recover over the gap in the room with the falling bridge. brqm's TAS also optimized all of the rooms without boos so well that I couldn't improve any of them. lemon optimized the big boo fight so well that I couldn't improve it either.Additionally, several folks (I think Iwer Sonsch originally, and also Tyler Kehne, Fraims, and probably some others) suggested the idea of skipping the door during timestop. It took many suggestions because I didn't really understand it at first, but I do now :) When a textbox is activated on JP, there's a glitch called timestop that makes it possible for Mario to spend time in certain actions, mostly air actions, without other objects being updated. Doors have walls when Mario is somewhat close to them, but not when he is far. This TAS initiates a slidekick far from the door in the painting room. The boo dies and queues up a textbox while Mario is in the air, so the door doesn't update for a while, and the walls don't exist. This allows Mario to go straight through where the door walls would normally be.The rest of the improvements were mostly due to better boo manipulation. You can see a couple of points in the run where Mario slowwalks or even brakes when entering a door to do RNG manipulation. The slowwalk doesn't cost speed. The brake does, but it's tied with the best I could do without braking, so the RNG manipulation doesn't cost any time.In the painting room, Mario dive grinds with 1/4 quarterframes along the platform to keep high speed without much displacement, since the boos go faster when Mario goes fast. Mario is able to hit the boos such that both eventually die on the same frame, which is why only one textbox shows up.
[TAS] Plunder in the Sunken Ship 0xA - 1:1878 fifdspence 2021-03-23 | Huge thanks to ds273 and to lemon, both of whom made very important contributions to this star.The downwarp and clip strategy were the brainchildren of ds273 ( youtube.com/user/Th3DekuScrub ). He investigated the usual clip on the other side of the ship (often used by RTA runners) with the 16 speed we can keep while B swimming, and didn't find a way to do that clip. He went on to investigate this clip on the other side, and found a few locations where Mario could clip in with precise positioning and angle. He sent me a list of locations where the clip can be done without turning too far to the left, and I had a swimming optimization tool aim for the best of those.lemon ( youtube.com/channel/UCoZ8JbiXlq_Zh0DGvaGRTDQ ) came up with the strategy and movement for the inside of the ship after the swimming was done. This approach takes advantage of the strange properties of shallow water to 1) Walk downhill on a slope where Mario would normally buttslide 2) Dive recover on slopes where he normally can't. lemon found all of the movement from the walking onward, and also optimized the final segment (from the top platform to the star) beyond what I was able to do.I relied heavily on the swimming optimization tools I've been using for the other JRB stars. The TAS was optimized in many segments one-by-one (start of swimming to shell, shell to over the hill, over the hill to downwarp, downwarp to clip, clip to ship entry, each chest one by one, and then the final chest to the right side of the ship). The earlier parts of the TAS look choppy because I was using a less stable approach at the time. Partway through, I started optimizing the difference between successive joystick frames instead of optimizing the joystick inputs themselves. Adding L1 regularization on this difference makes for a much smoother swim and slightly better results time-wise. I tried to improve the earlier parts with this technique, but couldn't save any frames, so that's why those still look choppy at times.
[TAS] Can the Eel Come Out to Play? 0xA - 4380 fifdspence 2021-02-17 | This star involves a lot of tricky swimming, and so I decided to make use of some automated TAS optimization tools made possible by bad_boot ( youtube.com/channel/UC280b7IZCFnzCR0RCwi6JYg ) and Krithalith ( youtube.com/user/Kiefertje ). bad_boot developed the core of what made this automation possible and wrote code to allow a Python interface to that core, and Krithalith demonstrated successful use of these tools to improve several stars (many of which are swimming stars). Thanks also to nim who explained how Unagi works to me, including the star spawn.Three times, I TASed a segment by hand, had the automated tool improve that segment, and then moved on to the next segment. These are the segments:1. First swimming frame to shortly after getting the shell. The tool was able to do this roughly 1f worth of distance ahead of what I was able to do. The clam is cycle-locked and Mario's swim punch cycle is also locked (if he wants to conserve his B speed), so the automated tool had no hope of grabbing the shell earlier, but it did grab it a bit closer to where Mario needs to be and with a good angle and angle velocity.2. The end of that to the eel activation. The tool was able to do this 1f ahead of what I was able to do. I don't quite know how that improvement worked - it is probably related to optimally interacting with floor triangles, but I'm not sure.3. Eel activation to star grab. This is the most complicated part, so I TASed it by hand several different times with slightly different approaches from above and with different directions (either doing a clockwise loop or a counterclockwise loop). I had the tool spend some time with each of these by-hand m64s I made, and it ended up several frames better than any of my attempts (which weren't especially optimized, since I hoped the tool could fix them up). The tool was able to improve almost all of the TASes I gave it to either 43"80 or 43"83.Some technical details - the tool uses simulated annealing where neighbors are perturbations of joystick angles on one or several frames. I didn't do anything particularly sophisticated or clever here, and there certainly could still be room for improvement on this star.
[TAS] Through the Jet Stream (JRB) 0xA - 5397 fifdspence 2021-02-08 | The first 0x way to collect this star, found by Plush ( youtube.com/c/PlushSM64 ) and implemented by pannenkoek2012 ( youtube.com/watch?v=0SIgn5lYgGg ), requires getting the metal cap, getting underneath the star in a precise area, waiting for the metal cap to expire, and then letting the current take Mario up into the star. BillyWar (channel: youtube.com/channel/UCBCqPFqPz2OUFOFdUZ9Bkyg ) found a clip where Mario can groundpound through the ship, and asked whether this could be used to get Mario to fall into the star.ds273 did a detailed analysis of how the clip works/what triangles were important, and was able not only to collect the star using this clip, but do so without pressing A: youtube.com/watch?v=ri_kpzmLynU Thanks to him for all of the work he spent on this! He explained to me how the clip works: "The floor passes from entirely under him to entirely over him during the GP when he cant collide with floors."This clip allows Mario to collect the star before the metal cap expires, saving time. I took note of the exact position/ship cycle frame where ds273 initiated the groundpound in his TAS and imitated that in this TAS.You might be wondering why Mario's speed reactivation took him all the way to the top of the ship in this TAS and only most of the way in the reds TAS. It's because the pillar Mario builds up speed on in the Jet Stream TAS is much larger than the one he builds up speed on in the reds TAS, and so he can get more speed. If Mario could build up speed twice on the small pillar in JRB reds, he could get more speed, but I believe there's not enough time to do that without missing the clam cycle (yes, clams opening are cycle-based in addition to distance-from-Mario-based!) and then the ship cycle.
[TAS] Red Coins on the Ship Afloat 0xA - 1:1477 fifdspence 2021-02-05 | This improvement of JRB reds is the result of an idea that was originally suggested by Tyler Kehne. Instead of framewalking all the way up the ship, it's faster to use sliding speed reactivation, provided good sliding speed is conserved with the appropriate sliding yaw. I wasn't sure whether it would be possible to conserve enough sliding speed in a good direction, still make it to the same ship cycle, and reactivate without sliding off the portion of the floor triangle that's in shallow water, but it is! (Note that the sliding speed is originally built along the sides of the pillar under the first red.) This saves a bit under 2s on the climb to the rim of the ship, which allows additional time savings later on because the red coins and the star are lower relative to the ship.Unfortunately, this and some recent testing of BBH 100+reds strongly suggest that SSL reds will be too slow to be in the run (with SSL reds+Ghost Hunt losing out to BBH 100+reds). A sufficiently optimized version of BBH 100+reds won't be made until we reach that point in the run, since it's RNG-dependent, but there's not a lot of hope for SSL reds unless it can be improved to be faster than JRB reds again or unless DDD chests is drastically slower than expected. I think SSL reds is a much cooler star, but at least JRB reds isn't too bad now with this reactivation.Also, Blast to the Stone Pillar is finished now. lemon made most of the TAS, which you can see here: youtube.com/watch?v=YPVlJHHaFBM lemon has been uploading some great 0xA TASes of upstairs stars, so be sure to check those out as well!
[TAS] Vanish Cap Under the Moat Reds+Switch 0xA - 4477 fifdspence 2020-11-15 | This TAS is only slightly faster than pannenkoek2012's excellent TAS of this course, shown here: youtube.com/watch?v=XBhDWqI4S7c . I believe pannenkoek2012's TAS reached the earliest possible elevator cycle. I did my best to optimize the beginning, but was not able to reach the previous cycle.Therefore, the only opportunity to save time was on the switch press/final red coin/star grab at the end. It's a bit faster to brake and turn into the switch misalignments instead of turning all the way around, and I was also able to save some time in the final sequence after breaking the box.You may be wondering why I walk for a few frames before slidekicking at the beginning. When Mario lands from a slidekick while he's still in the slidekicking action, he bounces, which is bad here. It's preferable to cancel the slidekick into a freefall, which lets him land on the ground instead of bouncing. One of the checks the game uses to determine whether a slidekick should be cancelled into a freefall is a check on how far above his floor Mario is. If I slidekicked a frame sooner, Mario wouldn't have been high enough above the slope for the slidekick to turn into freefall, so he would have bounced instead.
[TAS] Cavern of the Metal Cap entry + reds 0xA - 5627 (with switch) fifdspence 2020-10-28 | This TAS follows the same overall route as pannenkoek2012's TASes of CotMC entry, reds, and switch ( youtube.com/watch?v=Rql7kxQ9rOk and youtube.com/watch?v=qG7mPAQLnv0 ).The beginning of this star was kind of tricky. While doing the Eru drop off of the elevator, it's best to face as far to the right as possible to be able to land on Dorrie, and it's necessary to strain way left to get the clip, so that was a challenging tradeoff.The Dorrie part was the hardest to optimize. It's necessary to manipulate Dorrie while still getting Mario to a good spot to do the dive and dive recover onto the platform. Mario controls Dorrie via his facing angle as long as he is standing on Dorrie. The more to the left he faces, the more positive Dorrie's target yaw velocity is. So I have Mario face to the left to get Dorrie to turn left as long as I can, but not too long, since Mario needs to reach the top of Dorrie's head to do the dive and dive recover as soon as possible. Starting to run later would be slower, starting to run sooner wouldn't let Dorrie turn as far (at least without better optimization), turning later would keep Mario from getting to the appropriate spot in time, and turning sooner would make Dorrie turn back too fast (and also probably put Mario in a bad position to do the dive and dive recover from Dorrie's back onto Dorrie's neck). This was the part that required the most effort.CotMC itself was relatively straightforward. It would be difficult to fully optimize the crystal and switch part, but saving time there wouldn't actually make the star faster, since the technique to collect the star requires waiting for the metal cap to run out. So the only part that really needed to be optimized was the very end, falling off of the steep floor and getting in position to swim into the star.I should also note a TAS of this star by Super Watermelon 64 ( youtube.com/watch?v=wGFyjNA8rys ), which shows off a cool timestop, but unfortunately I don't think there's any way for timestop to help here.
[TAS] Metal Head Mario Can Move! 0xA - 1:1097 fifdspence 2020-10-22 | This TAS follows a blueprint by Iwer Sonsch ( youtube.com/watch?v=uVv3cb5h_fM ), which differs from pannenkoek2012's original 0x TAS of this star ( youtube.com/watch?v=69NMgKGFKC0 ) in a few respects:- It takes advantage of the spacing of the dive rollout to press the ! switch without the metal cap- It builds up speed using a different technique, avoiding quite a bit of swimming- It does not require Dorrie (and so is not cycle-dependent). Apologies to Dorrie's enthusiastic fan club.To understand what's happening with the speed building, you need to know some basics of speed reactivation. Mario's sliding speed is stored as a vector (so technically it's a sliding velocity, but the SM64 terminology here is entrenched). In some actions, such as swimming, sliding speed is not updated to match up with regular speed, and in some situations, regular speed can be replaced by sliding speed ("reactivated"). In this TAS, Mario starts with some speed along a slope, stores it as sliding speed while swimming, and then reactivates it (with a small tax) along a different slope by pressing z and c^. Mario gets forward speed if his facing angle during reactivation is within 90 degrees of the sliding speed vector, and he gets backward speed otherwise. During the first frame of the reactivation, the sliding vector gets added to Mario's position, and then the sliding vector gets updated to match with Mario's facing angle.That tells you enough to get a general idea of what happens: Mario stores some speed, reactivates it along a new slope that only differs by 45 degrees, and repeatedly does this until he gets enough speed.However, there are some complications. If Mario is facing downhill, he will gain speed, but if he spends a frame facing uphill, he will lose speed that frame, which is important to avoid. Therefore, it's necessary to be really conscious of where the speed is reactivated. Because the first frame of movement has the sliding vector added before its angle gets updated, Mario gets displaced quite a bit along the slope he was previously sliding on. So Mario must aim for a sweet spot when he reactivates the speed: not too far to the left, or he'll just splash into the water on the first-frame displacement, but not too far to the right, or the displacement will have him stay on the old slope and spend a frame facing uphill.Our control over spacing is very limited (especially with the awkwardness of swimming), so it's hard to enter this sweet spot. Quarterframes don't apply on ground-to-water movement, so Mario can only enter the water once every x units or so, where x is his speed. The higher Mario's speed is, the less likely we'll happen to get lucky and be within the sweet spot (which has a constant width). We have a little bit of control based on how we move the joystick when Mario exits the slope, but without doing some additional swimming, it's often impossible to hit the sweet spot.Therefore, I employed two strategies:1. Sometimes, I would get off of the slope on the early side, have Mario face upward, and press B to swim for a bit. The more Mario faces up, the less of his speed actually gets converted to horizontal displacement, so this lets us manipulate Mario's position very well. This is fairly quick to do.2. For the last two speed-building segments, I went all the way to the end of the slope instead of getting off partway along it. This requires Mario to swim in a loop to do the next reactivation, so it costs more time, but it does let him build up much more speed on the slope.What I've done is definitely not optimal, but I'm convinced that it's not possible to always get perfect spacing without employing these techniques, and I do believe it's best to use method 1 early on and use method 2 at the very end.This TAS does differ from Iwer's blueprint in that it clips through the elevator as it moves up rather than down. This is done via an Eru drop, a technique discovered by Eru-san to walk backward off of ledges ( youtube.com/watch?v=Tl1qDika4Bk ).The first room of this TAS is from Iwer Sonsch's Swimming Beast TAS ( youtube.com/watch?v=GXTHhWFzbr0 ).By the way, check out jongyon's phenomenal work on Navigating the Toxic Maze ( youtube.com/watch?v=0EJIsstnRnw )! That star is also fast enough to be in the run.
[TAS] Swimming Beast in the Cavern 0xA - 1743 fifdspence 2020-10-02 | The first room of this TAS is by Iwer Sonsch from his previous WR on this star: youtube.com/watch?v=GXTHhWFzbr0Overall, Iwer's existing TAS of this star was excellent. It had what I think is most likely a fully optimal first room. Next, it did a slidekick with the perfect angle: the wall next to the pole will cause Mario to bonk if he is facing any more counterclockwise than this, and the more counterclockwise he is, the better. Iwer's TAS did some extreme straining to avoid hugging the hallway wall after landing from the rollout, and went on to do the very cool elevator clip that lets Mario reach the star.This new TAS is a 6f improvement over the previous WR. The improvement is by making two dust frames after the big fall instead of straining hard to avoid the wallhug. (For those who don't know, if Mario is close enough to a wall and facing too much into the wall, he will wallhug and instantly drop to a measly 6 speed.) ds273 applied this trick in their excellent segment of the of Stand Tall on the Four Pillars 0xA WR: youtube.com/watch?v=Lh0HKRw1p5E The two dust frames allow Mario to change his angle enough to keep running after landing instead of wallhugging, even though he is really close to the wall. (One dust frame can't change Mario's angle enough.) This means instead of doing extreme straining, Mario can continue to accelerate, moving farther along the wall and keeping more speed.Using this trick modifies Mario's 0xA cycle of diving and dive recovering, which is really the only option for staying close to 48 speed in the absence of edges to grind on or slopes to slide on. Fortunately, the elevator is tall enough to cancel a dive recover on, so the extra distance we get using the dust trick does translate to a save by landing on the elevator significantly sooner.I should probably mention how the elevator clip happens. When Mario is diving underneath the elevator on most of the frames, he gets pushed down by the ceiling on the bottom of the elevator. When he reaches the edge of the elevator, though, he gets a sideways push from the elevator's wall before he gets pushed down by its ceiling. This wall push puts him far enough that the cavern's wall doesn't push him back inward, so he can continue on his diving path in the room with the star. (If you wonder why the ceiling doesn't push him too far down to experience the wall check, note that the ceiling check applies after the wall check. Take a look at pannenkoek2012's epic Walls, Floors, and Ceilings series if you want to know more detail, starting here: youtube.com/watch?v=UnU7DJXiMAQ ) As far as I know, Iwer was the first person to demonstrate that this clip is possible 0xA with the elevator moving up instead of having to wait for it to come back down.
[TAS] Footrace with Koopa the Quick 0xA - 1:2100 fifdspence 2020-10-01 | This is a 33f improvement over a previous TAS by me and IsaacA: youtube.com/watch?v=w2P42t7ONnk This improvement would not have been possible without an automated script to try out a bunch of different possibilities. The script was written by bad_boot ( youtube.com/channel/UC280b7IZCFnzCR0RCwi6JYg ) and Krithalith ( youtube.com/user/Kiefertje ), and modified slightly by me for 0xA. bad_boot provided the core and Krithalith is very skilled at using the core to automate improvements in various SM64 situations, and provided me with his script for non-0xA Koopa the Quick.Koopa is influenced by a number of different factors:- His starting angle- When Mario finishes the race- When Mario spends time in his radius after finishing the race- Bowling balls he encounters on his climbTesting different possibilities for these can be extremely time-consuming. In the previous run at this star, I did a very naive approach to choose a starting angle and then finished the race as soon as possible. After that, Isaac took over and selected bowling ball spawns and implemented them. Isaac did a phenomenal (possibly fully optimal, though I've not checked this) job of this.The script is able to consider all of the above factors at once. After evaluating many tens of thousands of possibilities, it found a solution which was 20f faster than most nearby angles if I could beat the race 1f faster than last time. I was able to do so. After beating the race, I implemented the script's save in the game by spawning the bowling balls when it said to and entering Koopa's radius for exactly one frame when the script said to.The bowling ball spawns are quite important. It's necessary to avoid and cause some spawns while Mario is still doing the race, both from the lower spawner and the upper spawner. The most powerful tool we have for manipulating RNG is when accelerating from low speeds - Mario can reach the same speed with either high or medium joystick magnitudes, but the high magnitudes generate dust (which affects RNG) and the medium magnitudes don't. In this TAS, I was able to affect some of the early spawns during the painting reentry. The spawns partway through the race were manipulated during the framewalk sections. Manipulating the spawns after the race was easier, since Mario doesn't have to hurry anywhere and we can make dust however we want.
[TAS] Bob-omb Battlefield 100 coins + Reds 0xA - 2:0110 fifdspence 2020-09-24 | After seeing Plush's BOB 100+Koopa the Quick TAS at youtube.com/watch?v=D8nJppKGu9g (in a non-ABC setting), I wanted to try to improve this star, since some ideas he used apply for 0xA as well. This TAS is an improvement of a little over 5 seconds compared to my last attempt at this star: youtube.com/watch?v=fCjxssrHUQ4 . The first 1:27 is the same, so you can skip to there if you have already seen the previous version.I'd like to note three ideas that went into this TAS that made it faster than the previous ones:- Using a dive recover from the sign to get between posts, which I realized was a possibility after seeing Plush do it. I do this from post 1 to post 2. Because of this and because of goomba manipulation, I only ever need to enter one post misalignment, which is good because the misalignments aren't always on corners which are fast to access.- Using an Eru drop from the corner of a post to dive, slide until Mario stops, then recover once his speed reaches 0 so that I can do an Eru post. I learned about this possibility when Plush did it on post 4 of the 100+koopa TAS. This uses two techniques found by Eru-san ( youtube.com/channel/UC6ua8wn5GgxDNINh3xcStUA ) which are not related to each other, but which work together here. An Eru drop is a maneuver that takes advantage of the fact that Mario walks with alternating positive and negative speed when the joystick is at a very small magnitude. This can allow Mario to walk backward off of a platform, in this case allowing him to walk backward off of a post and then dive into it. An Eru post is a maneuver which uses air movement to rapidly go around the center of a post from above, doing almost a half-turn each frame instead of having to run around the entire post. This beats what I did previously for these posts, doing a bunch of quickturns around the center.- Using the alternating positive and negative speed while turning to go around the center of a post without having to enter air movement. This is used on post 2 here. The idea was originally conceived by Iwer Sonsch ( youtube.com/channel/UCnJKlalNQWIco1DS0Uw_XXw ). During these maneuvers, Mario's speed trends toward becoming positive over time rather than oscillating between positive and negative values of the same magnitude, so I wasn't sure whether it would work. Fortunately, jongyon ( youtube.com/channel/UC6dj_9jki7Zbuo29CT7i7gA ) demonstrated an initial magnitude that led to a sequence of speeds that works, so this trick can in fact be done. Here are images of Mario's path, where the squares are 1 in-game unit (1 centimeter) on each side. cdn.discordapp.com/attachments/610551754064396290/758548869444665374/unknown.png cdn.discordapp.com/attachments/610551754064396290/758548955964768296/unknown.pngThere is certainly some more timesave available here with better RNG manipulation, but it's extremely challenging to perform, and it wouldn't necessarily be feasible to include a timesave in the work-in-progress TAS (for RNG and global timer reasons). Fortunately, this current version is now included in the WIP 70ABC TAS, along with an updated Koopa the Quick that I'll be encoding sometime soon. The resyncing process was rough, but it's finished now, so it's time to leave SSL and move on to castle stuff and HMC!
[TAS] Free Flying for 8 Red Coins 0xA - 1:1647 fifdspence 2020-09-12 | This star is a borderline star to be fast enough for the in-progress 70 star ABC TAS. Personally, I think this is fast enough, but you never know what will happen! Thanks to many users for their considerable contributions to this TAS:-Ystem ( youtube.com/channel/UC_d0m6f9Sx5PHwKxYUR2REA ) TASed the manipulation of the initial fly guy, which was extremely complicated and required RNG manipulation and deep understanding of the fly guy code. They saved several seconds over my initial attempt at the manipulation.-pannenkoek2012 came up with the route for this star, including the cloning at the end, and demonstrated it in youtube.com/watch?v=53yVvpNQHQg . This star wouldn't have even been worth considering for the 70 star ABC TAS without him finding this route.-Tyler Kehne ( youtube.com/channel/UClPjENu5gaTOaZ5_99AQL0g ) came up with the idea of using the vertical speed from the crazy box to get an extremely high tornado twirl, making it to the upper level of the pyramid without having to framewalk.-jongyon ( youtube.com/user/jongyon7192p ) came up with the idea of landing from the second fly guy bounce early to build up speed on the slope before collecting the oasis red.You may wonder why the crazy box allows Mario to twirl so high from the tornado. Mario's vertical speed is conserved while he's in it, so he keeps 96 vertical speed (which is really large) throughout his time in the tornado and then keeps positive vertical speed for a long time. This turns out to be really useful, since Mario can land on the pyramid, keep some speed into the fly guy bounce, and use this to land somewhat far up the sand slopes. My first attempt at this star didn't do the huge twirl or the sand slope landing, but they both turn out to save time!There were some other ideas that didn't end up in the TAS... at least not yet.-This TAS lands from the tornado twirl before bouncing on the fly guy. What if we could bounce on the fly guy without landing? Normally, if you twirl into a fly guy, you just kill it, but Tyler Kehne had the idea of getting burned by its flame while in midair and bouncing off of it while in the burning fall action. This turns out to be barely possible (and is pretty cool to see), but I couldn't get very close to collecting the red coin during the twirl. Tyler is investigating some ideas to try to make it far enough to get the red, so maybe this star will be improved!-jongyon's idea of landing on the sand slope opens up some other possibilities. For example, what about buttsliding into the oasis, swimming to the sand, and conserving speed with z+c^ and a punch? This is possible, but turns out to be a little bit slower. Maybe instead of quickstopping out of the backward DR, we can backward walk to the walkway the tox boxes are on, fall off of the edge to cancel the backward walk, and grind up to normal speed facing forward? Unfortunately, a tox box is in a really bad point in its cycle for this strat, and I think it would be a bit slower anyway. Regardless, I think the strat in the video is nice as well, and cooler to see than just landing next to the red and running (not to mention faster!)
[TAS] Red Coins on the Ship Afloat 0xA - 1:1977 fifdspence 2020-07-25 | The 70ABC TAS has reached a crossroads where it's necessary to figure out which stars will be in, since SSL reds is a borderline star and has similar times to several other stars. One of these is JRB reds, and since there's no RNG or global timer dependence here, I decided to go ahead and TAS it.This TAS follows an overall similar route to the previous WR by pannenkoek2012 ( youtube.com/watch?v=PchLf_D1vks ). It saves over 10 seconds, most of which is by making an earlier ship cycle and some of which is by improving the ending after getting on the ship.Each ship cycle is 8.53 seconds long. This TAS reaches the earlier one with some small routing changes at the beginning. While swimming 0xA, Mario can only get up to 7 speed from rest, but he can maintain up to 16 speed. So much of 0xA swimming is trying to avoid spending too much time far below 16 speed. Two techniques of getting close to 16 speed are used here:- Getting the shell, which brings Mario to 28 speed, of which he can keep 16 after she shell is gone- Getting speed from the pillar the first red is on, which was suggested by Tyler Kehne. When Mario has a lot of speed, he can keep a fraction of his speed in the water (as long as he doesn't spend too much time losing speed in the plunging animation).Because we get 16 hspd right at the beginning, it's not crucial to get the shell right away as pannenkoek2012 does, so this TAS instead gets one of the clam reds before grabbing the shell. With this route, Mario is able to reach the ship in time for the previous cycle. The swimming isn't especially optimized, but it shouldn't be possible to make the previous cycle, so it doesn't matter.It was possible to get significant savings on the ship itself by collecting the sixth red a different way. Instead of waiting for that side of the ship to rise back up, I'm able to slidekick to the star before that side of the ship falls too far. I found this quite challenging to do. I tried to do the ship clip on the earliest frame I could and as far right as I could. Then, I framewalked at a specific angle that would allow me to turn somewhat before reaching the top and still get on top on my desired frame. This enabled me to get far to the right (higher up along the railing and closer to the red) and turn to slidekick with just enough height and distance to collect the star.I'll probably revisit the end if this ends up being fast enough to include. Dive grind start after the sixth red is probably faster than slidekick start, assuming it's still possible to climb to the top part of the ship earlier.
[TAS] In the Talons of the Big Bird 0xA - 1700 fifdspence 2020-06-11 | This star uses a different route from the previous WR by Ethan White ( youtube.com/watch?v=GQcSAzlGE9Y ). The new route was found by CeeSZee (channel: youtube.com/channel/UCrmbllPtODE3dQ_aFLGr5sg ), who tested the route, came up with the movement to the pillar, and also TASed the quite precise twirling from 0:08 to 0:14.Iwer Sonsch (channel: youtube.com/channel/UCnJKlalNQWIco1DS0Uw_XXw ) found the slidekick strategy at the end to reach the star, which turns out to be very precise as well. He made a TAS which gets within 1.12 units of the star, and then I wrote a brute forcer to improve his air movement section to squeeze out the final bit of distance. I ended up needing to use a slightly different angle than Iwer used to reach the star. Some technical details about this brute forcer are discussed in a pastebin linked at the end of this description. I believe Iwer also TASed the divegrind here, though I can't remember for sure whose divegrind made it into this version.Thanks very much to Ystem, who did a lot of work helping me understand how to manipulate the fly guy. It was necessary to have the fly guy lunge, and activating the lunge without wasting too much time requires a few conditions that I certainly wouldn't have understood properly without his help. Thanks also to others who contributed ideas while I streamed the TASing of this star, like jongyon, IsaacA, and galoomba.An interesting thing about this star is that Klepto is essentially cycle-based, and that him being earlier in this cycle makes the star more difficult in two respects:1. The earlier Klepto is in his cycle, the more difficult it is to activate his star and still make it to the corner of the pyramid platform at 0:14, since he's more out of the way of Mario's ideal twirling path. On the frame shown in the video, I wasn't able to make it, but fortunately CeeSZee optimized it and made it to the corner. If I had reached the twirling part another frame or two earlier, it may not have been possible.2. The star actually spawns further away from the pyramid the earlier Klepto is in his cycle. This is because of an off-by-one error in the code that makes the star overshoot its intended position by a frame whenever there's a cutscene that causes the star to move. The earlier Klepto is in this cycle, the farther the star moves in this extra frame, and the more difficult it is to reach the star using the slidekick bounced from the pyramid. If I had activated the star another frame or two earlier, it may not have been possible.I'll put some technical details about the brute forcer in this pastebin, for anyone who is interested. pastebin.com/CYDBmjFcNOTE: I improved this star by 1f and uploaded it unlisted: youtu.be/3QXpbCHoQL0
[TAS] Shining Atop the Pyramid 0xA - 1070 fifdspence 2020-06-05 | This is a 1f improvement from the previous TAS by Doctor Dab ( youtube.com/watch?v=Vrp9uXpkeV4 ). The TASes are tied until the clip onto the pyramid entrance and the beginning part of the subsequent climb. By turning aggressively left during part of the framewalking, I was able to reach slidekicking speed on the upper platform 1f sooner but with a worse angle. Fortunately, it was still possible to get the star dance clip with this angle. Specifically, while turning left, I tried to make Mario's angle as left as possible while not triggering the 24 unit wall check. I turned right for a little bit at the end to still get a reasonable angle for the slidekick.It's actually possible to reach slidekicking speed an additional few frames sooner by turning all the way around on the pyramid (so that Mario is walking downhill and gains speed on the frame where he clips onto the flat floor), but the angle is too bad for it to be worthwhile. The alternatives in that situation, like walking straight to the star or doing the clip on the left side of the star, were each 1-2f slower than what is shown in the video.
[TAS] 8-Coin Puzzle with 15 Pieces 0xA - 1430 fifdspence 2020-06-02 | This TAS is pretty much the same as the previous TAS by Doctor Dab ( youtube.com/watch?v=88E7aP-oNF4 ). However, it slidekicks one frame sooner at the beginning, if I've counted frames in the video correctly. The rest should be tied.Next up is SSL!
[TAS] Hot-Foot-It Into the Volcano 0xA - 2363 fifdspence 2020-05-31 | The part of this TAS before Mario enters the volcano is by Iwer Sonsch.This star has changed significantly in recent months due to pannenkoek2012 and Iwer Sonsch. pannenkoek2012 developed a new strategy, which he demonstrated in youtube.com/watch?v=gNJGCAw8L5Y . The strategy takes advantage of out-of-bounds unit squares, which allow Mario to reflect his direction in midair during lava boosting. This enables Mario to reach the Hot-Foot-It area with lavafall boosts that begin much lower than was previously known possible. Iwer Sonsch used this approach to make the following TAS of this star: youtube.com/watch?v=SN7wfMKdKCs In doing so, he showed that it is possible to dive recover to the rock at the base of the lavafall, which allows Mario to reach the bridge without using the elevator.I branched off of Iwer's TAS on volcano entry. There are two main sources of timesave compared to his TAS:1. A different combination of dives and dive recovers at the beginning2. A dive grind along the bridgeThe bridge dive grind may be improvable. When I stopped at the end of the dive grind, I was 7 units away from the final coin, and I couldn't find a way to reduce this. (Note that Mario needs every bridge coin to avoid dying from the lavafall.) Therefore, I used a slower variant of the standard quickturn. In a standard quickturn, Mario walks 1.x units, then .x units, then stops, and then turns around. I had Mario walk 3.x units, then 2.x, then 1.x, then .x, then stop, then turn around. So given a few more units, one could save a frame, and given a few more units after that, one could save a second frame. (I can't simply prolong the dive grind for one more frame to save two, since the dive grinding is ended by a turn in the edge along which Mario grinds. Additionally, the direction in which Mario moves is very different from the direction from Mario to the coin after he comes to a stop, so that also makes it more difficult to save a frame here.)Next up is LLL reds. I may not be able to improve the existing WR ( youtube.com/watch?v=88E7aP-oNF4 ) by Doctor Dab. If not, this will be the first star in 70ABC to go unimproved in the current run.
[TAS] Hot-Foot-It Into the Volcano + 100 Coins 0xA - 1:2721 fifdspence 2020-05-28 | his is a test TAS made to determine whether LLL 100 should be paired with Bully the Bullies or Hot-Foot-It Into the Volcano. As I neared the end of this TAS, I realized that this pairing would be slower, so the ending is not very clean. (When adding up times to figure out which pairing was faster, I included the 100 coin star spawn for 100+Hot-Foot-It, but forgot its star dance celebration, so it's nowhere near as close as I thought.) I still wanted to post this video for reference purposes, though. Thanks to kierio04 for testing some alternative route pairings for LLL 100.Thanks to pannenkoek2012, who came up with and demonstrated a way to do Hot-Foot-It 0xA much more quickly than was previously known possible: youtube.com/watch?v=gNJGCAw8L5Y Thanks also to Iwer Sonsch, whose TAS of Hot-Foot-It using pannenkoek2012's idea was a valuable reference as I made this TAS. Iwer's TAS shows the faster way of doing the end of this star, lava bouncing onto the blocks near the star and using them to initiate an air dive.
[TAS] Bully the Bullies and 100 Coins 0xA - 1:2653 fifdspence 2020-05-22 | Next up in the 70ABC TAS is LLL 100. At the moment, it's uncertain whether 100 will be combined with Bully the Bullies or Hot-Foot-It Into the Volcano. The TASes so far suggest that it will be very close. This is the first one that I consider finished by my standards.Thanks to Kierio04 for doing some enormously helpful test TASes of other routes for this pairing. Thanks also to several folks for their ideas and help, including jongyon, pannenkoek2012, Iwer Sonsch, galoomba, Lord SNEK, Ystem, and others. (My apologies to anyone I've forgotten, as I did some LLL 100 work on stream a month ago, and I tend to lose track of who said what while streaming.)If all goes according to plan, I want to upload a 100+Hot Foot It video and a solo Bully the Bullies video, and then hopefully a comparison video if my anemic editing skills are up to the task. The recent TAS of solo Hot-Foot-It by Iwer Sonsch ( youtube.com/watch?v=SN7wfMKdKCs ) using an idea conceived of and demonstrated by pannenkoek2012 ( youtube.com/watch?v=gNJGCAw8L5Y ) is excellent and more than sufficient to get a good time estimate, so I don't plan to try to improve it before making the comparison.
[TAS] Red-Hot Log Rolling 0xA - 1780 fifdspence 2020-04-15 | The first part of this TAS (until Mario reaches the top of the stairs around 0:06 ) is hexed from a TAS by Iwer Sonsch (channel: youtube.com/channel/UCnJKlalNQWIco1DS0Uw_XXw ).The previous record by Iwer at youtube.com/watch?v=SlRw8r22c4A uses a route in which Mario uses two lava boosts to reach the platform just after the log and then navigates to the star via land. I tried a new route, and it turns out to be 10f faster than my best attempt at the old route.The maneuver on the log was quite tricky. It takes advantage of an exploit that allows Mario to maintain running speeds up to 48 on certain floor types (whereas typically Mario's running speed would decrease by 1 each frame until it reaches ~32). This applies on the log, but only for floor triangles which are sufficiently sloped. I use it to initiate the slidekick with 48 speed.To reach 48 speed, I use dives and dive recovers which are quickly cancelled on the log. I can't end the final dive recover on the top part of the log, since it's not sloped enough yet for Mario to maintain his speed. This is why I run back down the log before turning to dive and dive recover up it again. It's nice to be walking on the log anyway, since it helps the log roll in the direction of the star, and losing distance to de facto speed is actually helpful (more on this later).The slidekick to the corner where I cancel the lava boost required some precision with how I handled the log. The space to turn to a good angle is very limited, since Mario can't walk on areas which are too flat without losing speed. I take advantage of all of that space (you can see how close I am to a flatter part of the log at 0:16 before turning) and still don't have an angle which would allow me to move straight west. Therefore, it's beneficial to be fairly north on the log.Additionally, there's a strange behavior on the lava bounce which I don't fully understand. From the height I slidekicked from, bouncing in the middle of nowhere gave a y speed of 21 to start the bounce, but bouncing near the tiny triangular prism (at 0:17) gave a y speed of 22. My guess is that some part of the triangular prism which is submerged in lava is somehow used in a relevant calculation, but I'm not sure. Anyway, that extra y speed gives Mario more time to reach the corner for the lava cancel, and he is able to make it. This is another reason it's beneficial to be north on the log: I didn't have to strain much at all to get that extra y speed. (EDIT: galoomba pointed out that this is because of a phenomenon known as floor overshadowing or by a different, ruder name: ukikipedia.net/wiki/Floor_Cucking The lower part of the pyramid is used for collision rather than the lava triangle, and Mario is able to fall for an extra frame and get a bigger bounce afterward. galoomba's channel: youtube.com/channel/UChh6RNVE2N7Ibob0wsCe_VQ Also, JoshDuMan notes that since those triangular prisms only have collision on JP, this is a JP-exclusive strat unless someone can optimize it enough to do it on US, which may not be possible without rolling the log much more. JoshDuMan's channel: youtube.com/user/joshduman )I'm happy with how this star turned out! Now it's probably time to figure out the best pairing for the 100 coin star.
[TAS] Boil the Big Bully 0xA - 2073 fifdspence 2020-03-28 | After I uploaded a previous version of this star ( youtube.com/watch?v=ISqwgPLbQnA ), pannenkoek2012 suggested that I try knocking the bully off of the platform during lava boost. Previously, I had attempted to do this, but only succeeded in knocking the bully toward the center of the platform, as Mario was in front of the bully, which I incorrectly believed to affect the bully's trajectory.As was pointed out to me by nim and Iwer Sonsch, Mario should knock the bully in the direction he is facing regardless of whether he is in front of the bully or not. I tried to figure out how this differed from my observations, and eventually realized that I was hitting the bully twice in my previous attempts: once toward the edge, causing Mario's direction to reverse and hitting the bully toward the edge, and then again back toward the center, causing Mario's direction to reverse again. To avoid this, Mario needs to hit the bully from around the edge of the bully, hitting it toward the edge and letting Mario get redirected toward the center parallel to the bully (so he doesn't collide with the bully again).While optimizing, I decided that Mario should hit the bully on the first frame possible that still allows him to hit it toward the edge. This gives Mario more time to build up speed in the air before landing (ground movement has slower acceleration than air acceleration), which is useful for diving sooner and/or getting closer to the star. Tyler Kehne explained to me that Mario can only collide with the bully in lava boost while Mario is moving down, which allowed me to determine when I was hitting the bully on the first frame possible for the lava boost.There are a number of strategies at the end that are very close to each other: the strategy in this video, the strategy I used in the previous video, dive grinding along the right edge, and just diving off of the back edge. I'm getting that the strategy here is about 1-3f faster than all of the other strategies.This is a 19f improvement from my previous video. It still includes the textbox in the timing.
[TAS] Boil the Big Bully 0xA - 2137 fifdspence 2020-03-26 | The slidekick and subsequent divegrind at the beginning of this TAS are by Iwer Sonsch. The rest is by me.Overall, the route for this star is very similar to the previous record by Iwer: youtube.com/watch?v=Lj5PO2XNiAA There are two movement differences I'd like to note:1. The right side start allows Mario to turn left to cut the corner and reaches the top of the stairs a few frames sooner than a left side start.2. When I slidekick after killing the bully, I strain left to get closer laterally to the star and to slow down enough to slide off and dive. Gravity is stronger when Mario is in a dive state than when he is in a slidekick bounce state, so he falls more quickly this way, and ends up lava boosting on the same frame as he would if I hadn't entered a sliding state. He makes it closer to the star due to the speed from the dive, which turns out to be enough to reach the star after bouncing only once on the upper platform. Without this strategy, I would have had to either bounce twice or spend frames to initiate the slidekick closer to the star (either by running more or using a slower setup).(Note: The time for this star includes the course entry textbox.)
[TAS] [Outdated] Tower of the Wing Cap with Switch Press 0xA [JP] - 1963 fifdspence 2020-03-21 | This star has been improved by Alexpalix: youtu.be/Owytukf5Vi0Note: This TAS is on JP rather than US, and the JP textbox is 5f faster. Therefore, this TAS is only 1f faster than Plush's US TAS here youtube.com/watch?v=c2bd6Dunsls rather than 6f as the times would make it seem.In a way, Plush is an author of this TAS as well, because everything up to the eighth red coin was the output of a computer program I wrote. The program took Plush's TAS as an input and tried to make improvements to it, and in the end, it was enough units to save a frame. I redid the ending from the last red coin to the switch press and the star, since it was hard for the program to deal with the switch press.We're still trying to verify whether it will be best to press the switch. Not pressing it would save either 1f or 2f, but would likely add some lag. Regardless, I thought I should upload this version, since it would be useful for this time to be documented for future non-ABC projects that require the wing cap.
[TAS] Princesss Secret Slide Under 21 Seconds 0xA - 2480 fifdspence 2020-03-10 | The previous world record for this star was an RTA run (!) by bad_boot: youtube.com/watch?v=JGZl_isHh9Q Instead of using that strategy, which requires going down the slide backward and then doing three bounces, it is possible to use only two bounces, obviating the need to turn around.We tried to figure out a way to use this skip in combination with variations on an older skip, TASed by Gaehne D here: youtube.com/watch?v=pgo_1rrX3nw That skip is not compatible with this one. Some variations using the arch-like structures above the slide to slidekick to the next part seem close to working, but I couldn't get two frames on those structures to initiate the slidekick; I always skidded off after one frame.By the way, you might wonder what's going on with BitDW. MKDasher and I collaborated on TASing BitDW, and the plan is for him to upload it once we're both happy with it. Keep an eye on his channel for the BitDW TAS! youtube.com/user/mkdasher
[TAS] Blast Away the Wall 0xA - 1420 fifdspence 2020-02-27 | With this, WF is done! The stage was full of RNG manipulation, Whomp manipulation, and misalignments, so I'm looking forward to possibly some less technical TASing. However, BitDW is next and has a dive recover into a misalignment that has only ever been done once, so the technical aspects will stick around for part of the next star.Blast Away the Wall was the final ABC WR held by the 20 star no A button TAS (with the exception of 100 coin stars which are paired with suboptimal stars): youtu.be/A5eFLrG7vIQ?t=1073 Though the stars have all been improved on now, I want to emphasize how impressive the achievement of the 20 star authors was. Without modern tools (such as angle matching LUA and STROOP) or modern knowledge (such as misalignments and the information in decomp), they managed to create a fast, cool-looking, interesting 0xA TAS with lots of creative strats. It was well ahead of its time as a combination of speed-TASing and ABC TASing, and I think it deserves appreciation, even as we move onward to faster strats and more stars.
[TAS] Fall Onto the Caged Island 0xA - 3233 fifdspence 2020-02-26 | This TAS uses a strategy demonstrated in the 20 star no A button TAS: youtube.com/watch?v=A5eFLrG7vIQ&t=20m34s It has some optimizations, however. First, we can now make an earlier elevator cycle using the misalignments on the stairs and framewalking, as seen in e.g. the WF red coins TAS.Also, less time is spent luring King Whomp into place. The 20 star TAS takes advantage of the fact that when the King squashes Mario, he gets back up. The TAS allows Mario to get squished near the edge, and when the King gets back up, he continues to walk toward the edge for a bit before trying to squash Mario again. This lets the TAS lure the King very close to the edge.However, it takes some time to do this setup. This new TAS demonstrates that the extra squish isn't necessary; the King can be lured pretty directly into a suitable position. It turns out to be nice to lure him straight toward the edge and then have him turn toward Mario. This puts the King in a better position, closer to the edge and also set up so that Mario will be lined up well with the cage.Then it's a matter of manipulating the King's hitbox so that Mario can dive recover from the very corner of the King with as much speed as possible (a bit over 48 due to the way speed and dive grinds work). The King's hitbox behaves much more simply than the King's visual model: it doesn't move up and down at all, it just rotates left or right depending on whether Mario is to his left or right. In this video, I wait a bit on top to let him rotate left enough so that I can start running along his edge, begin divegrinding, let the dive grind sliding frames modify my angle to face me directly toward the star, and get my position close to the corner when I do the final dive recover. (The King starts turning right as I run, which is what allows the dive grinding to work.) This maneuver was delicate, especially since I tried to do it from as far as possible (to reduce the amount of time needed to lure the King).
![[TAS] Scale the Mountain 0xA - 2377](https://i.ytimg.com/vi/8fHgvh3IKig/maxresdefault.jpg)