tauKhan
| SSBM btt mismatch TAS Yoshi / Puff 10.60 [wr] @Peikotus91 | Uploaded April 2024 | Updated October 2024, 1 hour ago.
Not optimized much.
🔇 TAP TO UNMUTE
SSBM btt mismatch TAS Yoshi / Puff 10.60 [wr] @Peikotus91
![SSBM btt mismatch TAS Yoshi / Puff 10.60 [wr]](https://i.ytimg.com/vi/oNfHJC2Vab8/hqdefault.jpg "SSBM btt mismatch TAS Yoshi / Puff 10.60 [wr]
Not optimized much.")
![SSBM Break The Targets [TAS]: Peach [7.00] [WR]](https://i.ytimg.com/vi/uRKafd2qvc0/hqdefault.jpg "SSBM Break The Targets [TAS]: Peach [7.00] [WR]
The perfect 1:25 000 000 000 odds Peach complete is finally here! With starting seed 0x9CF5BC17, Peach TAS wr once again matches best time with forced item pulls.
Previous record(+5 frames): https://youtu.be/f4e2vMrtPX0
Best time with previous route (by PracticalTAS)(+42 frames): https://www.youtube.com/watch?v=g5HEmEVlZWk
Since everything is mostly same as with the previous record, except seed is now better and allowed uncompromised ending, the description is still relevant:
The route requires a non-item turnip pull, followed by 3 consecutive bomb pulls afterwards. The 4th pull could also be a sword. Additionally, theres 5/12 chance that the 4th item drops from Peach getting hit by the explosion on target 7. The naive chance to hit all those events with a set of inputs, starting from a random seed is therefore 127/128 * (1/384)^3 * 1/256 * 7/12, which amounts to roughly 1 in 25 000 000 000 probability.
Melee uses a simple LCG with a single 32-bit seed for all its RNG events, which is easy to emulate. Unfortunately though, the number of rng events occuring throughout a run is not constant, but depends on the random numbers generated for particular events in a chaotic manner. Because of that we cant yet fully predict exactly how RNG events progress. Additionally, the total number of available starting seeds is only 2^32 = 4 294 967 296, so for any given set of inputs, its very unlikely there even exists a starting seed that would complete the route.
To get around the limitation on number of starting seeds, I prepared slight alterations to run, that have no effect on finish time, yet they change rng events of the run. That essentially let me multiply the number of available starting seeds. To this date, I have 150 individual runs with different RNG, that would all yield 7.00 on perfect match.
I wrote python script to find seeds that would have highest likelyhood of succeeding on the pulls for different runs individually. As while I cant exactly predict the number of rng events happening during a run, each run still has a reasonably predictable distribution for number of iterations occurring between each pull. For each run, I filtered into around 10 000 most likely seed candidates based on the distance of bomb pulls with the seed compared to the average number of iterations occurring with the run. I then used another script to go through a list of candidates for a run by having the script load a savestate and then swap starting seed, then play the run. Rinse and repeat with next candidate until the list of candidates was exhausted.
Time saves:
Largest time save was of course the damage boost from exploding target 7. The events around target 7 are intricate: First at 5.55 Peach z-drops her bomb, then NILs on the ground below. Still on the same frame, the bomb hits and breaks target 7. Then on next frame, Peach is actionable as she just NILd, and immediately pulls another bomb. Then later that frame, the previous bomb explosion from hitting target 7 finally connects on Peach, removing the lag from the new bomb pulling animation entirely. So effectively, Peach gets to do SDI from hitlag, and damage boost from explosion knockback, while keeping a bomb for target 9 in hand. That saves enormous amount of frames compared to previous TAS.
At the beginning, a falling turnip pull followed by double jump into turnip throw for target 3 saves around 15 frames. This is mostly because double jumping before turnip throw has Peach ascend during the throw, instead of falling down very low. With previous TAS route it would take a lot more time to dj back up after the throw. The airdodge into edge-cancel following turnip throw also uses double sweetspotted edge-cancel, tech that I debuted in my Ness TAS wr. In this run, the edge-cancel is ridiculously precise and only a single airdodge angle, of the available 6456 different angles, actually works. Peach slips off the edge after landing only by a margin of 0.00009 ingame units. For reference, a phantom hit occurs in Melee if hitbox overlap is less than 0.01 units, for instance.
After first bomb pull, a slightly delayed airdodge cancelled bomb throw was used. The delay was done because of how ECB bottom offset fluctuates during fall animation after the throw: With later throw, Peach is at earlier fall animation frame when getting just above the platform, and is actually able to NIL while still moving slightly upwards due to ECB bottom expanding down. This saved 1 frame over the previous TAS wr.")
![SSBM Break the Targets[TAS]: Kirby [10.93][Former WR]](https://i.ytimg.com/vi/ynkivW12Quw/hqdefault.jpg "SSBM Break the Targets[TAS]: Kirby [10.93][Former WR]
Follows the route of previous record by AJPanton. Saves 1 frame by wavelanding onto the right stage section with an earlier airdodge.
Done on PAL.")
