WEBVTT Kind: captions Language: en 00:00:06.012 --> 00:00:13.920 I have spent 3 hours, 14 minutes, and 44 seconds  covering the world record history of a 3 minute   00:00:13.920 --> 00:00:22.040 long flash game made by a concrete mixer salesman  back in 2009. That’s how long it took to cover 41   00:00:22.040 --> 00:00:28.960 records set by 13 different runners over the  span of 12 years. In the previous part, Red   00:00:28.960 --> 00:00:36.120 Ball Speedruns - Beyond the 3 Minute Milestone, we  explored how the time went from a 2:57 by Caramel,   00:00:36.120 --> 00:00:44.400 the first sub 3, to a 2:48 by Yak27, the first  sub 2:50. This was made possible thanks to a few   00:00:44.400 --> 00:00:52.240 frame dependent strategies that revolutionized  the game, namely Bird 3, Max 6, Resetless 10,   00:00:52.240 --> 00:01:00.000 and PBird 11. However, these discoveries, enabled  by a tool-assisted speedrun mod created in 2021,   00:01:00.000 --> 00:01:05.640 were widely believed to be almost all that the  game had to offer in terms of viable real-time   00:01:05.640 --> 00:01:13.000 setups. Sure, the TAS of the game was a sub 2:30,  and the segmented sum of best wasn’t too far off,   00:01:13.000 --> 00:01:18.680 but the stuff that people could pull off in  actual runs wasn’t even close to matching   00:01:18.680 --> 00:01:24.200 those individual level accomplishments. However,  if you glanced at the length of this video before   00:01:24.200 --> 00:01:29.920 clicking, then you know that we have a lot to  discuss. For Red Ball’s time to be pushed lower,   00:01:30.440 --> 00:01:34.720 more and more of these setups needed to  be found and implemented in each level,   00:01:34.720 --> 00:01:39.880 replacing variable movement with binary  outcomes. This quest for perfection was   00:01:39.880 --> 00:01:46.360 the next chapter in Red Ball’s story, and  indeed, just a month after the first sub 2:50,   00:01:46.360 --> 00:01:52.775 there was already a massive discovery that blew  the game’s ceiling wide open. Let’s take a look.  00:02:07.960 --> 00:02:13.720 6 frame perfect inputs for around 2.5  seconds of time save. Four years earlier,   00:02:13.720 --> 00:02:18.800 UltraAug discovered that a frame perfect  bounce pad jump could get you to the top   00:02:18.800 --> 00:02:25.320 of the level earlier, and world record contenders  had been using the same route ever since. However,   00:02:25.320 --> 00:02:31.880 by thinking outside of the box, NotVeryGood was  able to combine the top and bottom routes through   00:02:31.880 --> 00:02:37.960 the level into something much greater, and it  became known as Mug 7. A version of this hybrid   00:02:37.960 --> 00:02:44.480 route had already made its way into the TAS,  but it required dozens of frame perfects. Now,   00:02:44.480 --> 00:02:50.320 it had a good chance of being incorporated into  runs, especially after BirdBanana reduced the   00:02:50.320 --> 00:02:56.360 frame perfect count to 5. Former world record  holder NorXor even managed to hit it twice in   00:02:56.360 --> 00:03:02.600 a row during practice later that month. However,  with the visual cues for most inputs being quite   00:03:02.600 --> 00:03:09.320 awkward and easier time saves being present  in the run, it wasn’t worth going for… yet.  00:03:09.320 --> 00:03:14.760 As is typical whenever a large barrier is  shattered, the hype around top level competition   00:03:14.760 --> 00:03:22.200 died down for a while after Yak’s 2:48. It wasn’t  until May of 2022 that we would see another world   00:03:22.200 --> 00:03:30.000 record set, that being a 2:47.935 by Big_Mukmuk.  He missed the corner boost setup in level 4   00:03:30.000 --> 00:03:35.520 to save a cycle on the death warp, but he still  had the potential to break the record by a wide   00:03:35.520 --> 00:03:41.800 margin going into level 11 thanks to his excellent  movement throughout the run. Unfortunately,   00:03:41.800 --> 00:03:48.360 he barely messed up the train strategy that he  was going for. Back in June of 2021, BirdBanana   00:03:48.360 --> 00:03:54.280 had discovered a setup to complete the train  level in under 16 seconds, collecting the flag   00:03:54.280 --> 00:03:59.920 before it went under the ground triangle. This  required just a handful of frame perfects, but the   00:03:59.920 --> 00:04:06.360 unclear visual cues made it very difficult to pull  off. Pearce and InsaneJetman later evolved the   00:04:06.360 --> 00:04:12.600 setup into a way to back up the traditional PBird  strategy if you mistimed your first right press,   00:04:12.600 --> 00:04:19.000 coining it PJet. Even though this variant required  an extra frame perfect, the more distinct visual   00:04:19.000 --> 00:04:25.600 cues for each input made it a lot easier, causing  top runners to consider incorporating it. Mukmuk   00:04:25.600 --> 00:04:30.760 nailed the beginning of the setup to get past  the large spikes, but he unfortunately mistimed   00:04:30.760 --> 00:04:36.920 this jump off the black platform, causing him to  miss the corner boost. This tiny mistake cost him   00:04:36.920 --> 00:04:43.640 two seconds, but he still scraped by with a new  world record. However, the most notable takeaway   00:04:43.640 --> 00:04:52.600 was that the sum of Mukmuk’s best segments was  a 2:37. Indeed, with Mug 7 and sub 16 train,   00:04:52.600 --> 00:04:59.200 a sub 2:40 was on the cusp of being feasible.  Another discovery or two is likely all that was   00:04:59.200 --> 00:05:04.840 needed to piece together a viable route, but in  the meantime, runners were focused on optimizing   00:05:04.840 --> 00:05:11.480 the strategies they already had at their disposal. On June 21st, NorXor made it to the late game on   00:05:11.480 --> 00:05:19.000 easy sub 2:50 pace, despite having lost 2 seconds  from missing Bird 3. After hitting resetless 10,   00:05:19.000 --> 00:05:24.148 his nerves shot through the roof.  It all came down to level 11…  00:05:48.840 --> 00:05:57.160 With a heart rate of 188, NorXor choked a sub  2:50 at the last possible moment. Regardless,   00:05:57.160 --> 00:06:03.000 in an era dominated by new age runners,  NorXor proved that us old timers still had   00:06:03.000 --> 00:06:08.680 the potential to make a splash. However,  across the board, Red Ball still had a   00:06:08.680 --> 00:06:14.720 distinct lack of enthusiasm that was holding  back its potential development. Thankfully,   00:06:14.720 --> 00:06:18.612 there was an event in the pipeline  that sought to shake things up.  00:06:31.920 --> 00:06:38.840 Super Mario 63 is one of the most iconic flash  games ever created. An ambitious mashup of   00:06:38.840 --> 00:06:45.600 several mario games in a single 2D package, it  is a hallmark of the bustling creativity that   00:06:45.600 --> 00:06:52.000 the Flash platform provided. With a world record  history that is almost as storied as Red Ball’s,   00:06:52.000 --> 00:06:58.400 it only made sense for the two game’s communities  to combine forces at some point. Dignity pitched   00:06:58.400 --> 00:07:04.760 the idea of having members from each community  learn and speedrun the other game for a week   00:07:04.760 --> 00:07:10.720 to see what kind of time they could get. The top  3 runs for each game would be averaged together,   00:07:10.720 --> 00:07:16.280 and whatever community was closer to the world  record from a percentage standpoint would win.   00:07:16.280 --> 00:07:22.440 The SM63 community had a lot of talent to draw  from, most notably Vertic, who had the game’s   00:07:22.440 --> 00:07:29.640 Any% world record by a mile. Vertic had also swept  every world record in the original Vex trilogy,   00:07:29.640 --> 00:07:34.400 another set of flash game platformers. It  would be interesting to see what kind of   00:07:34.400 --> 00:07:40.760 times himself and other Mario 63 runners would  be able to pull off in a limited time frame:   00:07:40.760 --> 00:07:49.269 June 24th to June 30th. Vertic speedran Red Ball  for the first time on day 1, and then on day 3… 00:08:18.600 --> 00:08:26.480 In just 391 attempts, Vertic got a sub 3, a feat  that had taken most runners thousands of tries and   00:08:26.480 --> 00:08:33.160 months of effort. Setting a time only 7% slower  than Mukmuk’s world record, this was equivalent   00:08:33.160 --> 00:08:40.800 to achieving a sub 6:45 in sm63 in the context  of the competition. The best time that the Red   00:08:40.800 --> 00:08:49.200 Ball community ended up setting was a 7:53 by me.  The competition was set to be an absolute blowout,   00:08:49.200 --> 00:08:54.238 but right before the final results were  tallied, someone ended up lessening the blow. 00:09:37.600 --> 00:09:45.360 2:44.484. Poguin shattered the world record by 3  and a half seconds, making it the largest record   00:09:45.360 --> 00:09:52.840 jump since rare.eg’s iconic 3:07. In the era of  tool-developed setups, something like this was   00:09:52.840 --> 00:09:58.800 completely unprecedented. The run had almost  no mistakes, and it even implemented a few   00:09:58.800 --> 00:10:05.760 new strategies. Firstly, he had a new level  6 setup called Star 6, which only required 2   00:10:05.760 --> 00:10:12.520 frame perfects instead of 3. Not only was this  strategy easier, it also completely removed the   00:10:12.520 --> 00:10:18.480 reaction time barrier that was present with Max 6,  since the inputs came much later after the level   00:10:18.480 --> 00:10:25.480 transition. Next, he implemented a 4 frame perfect  setup in level 7 that built up enough speed to   00:10:25.480 --> 00:10:32.240 skip the moving platforms without using the top  route. This was around a second slower than Mug 7,   00:10:32.240 --> 00:10:37.720 but it proved to be a decent bit easier. The  only mistake present in the run was missing   00:10:37.720 --> 00:10:44.400 the corner boost at the end of PJet 11, costing  around 2 seconds. It is also worth noting that   00:10:44.400 --> 00:10:49.760 this was the first world record performed on the  Tournament Edition of the game, which included an   00:10:49.760 --> 00:10:56.880 in-game timer and printed the date at the end of  each run. The timer truncated to 3 decimal places,   00:10:56.880 --> 00:11:04.960 which is why the milliseconds of Poguin’s record  read .483 instead of the properly rounded .484.   00:11:04.960 --> 00:11:11.120 While the timing of runs in the past had always  been a bit ambiguous, since a 31 frames-per-second   00:11:11.120 --> 00:11:18.960 game does not cleanly fit into either a 30FPS or  60FPS video, any run performed on this version   00:11:18.960 --> 00:11:25.320 would have a 100% accurate retime. Some runners  still preferred playing on the standard version,   00:11:25.320 --> 00:11:29.560 but the tournament edition would become  more and more commonplace moving forward.  00:11:29.560 --> 00:11:35.480 Even factoring in Poguin’s improvement to the  world record, the Super Mario 63 community still   00:11:35.480 --> 00:11:42.280 beat the Red Ball community by over 17% in the  competition. Vertic had proved himself capable of   00:11:42.280 --> 00:11:48.680 being a top Red Ball runner, but only time would  tell if he’d continue to push his personal best.  00:11:48.680 --> 00:11:54.680 After Poguin’s 2:44, another world record  wasn’t expected to happen any time soon,   00:11:54.680 --> 00:12:01.720 and if one was, it would more than likely be by a  tiny margin. Yak27 was attempting to reclaim his   00:12:01.720 --> 00:12:07.840 record with the same route at Poguin, hoping to  gain an edge by nailing PJet. He got a run that   00:12:07.840 --> 00:12:13.760 missed the corner boost during bottom route  7, which lost around 1.5 seconds. However,   00:12:13.760 --> 00:12:20.522 his levels 8-10 were optimal, and he  had the ability to close the gap in 11. 00:12:40.720 --> 00:12:46.600 Yak27 beat the world record by just 6  frames. Though he was now back on top,   00:12:46.600 --> 00:12:53.040 he wanted to push the record further, as did the  man that he just bested. Poguin and Yak both knew   00:12:53.040 --> 00:12:59.160 that the current route could get a 2:42, but they  both decided to implement Mug 7, pushing that   00:12:59.160 --> 00:13:04.920 potential even lower. In August, they were still  far ahead of everyone else on the leaderboard,   00:13:04.920 --> 00:13:11.400 and the record was ready to fall at any moment. On  August 15th, Yak had a run that missed resetless   00:13:11.400 --> 00:13:17.800 4, meaning that he had to wait to get the death  warp on the back of the axe. He nailed Mug 7,   00:13:17.800 --> 00:13:22.920 but he unfortunately didn’t press left for long  enough before the super bounce, causing him to   00:13:22.920 --> 00:13:29.280 bonk the red wall. These two mistakes combined  lost around 2.5 seconds, but since he had messed   00:13:29.280 --> 00:13:35.480 up a slower level 7 strategy in his world record,  it was a bit of a wash. He took the lead after   00:13:35.480 --> 00:13:41.240 nailing a new strategy called Carline in level  8. With a frame perfect left to right switch at   00:13:41.240 --> 00:13:46.360 the start of the level and a jump on the ramp  in a two frame window, Red Ball gets enough   00:13:46.360 --> 00:13:52.040 height to clear this barrier without slowing  down, saving almost a second. He extended his   00:13:52.040 --> 00:13:58.885 lead after getting a near perfect plunger jump,  and a perfect PJet put him on pace for a 2:43. 00:14:22.950 --> 00:14:29.880 That accidental corner jump in level 12 cost  him the 2:43, but LiveSplit showed a new world   00:14:29.880 --> 00:14:37.520 record by 0.09 seconds. However, upon frame  counting, the real time of the run was revealed   00:14:37.520 --> 00:14:48.280 to be 2:44.258: just a single frame faster. A week later, Poguin had a run that got Star 6   00:14:48.280 --> 00:14:54.800 on the second try and nailed all the other setups.  However, since his hood launch and plunger were   00:14:54.800 --> 00:15:03.680 a bit less than optimal, the run clocked in  at 2:44.065, another 6 frame improvement. So,   00:15:03.680 --> 00:15:10.560 despite three new world records being set in the  past month, the time only dropped by 13 frames,   00:15:10.560 --> 00:15:18.040 or 0.419 seconds. It truly seemed like the game  was running into a wall in terms of optimization,   00:15:18.040 --> 00:15:24.840 and hope of another sizable record ever being  set was dwindling by the day. Yes, an optimal   00:15:24.840 --> 00:15:30.760 run with Poguin and Yak’s route could certainly  get a 2:41, but it seemed like the road to get   00:15:30.760 --> 00:15:39.371 there would be extremely long. However, that’s not  the path that the record history ended up taking.  00:16:28.960 --> 00:16:36.240 Well, that’s more like it. In one run, Yak27  improved the world record by six times what   00:16:36.240 --> 00:16:44.480 the previous three records had combined. He  nailed 2 corner jump, resetless 4, Star 6,   00:16:44.480 --> 00:16:51.640 Mug 7, resetless 10, and PJet 11 first  try. He did omit Carline this time around,   00:16:51.640 --> 00:16:57.080 but he added in Cookie 3, which saves  around a third of a second over Bird 3.   00:16:57.080 --> 00:17:02.000 His hood launch and plunger weren’t perfect,  but he still achieved a run that was near the   00:17:02.000 --> 00:17:08.120 limit for this route. The implementation of  another strategy or two is likely all that   00:17:08.120 --> 00:17:14.560 a sub 2:40 would need. He included a bold  prediction in the description of the run:   00:17:14.560 --> 00:17:21.920 “3x will happen in 2022!”. With only 4 months left  in the year, it would almost certainly be himself   00:17:21.920 --> 00:17:30.654 or Poguin claiming this final milestone should his  prediction become a reality. Well, not so fast. 00:18:23.000 --> 00:18:31.400 By the time Yak set his 2:41, Vertic had shot up  to 3rd place on the leaderboard with a 2:44.613,   00:18:31.400 --> 00:18:38.040 placing the top 3 in a dead heat. After Yak  demolished the record, Vertic knew that he   00:18:38.040 --> 00:18:44.040 had to implement more strategies to increase his  odds of breaking the record and possibly setting   00:18:44.040 --> 00:18:51.120 the first sub 2:40. He added Mug 7 to match Yak’s  route, and he brought Carline 8 back in from the   00:18:51.120 --> 00:18:58.960 sidelines. He also incorporated a setup in level  12 that NotVeryGood used in a TAS back in July   00:18:58.960 --> 00:19:05.680 of 2022. With a frame perfect jump release at  the start and a frame perfect jump on the pillar,   00:19:05.680 --> 00:19:10.720 Red Ball can bounce in the spike pit  without dying. By jumping after the recoil,   00:19:10.720 --> 00:19:16.400 Red Ball will avoid bonking into this ceiling,  saving around half a second. Furthermore,   00:19:16.400 --> 00:19:22.280 by frame perfectly jumping out of the tunnel,  Red Ball will be set up for a corner jump off   00:19:22.280 --> 00:19:28.840 this stair, saving another full second. This  strategy was the final piece in the puzzle   00:19:28.840 --> 00:19:35.560 of a viable 2:3x route, and since it was also  the very last thing that had to be performed,   00:19:35.560 --> 00:19:42.560 it was aptly named 3x choker. Vertic was the  most prepared out of anyone to shatter the game’s   00:19:42.560 --> 00:19:50.177 final 10 second barrier, and just a week after  setting his 2:44, he had his first opportunity.  00:20:04.400 --> 00:20:12.160 Going into level 11, he was only one and a half  seconds behind his sum of best of 2:37.95. With   00:20:12.160 --> 00:20:20.002 PJet and 3x choker, this could easily be  the one. Let’s see what Vertic made of it.  00:20:42.200 --> 00:20:49.040 By releasing jump a frame early at the start,  Vertic ended up dying in the spike pit, choking   00:20:49.040 --> 00:20:55.760 3x choker. He also flubbed his second attempt,  requiring him to implement a frame-dependent   00:20:55.760 --> 00:21:02.080 backup to still get the corner jump on the stairs.  Despite the heartbreaking failure, the run still   00:21:02.080 --> 00:21:08.200 had significant context in the overall quest  for perfection with Red Ball speedruns. Never   00:21:08.200 --> 00:21:13.920 in the game’s history had a top-level runner  been able to play the first 11 levels of the   00:21:13.920 --> 00:21:22.040 game anywhere close to within 1.5 seconds of  their theoretical cap. In fact, rare.eg’s 3:07,   00:21:22.040 --> 00:21:28.720 a run that was hailed as essentially being perfect  for the strategies used, was still the closest   00:21:28.720 --> 00:21:36.320 a world record had ever gotten to the runner’s  sum of best, over 3 seconds away. At this point,   00:21:36.320 --> 00:21:42.800 every single level in the game involved a  frame-dependent strategy except for 1, 5, and   00:21:42.800 --> 00:21:49.720 9. And, since level 1 just involves a single jump  over this barrier, and level 5 just requires a   00:21:49.720 --> 00:21:56.440 jump up this slope and then to the flag, the only  level that fully relied on non-fixed movement was   00:21:56.440 --> 00:22:03.200 level 9. While it had taken a while for runners  to master this new era of Red Ball speedrunning,   00:22:03.200 --> 00:22:09.480 it finally seemed like the end game had  arrived. If you wanted to set a top level time,   00:22:09.480 --> 00:22:15.120 you simply selected a frame-dependent setup for  each level, added the times that they yielded,   00:22:15.120 --> 00:22:20.000 left a second or two of wiggle room  for the bits that were still variable,   00:22:20.000 --> 00:22:24.880 and that is the time that you could expect  to get. All that it came down to was whether   00:22:24.880 --> 00:22:32.520 you hit all of the frame perfects or not, of  which, the sub 2:40 route had roughly 23. And,   00:22:32.520 --> 00:22:36.960 on the subject of frame perfect counts,  every single time that I mention them,   00:22:36.960 --> 00:22:43.520 I get plenty of people commenting the exact same  thing. “Are you serious? Geometry Dash has levels   00:22:43.520 --> 00:22:50.680 with over a hundred frame perfects. Hitting 23 is  so easy!” Alright, alright, geez dude calm down.   00:22:50.680 --> 00:22:57.520 Give me a few minutes to explain why analyzing the  difficulty of frame perfects isn’t that simple. 00:22:57.520 --> 00:23:05.480 Red Ball is a 31FPS game, meaning that each frame  should last for about 32 milliseconds. In theory,   00:23:05.480 --> 00:23:11.840 hitting 23 frame perfect inputs in a row with  that window shouldn’t be too bad. In fact,   00:23:11.840 --> 00:23:18.323 here’s a video of me hitting 23 “frame  perfects” in a row on my first try! 00:23:20.467 --> 00:23:25.680 Yeah, every input may have had equal  spacing with a distinct visual cue,   00:23:25.680 --> 00:23:33.760 but that was still 23 inputs, each with a 32  millisecond window. Now, here’s me hitting   00:23:33.760 --> 00:23:40.160 200 “frame perfects” in a row… so I guess I  must be ready to beat Tidal Wave! Alright,   00:23:40.160 --> 00:23:46.320 so the context behind frame perfect inputs clearly  matters when determining how difficult they are,   00:23:46.320 --> 00:23:49.640 so let me break down the factors  that make frame perfect inputs in   00:23:49.640 --> 00:23:55.360 Red Ball much different from other games. First of all, Red Ball has four buttons to   00:23:55.360 --> 00:24:00.720 keep track of, while the examples that I  have shown thus far only require one. The   00:24:00.720 --> 00:24:05.440 more inputs that you have to keep track  of, the harder your job is going to be. 00:24:05.440 --> 00:24:11.160 Next up, you have the impact of rhythm. Generally,  the more rhythm-based the inputs are, the   00:24:11.160 --> 00:24:17.480 easier they are to perform. Even complex “click  patterns,” as they are called in Geometry Dash,   00:24:17.480 --> 00:24:23.080 can produce a distinct rhythm that can become more  and more refined through practice. With many of   00:24:23.080 --> 00:24:28.480 the setups in Red Ball, the frame perfect  inputs can take place seconds apart from   00:24:28.480 --> 00:24:33.720 each other, yielding no such pattern. Rhythm is also related to the audio and   00:24:33.720 --> 00:24:39.920 visual cues that the game provides. For my a dance  of fire and ice demonstration, I just had to hit   00:24:39.920 --> 00:24:46.560 a button with the consistent beat of a metronome…  it doesn’t get any easier than that. Even without   00:24:46.560 --> 00:24:53.320 audio, there is a very distinct visual cue to look  out for, the ball overlapping the path. With the   00:24:53.320 --> 00:25:00.360 balls moving extremely fast, this only happens  for a brief moment in time. Compare that to the   00:25:00.360 --> 00:25:06.880 first right press in PJet 11. This is the frame  before, this is the frame that you need to hit it,   00:25:06.880 --> 00:25:12.840 and this is the frame after. There is almost  no distinction. Even with Geometry Dash,   00:25:12.840 --> 00:25:18.680 the visual change over the same time  frame is often immense in comparison.  00:25:18.680 --> 00:25:25.160 Regardless, the best proof that “frame perfect”  counts aren’t a universal measure of difficulty   00:25:25.160 --> 00:25:31.040 comes from the results of people who have actually  played multiple frame-dependent games at a high   00:25:31.040 --> 00:25:37.320 level. Thankfully, we have a great example of  this, since one of the world’s best Geometry Dash   00:25:37.320 --> 00:25:45.120 players, Trick, managed to get a mid 4:56 in Super  Mario Bros., a feat which requires roughly 20   00:25:45.120 --> 00:25:52.560 frame perfects at 60FPS. On paper, this is so much  easier than his completion of Tidal Wave that he   00:25:52.560 --> 00:25:58.400 should have been able to just boot up the game and  get it on like his 10th try or something. But no,   00:25:58.400 --> 00:26:03.960 it actually took him weeks of practice and  dozens of hours of attempts to pull off. Don’t   00:26:03.960 --> 00:26:11.320 get me wrong, achieving a mid 4:56 that fast  is still extremely impressive and practically   00:26:11.320 --> 00:26:17.600 unheard of… but it didn’t happen instantly,  like this graph would lead you to believe.  00:26:17.600 --> 00:26:23.400 Going even further, there is something baked  into Flash Player itself that makes frame perfect   00:26:23.400 --> 00:26:29.000 inputs less consistent than they otherwise should  be. If you have seen my Red Ball TAS explained   00:26:29.000 --> 00:26:35.120 video, then you should be aware that the time  that each frame takes up is not the same. The   00:26:35.120 --> 00:26:41.520 frame rate that is specified for any Flash game  is simply a target that it is trying to hit. In   00:26:41.520 --> 00:26:48.520 other words, frames can be shorter or longer than  32 milliseconds, so long as the average framerate   00:26:48.520 --> 00:26:55.920 comes out to around 31. On good machines, this  error is usually just a few milliseconds or so.   00:26:55.920 --> 00:27:01.960 However, upon opening Adobe Scout, an official  tool for profiling the performance of Flash   00:27:01.960 --> 00:27:09.920 Player, the truth is revealed to be much more  severe. On my computer, every 15-16 frames,   00:27:09.920 --> 00:27:16.360 Flash throws in a frame that is half the length,  compensating for all of the other frames being   00:27:16.360 --> 00:27:22.000 a bit longer than they should be. As you can  imagine, if you need to perform a frame perfect   00:27:22.000 --> 00:27:28.320 input on any of these half frames, you are  far more likely to miss it. And, for slower   00:27:28.320 --> 00:27:35.240 machines like my laptop, the spikes and dips  are even worse. To demonstrate how bad this is,   00:27:35.240 --> 00:27:41.840 I will examine the exact input and frame time  data when performing Resetless 4 on both my   00:27:41.840 --> 00:27:49.120 desktop and my laptop for 5 runs each. You can  see that jump must be pressed on frame 149,   00:27:49.120 --> 00:27:56.680 released on frame 150, and then pressed again  on frame 152. However, if we extend the exact   00:27:56.680 --> 00:28:04.080 input timings from the first trial to each of the  others, you can see that only 5 of the remaining   00:28:04.080 --> 00:28:10.600 9 would have succeeded. Yes, this is a small  sample size, but the overall lesson here is   00:28:10.600 --> 00:28:16.560 that a certain level of randomness is involved  with almost every single frame perfect input   00:28:16.560 --> 00:28:23.080 in Red Ball. We will discuss how this impacts  the human limit of the game later in the video,   00:28:23.080 --> 00:28:29.760 but just know for now that any frame perfect  count that I list in this video is not directly   00:28:29.760 --> 00:28:38.240 comparable to other games. Now, with that out of  the way, let’s merge back on the road to 2:3x. 00:28:38.240 --> 00:28:44.400 Though Vertic was still quite a bit behind  Yak, his 2:43 proved that he was likely the   00:28:44.400 --> 00:28:50.680 favorite to be the first to get a sub 2:40.  He just needed to get that pace again and   00:28:50.680 --> 00:28:56.800 then hit 3x choker. How hard could it  possibly be? Well, the very same day,   00:28:56.800 --> 00:29:03.120 he got a run that missed the first frame perfect  of Cookie 3, so he reset the level to get it on   00:29:03.120 --> 00:29:09.680 his second try. Similarly, he messed up the left  right switch at the start of Carline 8 and had to   00:29:09.680 --> 00:29:15.760 quickly reset to fix it. Another second lost  to his sum of best. Even though 3x was dead,   00:29:15.760 --> 00:29:21.640 he still continued the run since the world  record was barely within his grasp. Plunger?   00:29:21.640 --> 00:29:31.252 Check. Resetless 10? Check. PJet? Check. All he  needed was 3x choker, and the record would be his… 00:29:44.440 --> 00:29:48.360 Well, he messed up the start, but he  nailed the backup to still get the   00:29:48.360 --> 00:29:54.360 corner jump off the stairs. Still, he  found himself 4 frames off the record.  00:29:54.360 --> 00:30:02.920 At this point, the only people who had ever gotten  a sub 2:50 were Yak, Poguin, Vertic, Mukmuk,   00:30:02.920 --> 00:30:09.560 and Caramel. Though there was obviously a clear  time gap between the top 3 and everyone else,   00:30:09.560 --> 00:30:16.040 Mukmuk wasn’t as active as he used to be, and  Caramel had long since left the game behind.   00:30:16.040 --> 00:30:21.680 There were just three guys that had a chance of  breaking that milestone, and it seemed pretty set   00:30:21.680 --> 00:30:28.120 in stone that no one else would get to their level  before the glass was finally shattered. However,   00:30:28.120 --> 00:30:36.350 that certainty was eroded the very next day, as  another sub 2:50 came from an unlikely source… 00:30:49.840 --> 00:30:56.640 To explain why on earth Niftski, the greatest  Super Mario Bros. speedrunner to ever live,   00:30:56.640 --> 00:31:03.400 set a 4th place run in Red Ball out of seemingly  nowhere, let’s wind the clock back to January   00:31:03.400 --> 00:31:09.400 2021, when the rise of Red Ball began. As  thousands of speedrunners flooded to the Red   00:31:09.400 --> 00:31:15.760 Ball community, largely as a result of my videos  being thrust into the good graces of the YouTube   00:31:15.760 --> 00:31:22.680 algorithm, Niftski was one of the runners who  decided to give the game a shot. On January 19th,   00:31:22.680 --> 00:31:29.240 he followed the beginner speedrun tutorial that  I had created to get a sub 4-minute time on his   00:31:29.240 --> 00:31:35.600 first day. This eventually led to many other Super  Mario Bros. speedrunners trying the game out,   00:31:35.600 --> 00:31:43.600 including AldynSpeedruns, JeremyMKW,  Saywer0622, Hitzcritz, and more. Miniland,   00:31:43.600 --> 00:31:48.520 who had the world record in Super Mario  Bros. at the time, even popped into the   00:31:48.520 --> 00:31:53.040 Red Ball Discord for a little while, though  he wasn’t interested in playing the game.  00:31:53.040 --> 00:31:59.800 Red Ball had so many new players that, near the  start of February, it had over double Super Mario   00:31:59.800 --> 00:32:06.240 Bros.’ active player count on speedrun.com. While  the two games were very different in their design,   00:32:06.240 --> 00:32:11.400 one was a physics sandbox while the other was  a fixed-speed platformer, they were similar in   00:32:11.400 --> 00:32:17.800 the fact that they were short, easy-to-pick-up  speedgames with no RNG. Well, aside from spike   00:32:17.800 --> 00:32:24.982 glitch, since this was the pre-deterministic  patch era. Niftski was obviously capable of   00:32:24.982 --> 00:32:31.840 getting much more than a 3:53, and on February  18th, he improved his PB to a 3:27. For the   00:32:31.840 --> 00:32:37.560 next two months, he put his full focus towards  grinding Super Mario Bros., culminating with   00:32:37.560 --> 00:32:44.800 him setting the first 4:54, shattering the game’s  final second barrier. He eventually found his way   00:32:44.800 --> 00:32:51.560 back to Red Ball, and in June of 2021, he hit the  infamous sub 20 train strategy for the first time,   00:32:51.560 --> 00:32:59.320 using the newly found PBird setup. Four  months later, he managed to set a 3:17.065,   00:32:59.320 --> 00:33:05.880 a good time that would have flown largely under  the radar… had it not been for him hitting sub 20   00:33:05.880 --> 00:33:13.440 train in the run. This actually set the record for  the slowest personal best with a first try sub 20   00:33:13.440 --> 00:33:23.600 train. Surprisingly, it wasn’t until June 20th,  2022 that he came back again, setting a 3:11.645   00:33:23.600 --> 00:33:31.560 and a 3:05.968 on the same day. The keen eyed  among you may have noticed that, despite one and   00:33:31.560 --> 00:33:40.000 a half years passing, Niftski’s attempt counter  sat at just 544. Niftski found massive success by   00:33:40.000 --> 00:33:46.560 practicing way more than doing attempts, the same  approach that he took to Super Mario Bros. As a   00:33:46.560 --> 00:33:53.280 hint to any up and coming speedrunners out there,  in platformer games without RNG, this is always   00:33:53.280 --> 00:33:59.709 a winning strategy. Niftski’s approach would  continue to pay dividends, as that same night… 00:34:21.478 --> 00:34:27.840 Niftski got a sub 3 in just 602 attempts,  clutching it out with a gold level 11   00:34:27.840 --> 00:34:34.520 and a gold level 12. This was the record for  the least attempts to get a sub 3 until Vertic   00:34:34.520 --> 00:34:43.760 did it in 391 a week later during the SM63  crossover event. In the span of just 8 hours,   00:34:43.760 --> 00:34:51.400 Niftski pushed his personal best from a 3:17 to  a 2:59, something that was completely unheard   00:34:51.400 --> 00:34:57.440 of and almost unimaginable until he somehow  made it happen. As you heard from that popoff,   00:34:57.440 --> 00:35:02.600 Niftski genuinely thought that he was  never going to touch the game again,   00:35:02.600 --> 00:35:07.320 but Red Ball has a unique  ability to drag people back in. 00:35:07.320 --> 00:35:15.000 In late August, Niftski got a sudden spike in  motivation to go for sub 2:50. On August 30th,   00:35:15.000 --> 00:35:19.294 he jumped all the way down to a 2:52.903. 00:35:29.520 --> 00:35:37.875 Indeed, PJet and a better level 12 is all that he  needed for the 4x. Two days later, he got this. 00:35:51.467 --> 00:35:56.320 Niftski started livestreaming on his  alternate account, and 11 attempts later,   00:35:56.320 --> 00:36:02.080 he had a run that was a bit behind  going into level 12. 3x choker could   00:36:02.080 --> 00:36:07.160 have granted him another PB, but  he didn’t know it or the backup…  00:36:21.160 --> 00:36:27.720 A week later, he was easily on 4x pace going  into level 11, and he opted to do PBird   00:36:27.720 --> 00:36:33.760 instead of PJet since he thought that he had  enough wiggle room. However, he forgot that he   00:36:33.760 --> 00:36:40.720 had done the setupless stair jump in his personal  best, and he ended up coming up just short. But,   00:36:40.720 --> 00:36:48.640 as you saw earlier, he managed to get the  2:47.871 later that same day. With how fast   00:36:48.640 --> 00:36:53.920 Niftski had been able to improve each time  that he returned to the game, it truly seemed   00:36:53.920 --> 00:37:01.080 like he may position himself as a 3x contender.  However, he again stated at the end of the run…  00:37:08.160 --> 00:37:14.720 Whether he was being truthful this time had yet to  be seen. Still, Niftski’s emphatic return to the   00:37:14.720 --> 00:37:21.480 game had finally motivated another high-profile  mario speedrunner to give the game a shot. 00:37:50.000 --> 00:37:57.880 On September 19th, Miniland managed to get  a sub 3 in just 185 attempts, demolishing   00:37:57.880 --> 00:38:04.920 Vertic’s record. Believe it or not, it only  took him 17 attempts to get to level 11 on sub   00:38:04.920 --> 00:38:12.680 3 pace for the first time. Again, an experienced  mario 1 speedrunner had found enormous success by   00:38:12.680 --> 00:38:18.680 applying a practice-heavy approach to Red Ball  speedruns. And by heavy, I mean that Miniland   00:38:18.680 --> 00:38:25.040 spent an entire week practicing the game before  doing any attempts. Miniland estimates that his   00:38:25.040 --> 00:38:32.960 ratio of time spent practicing versus doing  runs ended up being around 90 to 10. However,   00:38:32.960 --> 00:38:38.720 it wasn’t just the sheer amount of time spent  practicing that brought him success. What you   00:38:38.720 --> 00:38:44.600 practice and how you practice it is even  more important. Miniland made sure to laser   00:38:44.600 --> 00:38:51.800 in on the most difficult pieces of the run,  namely Cookie 3, Resetless 4, 8 Cycle Skip,   00:38:51.800 --> 00:38:58.880 and PBird, utilizing savestates, the frame offset  display, and the heads-up display mod to make his   00:38:58.880 --> 00:39:05.560 practice as efficient and effective as possible. In the description of his run, Miniland stated,   00:39:05.560 --> 00:39:10.040 “I will be going for a time in the 2 minute 40  second range, which would place me comfortably in   00:39:10.040 --> 00:39:15.760 the top 10, and who knows what will happen after  that.” At this point, Red Ball speedrunning seemed   00:39:15.760 --> 00:39:21.200 like it had all the momentum in the world, with  multiple people actively gunning for the game’s   00:39:21.200 --> 00:39:27.400 most formidable milestone. Given all the hype,  the community felt like it was a great time to   00:39:27.400 --> 00:39:34.600 organize a special event to showcase everyone’s  talent. Welcome to Red Ball: Break the Record. 00:39:34.600 --> 00:39:41.160 On October 1st and 2nd, the Red Ball Community  ran an event that featured several top runners   00:39:41.160 --> 00:39:49.680 aiming to beat the 12 Levels world record. The  talent included TheEgglet, Pearce, NorXor, Sprong,   00:39:49.680 --> 00:39:58.520 Mukmuk, Poguin, Yak, and, of course, Vertic. For  8 hours each day, the runners could put up as many   00:39:58.520 --> 00:40:05.000 attempts as they wanted, with the goal of setting  the best time possible. The 4 players with the   00:40:05.000 --> 00:40:11.920 best runs during the event would get a portion of  the prize pool, with $100 reserved for anyone who   00:40:11.920 --> 00:40:19.160 set the first 2:3x during the stream. This kind  of event is exactly what the Tournament Edition   00:40:19.160 --> 00:40:25.880 of the game was made for. The built in timer  avoids having to retime runs, and the date display   00:40:25.880 --> 00:40:32.560 ensures that nobody is streaming pre recorded  attempts. While the event was a success, it's hard   00:40:32.560 --> 00:40:39.040 to say much about it, since only a small amount  of video has survived. The most notable thing   00:40:39.040 --> 00:40:47.960 that happened was Vertic getting a world record  paced run into level 12 that ended as a 2:42.129,   00:40:47.960 --> 00:40:57.760 just short again. Soon after, he managed to set  a 2:43.774, which he evidently wasn’t very happy   00:40:57.760 --> 00:41:05.280 about. Still, he ended up finishing first place  in the event by a mile, clearly cementing himself   00:41:05.280 --> 00:41:12.320 as the frontrunner to set the first sub 2:40. Off the heels of this successful and entertaining   00:41:12.320 --> 00:41:18.400 showcase, I started planning another Red Ball:  Break the Record that would feature a larger   00:41:18.400 --> 00:41:24.520 prize pool, more runners and commentators, and  a more professional broadcast. While the event   00:41:24.520 --> 00:41:32.880 was set to be much shorter, just 5 hours instead  of 16, the greater presence of hype and talent   00:41:32.880 --> 00:41:40.414 meant that a new world record could easily be  set. Well, on the very first run of the event… 00:41:44.400 --> 00:41:52.520 Vertic got a 2:55 on demand, and Pearce  also managed to get a sub 3. Viewers were   00:41:52.520 --> 00:41:58.840 certainly in for a treat, and it didn’t take  long for Vertic to move the goalposts further.  00:42:06.040 --> 00:42:11.080 Around the two hour mark, the top of the  event leaderboard was starting to get a   00:42:11.080 --> 00:42:16.480 bit crowded. However, when Vertic  got an opportunity to pull away…  00:42:26.760 --> 00:42:32.240 After being the only person to set a  sub 2:50 in the first Break the Record,   00:42:32.240 --> 00:42:39.080 Vertic became the first to do it here, a time  that likely could have won him the event. However,   00:42:39.080 --> 00:42:44.240 there was still three hours left for  runs, and his aspirations were much   00:42:44.240 --> 00:42:50.325 larger than a 1st place spot. Here’s a  run that he got less than an hour later. 00:43:18.840 --> 00:43:26.520 The man had a 3x paced run past the large spikes  in level 11, but he didn’t get the corner boost.   00:43:26.520 --> 00:43:32.600 He didn’t quite nail the starting inputs for  3x choker, and he just took the ending safe   00:43:32.600 --> 00:43:40.400 instead of resetting at the checkpoint flag to  attempt the backup. 2:43.516. This is the best   00:43:40.400 --> 00:43:46.280 run that anyone ended up getting during the  event, meaning that Vertic was in first place   00:43:46.280 --> 00:43:53.320 for the entire thing. Sprong ended up being  the only other runner to even get a sub 2:50,   00:43:53.320 --> 00:43:59.400 further cementing Vertic’s perceived dominance  over the category, despite him still not having   00:43:59.400 --> 00:44:06.308 the world record. Well, the next weekend,  he had an excellent opportunity to fix that.  00:44:20.400 --> 00:44:27.880 The strategy you just saw is known as 9x3, a  4 frame perfect setup to jump off the plunger   00:44:27.880 --> 00:44:34.280 a whole cycle early. Since he got the Star 6  inputs on his second try, Vertic needed that   00:44:34.280 --> 00:44:38.218 to stay on 3x pace. Let’s watch the rest. 00:45:14.842 --> 00:45:26.156 The ending of PJet trolled him yet again,  but he still clutched 3x choker to claim a new world record of 2:40.355. 00:45:26.156 --> 00:45:32.880 Now, it may look like Vertic was pretty far away from  collecting the flag before it went underground,   00:45:32.880 --> 00:45:38.200 but Red Ball’s hitbox is often larger than  it should be due to a certain box-in-box   00:45:38.200 --> 00:45:49.960 phenomenon. By replaying his inputs as a TAS, we  find that, on this frame, Vertic was 0.0017 units   00:45:49.960 --> 00:45:57.280 away from contacting the flag. We can convert this  value to find that the gap which separated Vertic   00:45:57.280 --> 00:46:06.920 from Red Ball glory was around a seventh of a  pixel at 1080p. This man could not catch a break.  00:46:06.920 --> 00:46:12.360 Still, Vertic continued streaming attempts for  the rest of October and all through November,   00:46:12.360 --> 00:46:17.600 but he couldn’t quite close anything out.  He even had another run make it to level   00:46:17.600 --> 00:46:25.160 12 on 3x pace that missed 3x choker. It wasn’t  until December that the top of the leaderboard   00:46:25.160 --> 00:46:31.960 saw any kind of movement. You’d probably expect  Yak or Poguin to be the one to finally shake up   00:46:31.960 --> 00:46:39.120 the race to 3x, but that momentum ended  up coming from a very unlikely source.  00:46:39.120 --> 00:46:44.120 Remember when Miniland said that he would  go for a sub 2:50 at some point in the   00:46:44.120 --> 00:46:51.120 description of his sub 3? Well, he ended  up achieving that goal just 8 days later.  00:46:51.120 --> 00:47:00.440 Indeed, this 2:46.645 launched him to 4th place,  just ahead of Niftski. The run was nearly flawless   00:47:00.440 --> 00:47:06.080 for the route he was going for, which included  a train strategy that saved half a second over   00:47:06.080 --> 00:47:13.320 PJet. On paper, this strategy was more difficult,  since it had 5 frame perfects instead of 4, but   00:47:13.320 --> 00:47:19.760 for whatever reason, Miniland and Niftski found  it to be a good bit easier. Since the Super Mario   00:47:19.760 --> 00:47:27.040 Bros. speedrunners had found more success with it  compared to everyone else, it was coined SMB11.   00:47:27.040 --> 00:47:32.320 Miniland stated that he could comfortably quit  the game after this run and still be satisfied,   00:47:32.320 --> 00:47:36.880 but we all know how that usually ends  up going. Towards the end of November,   00:47:36.880 --> 00:47:42.880 Miniland had a sudden burst in motivation to  hunt for the first sub 2:40. He spent some   00:47:42.880 --> 00:47:48.560 time learning and practicing strategies  from Vertic’s 3x route, including Mug 7,   00:47:48.560 --> 00:47:56.320 Carline 8, and 3x choker. However, he decided  to switch out Cookie 3 with a setup for TAZ 3,   00:47:56.320 --> 00:48:04.080 adding 2 frame perfects to save 0.8 seconds. TAZ  3 involves getting a boost off the platform itself   00:48:04.080 --> 00:48:10.080 instead of from the back corner, something which  is extremely precise and almost impossible to do   00:48:10.080 --> 00:48:16.880 without a frame-dependent setup. It actually  has a rich history dating back to early 2021,   00:48:16.880 --> 00:48:23.200 with InsaneJetman having made a 20 minute long  video documenting the strategy’s discovery by   00:48:23.200 --> 00:48:28.800 BirdBanana and his own attempts to replicate  it in real time without a frame-dependent   00:48:28.800 --> 00:48:35.640 setup. With the extra wiggle room that TAZ  3 provided, Miniland also subbed out 9x3   00:48:35.640 --> 00:48:42.120 for an easier strategy called EasyJet 9, which  avoids the plunger entirely and loses around a   00:48:42.120 --> 00:48:48.040 second. With his new route prepared, Miniland  was ready for runs, and on November 30th,   00:48:48.040 --> 00:48:56.293 he got a 2:43.645, moving him to 3rd place  ahead of Poguin. And then, on December 4th… 00:49:34.560 --> 00:49:41.880 The race to 3x had suddenly turned into a dead  heat between Vertic and Miniland. The only reason   00:49:41.880 --> 00:49:48.880 that this run wasn’t a 3x is because Miniland  messed up resetless 4, losing around 2 seconds.   00:49:48.880 --> 00:49:55.800 Miniland was rapidly improving, and a sub 2:40  was likely to be his in a matter of days. If   00:49:55.800 --> 00:50:03.120 Vertic wanted to be the first, he had to act now.  After spending several months just barely coming   00:50:03.120 --> 00:50:10.000 up short, it was time for him to show off his full  potential. Let’s see what happened the next day. 00:50:37.280 --> 00:50:42.080 Well, that missed cycle skip and bad  hood launch in 8 was practically a   00:50:42.080 --> 00:50:50.760 death sentence. Even a perfect ending with  9x3, Resetless 10, PJet 11, and 3x Choker   00:50:50.760 --> 00:50:56.990 wouldn’t be enough for the sub 2:40. That  is, unless he had a trick up his sleeve. 00:51:59.960 --> 00:52:05.760 When it mattered most, Vertic clutched up.  The man pulled out a strategy that seemed   00:52:05.760 --> 00:52:11.880 forbidden for us normal speedrunners to use,  and it saved his run. I guess his experience   00:52:11.880 --> 00:52:18.720 with Super Mario 63 really did pay off. The  accomplishment brought much celebration, and   00:52:18.720 --> 00:52:26.840 Yak’s prediction that a sub 2:40 would happen in  2022 had become a reality. And, as it turned out,   00:52:26.840 --> 00:52:32.568 another monumental moment for the Red  Ball Community was just around the corner. 00:52:47.320 --> 00:52:52.960 So yeah, the day after Red Ball’s last  10 second barrier was finally shattered,   00:52:52.960 --> 00:52:58.920 I hit 100,000 subscribers on YouTube! I was  able to celebrate the occasion with dozens   00:52:58.920 --> 00:53:05.280 of others on Red Ball Online, a multiplayer mod  of the game that was created by smartcube, Sage,   00:53:05.280 --> 00:53:12.760 and ImPurpl3. Fittingly, I hit 100k just as I  finished Train, the level that was responsible   00:53:12.760 --> 00:53:19.720 for the video that thrust my channel into the  spotlight back in 2021. I spent most of the stream   00:53:19.720 --> 00:53:26.800 running simultaneous races with everyone for fun,  and included in that swarm of balls was Niftski,   00:53:26.800 --> 00:53:32.320 who hadn’t done attempts for a few months.  However, for the past week, he had put in some   00:53:32.320 --> 00:53:41.000 practice with the newer strategies, namely TAZ  3, EasyJet9, SMB11, and 3x Choker. My celebration   00:53:41.000 --> 00:53:47.680 stream was the last bit of motivation that he  needed to finally return to doing runs, and he hit   00:53:47.680 --> 00:53:54.274 the ground running right after the stream ended.  And, just 30 minutes into his attempt session…  00:54:05.200 --> 00:54:11.560 This dude is a machine. Yeah, he took  quick resets for TAZ 3 and Star 6,   00:54:11.560 --> 00:54:17.840 he had a less than optimal hood launch in  level 8, and he hit the backup for 3x choker,   00:54:17.840 --> 00:54:23.480 but everything else was flawless. This  was a good first step towards a sub 2:40,   00:54:23.480 --> 00:54:28.840 but he still had a long grind ahead of  him if he chose to take that path. I mean,   00:54:28.840 --> 00:54:35.320 he was still in 5th place after all. Niftski  could have easily called it a night here, but,   00:54:35.320 --> 00:54:42.771 for whatever reason, he decided to keep going.  And, it’s certainly a good thing that he did. 00:55:21.760 --> 00:55:28.360 Alright, so remember earlier when I said that  it was “almost unimaginable” that Niftski had   00:55:28.360 --> 00:55:36.720 improved his PB from a 3:17 to a 2:59 in  a span of just 8 hours? That doesn’t even   00:55:36.720 --> 00:55:43.720 hold a candle to the miracle that he pulled off  during this attempt session. He demolished the   00:55:43.720 --> 00:55:50.920 world record by over two seconds, leaping  all the way from 5th place to 1st. He had   00:55:50.920 --> 00:55:59.400 improved his PB by over 10 seconds in a span of  less than 2 and a half hours. The cherry on top   00:55:59.400 --> 00:56:05.600 is that he could have still gotten the record  without 3x Choker, but he went for it anyways,   00:56:05.600 --> 00:56:13.040 making it the largest world record jump  since Poguin’s 2:44.484. It’s no wonder   00:56:13.040 --> 00:56:19.720 that Niftski cited this run as possibly being  the biggest fluke in his speedrunning career.  00:56:19.720 --> 00:56:25.560 This was the first time in Red Ball’s speedrunning  history that someone had made it to the precipice   00:56:25.560 --> 00:56:31.560 of full-game perfection. Aside from getting an  even better hood launch than the already excellent   00:56:31.560 --> 00:56:37.800 one that Niftski got, you could not improve this  run without incorporating harder setups into the   00:56:37.800 --> 00:56:43.880 route. The good news is that there were still a  few more strategies on the sidelines waiting to   00:56:43.880 --> 00:56:50.720 be thrown in. We had already seen Carline  8 and 9x3 used in previous world records,   00:56:50.720 --> 00:56:57.840 leaving at least a second and a half of time save  on the table. However, that just wasn’t enough to   00:56:57.840 --> 00:57:03.880 compel people to keep going for the record, and  incorporating both setups would come at the cost   00:57:03.880 --> 00:57:11.320 of 6 frame perfects. Red Ball desperately needed  another strategy to push its human ceiling a bit   00:57:11.320 --> 00:57:18.080 higher and give any record-hopeful runners enough  wiggle room to justify going for it. Thankfully,   00:57:18.080 --> 00:57:24.302 on December 19th, that need was filled,  as Carline 8 would become supercharged.  00:57:29.720 --> 00:57:35.440 With 3 frame perfect one-frame-long jumps,  Red Ball could get an insane launch off   00:57:35.440 --> 00:57:40.640 the tip of the car. However, unlike most  frame dependent strategies in the game,   00:57:40.640 --> 00:57:46.200 all 3 of these inputs had pretty distinct  visual cues, and if you missed any one of them,   00:57:46.200 --> 00:57:52.000 you could usually still get a decent hood launch  and continue your run. Miniland immediately took   00:57:52.000 --> 00:57:57.280 notice of the strategy, and he started getting  the hang of it pretty quick. Funnily enough,   00:57:57.280 --> 00:58:04.360 he began drawing some similarities between this  setup and a strategy known as pl8-1 in Super   00:58:04.360 --> 00:58:10.320 Mario Bros. It had a similar position in the run,  and the left to left to right inputs at the start   00:58:10.320 --> 00:58:17.280 of the level were somewhat similar to the perfect  fast acceleration that pl8-1 required. Miniland   00:58:17.280 --> 00:58:24.240 suggested that the strategy be called pl8-8 as  a result, and it instantly stuck. (in a plate?).  00:58:24.240 --> 00:58:31.600 Miniland titled his first perfect pl8-8 “a sneak  peek of the next red ball wr,” indicating that   00:58:31.600 --> 00:58:38.280 he had clear intentions of claiming it with this  strategy. That very same day, he had a run that   00:58:38.280 --> 00:58:44.720 got the Carline portion of pl8-8 but failed the  ending, still placing him ahead of the record.   00:58:44.720 --> 00:58:52.800 He extended his lead to 1.1 seconds by nailing  9x3. He got resetless 10 but had a slow ending,   00:58:52.800 --> 00:58:59.840 shrinking the lead to .2. Still, he had just  gotten the best ever exits out of levels 8,   00:58:59.840 --> 00:59:04.118 9, and 10, and he readied himself for SMB11. 00:59:06.757 --> 00:59:12.942 So… Miniland clearly pressed the up button, as indicated by his input display, 00:59:12.942 --> 00:59:20.350 but Red Ball just didn’t jump. Unfortunately, most flash games, including Red Ball, 00:59:20.350 --> 00:59:26.160 poll inputs in a flawed manner  that makes this possible to happen. If an input is   00:59:26.160 --> 00:59:33.280 pressed and released within the span of the same  frame, the game won’t register the input at all.   00:59:33.280 --> 00:59:40.600 This jump in SMB11 must be a single frame long, so  Miniland is aiming to press the button for around   00:59:40.600 --> 00:59:47.880 32ms to give him the best chance of having his  input pulled just a single time. However, this   00:59:47.880 --> 00:59:55.640 is extremely difficult to do, and even if you were  a robot that could press a button for 32ms every   00:59:55.640 --> 01:00:02.040 single time, it still wouldn’t be 100% consistent.  Because of all the frame time shenanigans that I   01:00:02.040 --> 01:00:09.920 mentioned earlier, a perfect 32ms input still has  a chance of being pulled for 0 frames or 2 frames.  01:00:09.920 --> 01:00:16.920 The same day, he got a run that hit all the frame  perfects in pl8-8 except for the last, placing him   01:00:16.920 --> 01:00:23.760 a second ahead of the record. Again, he had the  fastest ever exit from level 8 but failed 9x3.   01:00:24.840 --> 01:00:32.760 The next day, he finally managed to hit all of  pl8-8 in a run, but 9x3 eluded him yet again.  01:00:32.760 --> 01:00:37.040 For the rest of December, Miniland  kept pouring in attempts, steadily   01:00:37.040 --> 01:00:44.486 making progress towards the record. And, on the  28th, he had another opportunity at level 11.  01:01:10.560 --> 01:01:17.600 Miniland had a 3x choker attempt that looked  almost exactly like the real thing, but sadly,   01:01:17.600 --> 01:01:23.800 it was a counterfeit. After some investigating,  he found that this can only happen if the jump   01:01:23.800 --> 01:01:31.320 off the pillar is a single frame long, so he made  sure to always do a 2 or 3-framer in the future.  01:01:31.320 --> 01:01:37.640 Miniland continued to post steady progress through  February of the next year, even streaming on   01:01:37.640 --> 01:01:43.880 Niftski’s alternate Twitch account a few times.  While he had achieved various best exit records,   01:01:43.880 --> 01:01:51.000 he still hadn’t closed out a sub 2:40. He was  overdue for a massive pb, and he got confident   01:01:51.000 --> 01:01:58.640 enough to stream 2:35 attempts to his main Twitch  account. 2:35 was the upper limit of the route   01:01:58.640 --> 01:02:06.080 he was going for, which, I’ll remind you, was  Niftski’s 2:37 route with 9 more frame perfects   01:02:06.080 --> 01:02:13.920 added from pl8-8 and 9x3. It was incredibly  unlikely that Miniland would be able to jump   01:02:13.920 --> 01:02:20.040 straight to a time like that, but he was certainly  capable of achieving it a few personal bests down   01:02:20.040 --> 01:02:27.654 the road. Well, on his 3rd main channel  stream, he finally got his first sub 2:40.  01:03:17.200 --> 01:03:24.560 The man skipped straight to the 2:35, getting it  by a single frame. A quick reset at the start of   01:03:24.560 --> 01:03:30.640 8 to retry the left-right inputs was the only  mistake in the entire run. 2 corner jump,   01:03:30.640 --> 01:03:42.240 TAZ 3, Resetless 4, Star 6, Mug 7, pl8–8,  9x3, Resetless 10, SMB 11, and 3x Choker.   01:03:42.240 --> 01:03:49.280 It all came together. Funnily enough, the top  of the Red Ball leaderboard was now… almost   01:03:49.280 --> 01:03:55.880 identical to the top of the Super Mario Bros.  leaderboard. Their hostile takeover was now   01:03:55.880 --> 01:04:04.000 complete. With the reservoir of RTA viable setups  having finally dried up, Miniland’s record stood,   01:04:04.000 --> 01:04:13.640 and stood, and stood. Eventually, it became the  first world record since Flower’s 4:04 back in   01:04:13.640 --> 01:04:20.960 2016 to stand for over a year. Vertic continued  doing attempts from time to time, eventually   01:04:20.960 --> 01:04:28.120 getting a 2:37 and some 17 Levels records, but  the top two spots on the 12 Levels leaderboard   01:04:28.120 --> 01:04:34.840 still eluded him. Though top-level play was  expected to remain stagnant for even longer,   01:04:34.840 --> 01:04:42.000 as I have said many times in this video, Red Ball  has a unique ability to draw people back to the   01:04:42.000 --> 01:04:49.240 game, even those who have claimed to be completely  done with it. A small spark is often all it takes   01:04:49.240 --> 01:04:57.360 to reignite that passion, and for me, creating  the 3-minute milestone video was exactly that. 01:05:19.360 --> 01:05:25.280 For almost a month, I had been practicing the  game in secret, equipping myself with the tools   01:05:25.280 --> 01:05:31.160 that I needed to contest for the record. I kept  track of my consistency with each strategy on a   01:05:31.160 --> 01:05:35.960 spreadsheet, which highlighted my weak spots  and estimated how many attempts it would take   01:05:35.960 --> 01:05:41.760 me to beat the record. The only difference  between my route and Miniland’s was RBZ9,   01:05:41.760 --> 01:05:49.000 a new setup for level 9 that was a good bit easier  than 9x3 without losing any time. I had done a   01:05:49.000 --> 01:05:56.640 handful of attempts offline as a sanity check, and  I ended up getting an optimal 1-6 and 1-7 pretty   01:05:56.640 --> 01:06:02.960 quickly. I didn’t want to place too much pressure  on myself until I was truly ready to dive in,   01:06:02.960 --> 01:06:09.280 so the only person that I kept in the know about  what I was up to was Niftski. At this point,   01:06:09.280 --> 01:06:14.680 the two of us had become close friends, and I  started motivating him to return to the game as   01:06:14.680 --> 01:06:22.360 well during my preparation. And, on March 30th,  he sent me the exact message that I’d been hoping   01:06:22.360 --> 01:06:27.560 for. We agreed that some friendly competition  would be a great way to motivate each other,   01:06:27.560 --> 01:06:33.040 and I finally made my announcement. The  race to world record was officially on.  01:06:33.040 --> 01:06:40.080 With my personal best still being the famous  2:55, I had a lot of catching up to do. On day   01:06:40.080 --> 01:06:43.384 3 of attempts, I got my first sub 2:50. 01:06:46.267 --> 01:06:49.425 Though I had beaten my PB by 6 seconds, 01:06:49.440 --> 01:06:55.760 the pop off was almost non-existent because  I was still so far off of my goal. As I got   01:06:55.760 --> 01:07:02.400 distracted by school, my grind started tapering  off, while Niftski went full steam ahead. Though   01:07:02.400 --> 01:07:09.080 our routes were almost identical, he included some  additional frame perfects at the end of levels 7   01:07:09.080 --> 01:07:16.200 and 10 that saved around half a second combined,  giving himself 2:34 potential. Since he also went   01:07:16.200 --> 01:07:23.160 for optimal jumps in levels 1 and 5 as well, this  meant that every single piece of his route was   01:07:23.160 --> 01:07:29.600 entirely frame dependent except for the plunger  jump in level 9. Aside from any quick resets,   01:07:29.600 --> 01:07:36.320 it was essentially all or nothing. On April 8th,  he had a run with a second try left right switch   01:07:36.320 --> 01:07:42.400 in level 8, matching Miniland in that regard. His  additional optimizations meant that he was still   01:07:42.400 --> 01:07:48.320 on record pace, but after nailing the first  jump of resetless 10… he somehow missed the   01:07:48.320 --> 01:07:54.800 pillar. This came as no fault of his own; he  buffered right through the level transition,   01:07:54.800 --> 01:08:01.280 but the game didn’t pick up the input until frame  2, throwing off the setup. This has a chance of   01:08:01.280 --> 01:08:06.000 happening whenever the level transition  lands on one of those half frames that I   01:08:06.000 --> 01:08:12.080 mentioned earlier. Shoutouts again to Flash for  being a technological dumpster fire. Despite   01:08:12.080 --> 01:08:18.440 having to take a costly reset there, he still  finished the run a few frames ahead of his PB.  01:08:18.440 --> 01:08:25.560 The next day, he got a run that missed the 4th  frame perfect in pl8-8, causing a 1.3 second   01:08:25.560 --> 01:08:33.040 time loss. After a perfect levels 9, 10, and 11,  he barely messed up the start of 3x Choker and had   01:08:33.040 --> 01:08:34.150 to hit the backup.   01:08:38.213 --> 01:08:43.760 Since this was another small  PB, he was able to keep his composure and lock 01:08:43.760 --> 01:08:51.360 in on 17 levels. At this point, the world record  in that category was a 4:04, exceptionally close   01:08:51.360 --> 01:08:57.320 to that 4 minute milestone. This was the final  minute barrier that the game had to offer, and   01:08:57.320 --> 01:09:04.561 since Niftski’s level 12 split was over 7 seconds  ahead of the record, he was easily on pace for it. 01:09:24.680 --> 01:09:29.520 The missed jump from the moving  platform in level 13 killed the sub 4,   01:09:29.520 --> 01:09:37.640 but he still set a new 17 Levels world record of  4:01.387. Knowing that a 12 Levels world record   01:09:37.640 --> 01:09:43.320 could happen any run now, he began streaming  attempts to his main Twitch account the next   01:09:43.320 --> 01:09:49.440 day. While Niftski demonstrated incredible  consistency on his first and second streams,   01:09:49.440 --> 01:09:54.800 he wasn’t quite able to put together the  run that he was looking for. However,   01:09:54.800 --> 01:10:01.418 perfection can only elude this man for  so long. Here’s what happened on day 3. 01:11:07.280 --> 01:11:15.560 While Niftski’s 2:37.710 had come extremely close  to perfection for that route, this masterpiece was   01:11:15.560 --> 01:11:22.960 the real deal. You could not beat this run by  more than a few frames without faster setups.   01:11:22.960 --> 01:11:30.120 Niftski got the fastest ever exit from level  8, and he never looked back. This route had   01:11:30.120 --> 01:11:37.920 over 35 frame perfects, and he hit every single  one first try. The quest to perfect a full-game   01:11:37.920 --> 01:11:44.560 route developed from the TAS hack had concluded.  Knowing that it was impossible for me to beat this   01:11:44.560 --> 01:11:50.760 run with my current strategies, I threw in the  towel on the most lopsided world record battle   01:11:50.760 --> 01:11:57.000 in Red Ball history. Though this run could  have easily demolished the 17 levels record,   01:11:57.000 --> 01:12:01.720 he understandably struggled to keep his  composure through that pop off. There’s   01:12:01.720 --> 01:12:08.534 a reason that the last time that someone  got both records in the same run was Jonas’  01:12:10.000 --> 01:12:15.520 back in 2021. Niftski had no interest in running 17 levels   01:12:15.520 --> 01:12:22.440 on its own, meaning that the door to claim the  first sub 4 was still wide open. So, that’s the   01:12:22.440 --> 01:12:30.120 goal that I shifted my gears towards. On the road  there, I got my first 3x in 12 levels, I scored   01:12:30.120 --> 01:12:39.200 3rd place with a 2:36, I set the 17 levels record  with a 4 flat, and I choked a sub 4 to the final   01:12:39.200 --> 01:12:47.400 frame-dependent input in the run. On May 5th, the  legend rare.eg had tuned into my stream for the   01:12:47.400 --> 01:12:55.120 first time since my return to the game. Fittingly,  four runs later, I got my consolation prize.  01:13:03.200 --> 01:13:10.440 While my story had certainly found a happy ending,  Vertic had yet to find his. The game had cheated   01:13:10.440 --> 01:13:17.600 him time and time again, but he was persistent,  still pouring in runs whenever he had the time. At   01:13:17.600 --> 01:13:24.480 this point, all he wanted was to beat Miniland’s  time in 12 levels, a goal that had eluded him   01:13:24.480 --> 01:13:29.034 almost a year and a half. Well, on July 29th… 01:13:39.018 --> 01:13:41.950 That’s not quite what he was after, but beating  01:13:41.950 --> 01:13:50.320 my 17 levels PB by a single frame was certainly  welcome. Now, here’s what happened on August 6th.  01:14:09.480 --> 01:14:16.200 A sizable 17 levels record… this man was  on a roll. He just needed to translate   01:14:16.200 --> 01:14:22.866 that success to the first 12 levels as  well. Thankfully, the very next day…  01:14:44.880 --> 01:14:53.200 He achieved his goal by a single frame. After a  year and a half of just barely coming up short,   01:14:53.200 --> 01:14:57.120 Vertic was completely overcome with emotion. His   01:14:57.120 --> 01:15:02.680 journey was over. Vertic was  finally free from Red Ball. 01:15:02.680 --> 01:15:13.080 And, that leaves Niftski’s 2:34.613 in 12 Levels  and Vertic’s 3:57.452 in 17 Levels as the current   01:15:13.080 --> 01:15:19.920 world records in the game. It took over 4 hours  of scripted commentary to get here, but we have   01:15:19.920 --> 01:15:26.160 finally arrived at where the game may remain for  a good while. However, that doesn’t mean that this   01:15:26.160 --> 01:15:32.230 is the end of Red Ball’s story. Though sub 2:40  was widely believed to be the game’s final 10   01:15:32.230 --> 01:15:38.280 second barrier for a few years, we have thankfully  learned enough about Red Ball to know that there   01:15:38.280 --> 01:15:45.600 is still more to be done. Sub 2:30 is possible,  and I wholeheartedly believe that an extremely   01:15:45.600 --> 01:15:52.360 dedicated speedrunner will accomplish it one day.  However, to give you a sense of how far off we   01:15:52.360 --> 01:15:58.507 still are, let me show you one of the strategies  that you’d need to implement to make it happen.  01:16:04.680 --> 01:16:12.040 With 18 frame perfects, you can escape the opening  tunnel early and complete the level almost 3   01:16:12.040 --> 01:16:18.560 seconds faster than SMB11. The movement at the  beginning of the level is so precise that you   01:16:18.560 --> 01:16:25.000 have to keep the game at a pixel-precise window  size to make it work. What you witnessed there   01:16:25.000 --> 01:16:31.240 is just a taste of the mastery that speedrunners  have managed to achieve on an individual level   01:16:31.240 --> 01:16:37.480 basis. Many other crazy strategies that were  once thought to be impossible in a real-time   01:16:37.480 --> 01:16:44.360 setting have been pulled off, such as this insane  death warp in level 9 that involves inverting the   01:16:44.360 --> 01:16:51.000 plunger to get it in just the right position.  The sum of every individual level world record,   01:16:51.000 --> 01:17:00.320 the combination of thousands of hours of effort,  comes out to 2:20.065 for 12 Levels and 3:29.516   01:17:00.320 --> 01:17:09.400 for 17 Levels. The Tool-Assisted speedrun clocks  in just 13 and 14 frames faster respectively. 12   01:17:09.400 --> 01:17:15.680 out of 17 levels in the TAS have been tied to  the frame, and there’s a good chance that the   01:17:15.680 --> 01:17:21.760 human sum of best will come to match it one day. While the crazy accomplishments people have made   01:17:21.760 --> 01:17:27.400 from an individual level standpoint are extremely  impressive in their own right, it doesn’t   01:17:27.400 --> 01:17:33.560 necessarily translate over to full game success.  Attempting a single level of the game tens of   01:17:33.560 --> 01:17:39.040 thousands of times is completely different from  tackling the whole thing, and there are plenty of   01:17:39.040 --> 01:17:45.880 runners that are great at one but not the other.  If you look at the 10 best individual level   01:17:45.880 --> 01:17:52.920 players and the 10 best full-game players, only 5  make an appearance on both lists. While the human   01:17:52.920 --> 01:17:59.520 sum of best certainly proves that the game still  has a long theoretical runway, it simultaneously   01:17:59.520 --> 01:18:06.280 shows that a sub 2:20 is not possible, at  least using the versions of Flash Player that   01:18:06.280 --> 01:18:12.440 we currently have available. Let’s generously  assume that Flash’s inconsistent frame times   01:18:12.440 --> 01:18:20.400 result in 5% of otherwise successful frame perfect  inputs failing. To get a sub 2:20, you need to hit   01:18:20.400 --> 01:18:27.920 over 190 in a row. Even if all of those inputs  would have worked given consistent frame times,   01:18:27.920 --> 01:18:37.800 you’d only have a 0.006% chance of them all  succeeding. Thankfully, some recent developments   01:18:37.800 --> 01:18:43.560 have helped improve consistency and level the  playing field. Firstly, I discovered that the   01:18:43.560 --> 01:18:49.680 minimum keypress length of certain keyboards  could be modified. You see, all keyboards have   01:18:49.680 --> 01:18:56.040 a built-in debounce time to ensure that single  keypresses aren’t registered as multiple,   01:18:56.040 --> 01:19:04.520 usually defaulting to 5ms. By changing it to Red  Ball’s median frame time, 33ms, it makes 1 frame   01:19:04.520 --> 01:19:11.360 inputs a lot more consistent. Then, a few months  ago, I confirmed that some older Flash player   01:19:11.360 --> 01:19:18.040 versions yield more consistent frame times in the  Red Ball games compared to newer ones, especially   01:19:18.040 --> 01:19:24.800 on slower machines. I spent days tracking down  the optimal version, which came in the form of   01:19:24.800 --> 01:19:31.320 an ancient internet plugin, and then developed  my own program to use it without having to boot   01:19:31.320 --> 01:19:38.000 up Internet Explorer. As a bonus, I was able to  add custom features that the stock Flash Player   01:19:38.000 --> 01:19:45.280 lacked, making it a win-win. Finally, we’ve also  found a way to prevent 2nd frame accelerations,   01:19:45.280 --> 01:19:51.360 which have killed many world record paced runs,  from ever happening again by shortening the key   01:19:51.360 --> 01:19:58.360 repeat delay below the normal range in Windows. These recent developments have been beneficial to   01:19:58.360 --> 01:20:04.640 Red Ball and the flash game speedrunning scene  as a whole. We can only hope that, one day,   01:20:04.640 --> 01:20:12.040 we will have a way to play a 15-year-old,  600kb game without performance issues to   01:20:12.040 --> 01:20:19.000 push its potential as low as possible. But,  for now, this is where our story ends. From   01:20:19.000 --> 01:20:26.080 casual playthroughs, to initial optimization,  to movement mastery, to the quest for sub 3,   01:20:26.080 --> 01:20:31.840 to the entrance of frame-dependent setups,  and finally, the quest for perfection,   01:20:31.840 --> 01:20:37.920 this game has certainly been through a lot. Of  all the tens of thousands of Flash games that   01:20:37.920 --> 01:20:43.720 made their way to the internet, who could’ve known  that this is the one that would arguably hit it   01:20:43.720 --> 01:20:50.560 the biggest from a modern speedrunning standpoint.  Let’s just hope that there is still more history   01:20:50.560 --> 01:20:56.350 yet to be written. Thanks for following  the complete story of Red Ball speedruns.