Using the same methods I used to catch sir_farcelot to analyze DapperScratcher, and I found quite a bit.
-
She has submitted 4 TASed runs to speedrun.com:
-
Her full game run appears to be sped up by x1.01 as seen here:
I did the math, and the time before it was sped up would be about 3:53.84. This is a 1% speed up. This typically doesn’t happen.
- She hasn’t submitted many runs at all, so there is no way for me to know how skilled she is. This would normally be fine, but 3:51 is an insane time that would take many hours of grinding to achieve, and there is no footage of the grind other than her personal best? This is ridiculous.
- Back in early 2024 basically every run was accepted because chicken was the only mod and they don’t seem to analyze runs very well, but now, I’m a mod, and I hate it when people make a TAS or splice together a run because that completely undermines the competitiveness of speedrunning.
- Somehow, DapperScratcher has managed to get all of her good runs on the first try, or at least she has never improved on a run she has submitted, and she has no obsolete runs. This is not how speedrunning works. You grind to improve and slowly work your way to the top. I have been grinding full game for years and have still not managed to get below 3:55, and you are telling me that DapperScratcher thought that her 3:51 was the first time worth submitting? This is insane, even a 4:0x is a good time, and there is now way you can get a 3:51 before getting any other good times.
- When you take a look at DapperScratcher’s runs, you see two clear types, some runs look almost perfect, and some runs look slightly above average. This is odd because if you can get one second away from the world record for one of the most competitive categories in Appel, you would think that you would submit more good runs and improve your bad runs, but DapperScratcher did not do this.
- DapperScratcher is very capable when it comes to TASing, she has many projects on her scratch profile that are just TASes. This means that she is very capable of TASing a full game run.
In conclusion, Dapper_Scratcher has cheated in the past, hasn’t shown enough game play at the skill level she claims to be at, and may have sped up her full game run, so we have decided to remove her run from the leaderboard.
Note that LionCoder and BringUpYourPost helped me find evidence.
The other day, I was messing around with Appel’s replay system and graphing random runs in Appel when I came across something odd. All of the Level 5 runs looked similar except for Sir Farcelot’s. His inputs were very robotic and unnecessarily precise, with every sidejump being identical down to the frame. This level of precision is not necessary to tie the TAS on the human route for Level 5. This oddity led me to investigate his runs further, resulting in the following findings.
Findings:
- On his Level 5 personal best (PB), he presses and releases the right key while sidejumping, despite it being optimal to hold the key down. This is not definitive proof, but it is an odd deviation from standard techniques.
- I graphed the time intervals between keypress changes and observed a perfectly repeating pattern throughout the sidejumping sequence. Other top players (except Michael Chan) exhibited natural variations in their inputs. This strongly suggests a scripted TAS loop.
- In nearly all his runs, the replay releases all keys after completing the level. This is not possible without modifying the replay system. Normally, the keys that you were pressing when you finished the level are the last recorded keypress in the replay code, but Sir Farcelot’s replays let go of the keys after the end.
- While not direct evidence, Sir Farcelot has experience with TASing in the game Appel, meaning he possesses the necessary skill to create a convincing TAS run.
- He has only submitted any% runs, which may indicate that other categories are too tedious to TAS, though this is not conclusive evidence.
- He acknowledges that his sidejumping method is unusual and claims that he thought it was optimal when he TASed the level, so he implemented it. However, it isn’t faster, so this doesn’t make sense.
- He says that he used pause buffering to achieve his perfect sidejumping. However, I believe that would be just as difficult and as unnecessary, so it doesn't change much.
Based on the evidence, I remain convinced that Sir Farcelot used TAS methods in his runs. The robotic input patterns, impossible replay behavior, and lack of natural variance strongly indicate external assistance. If he wishes to clear his name, I have requested that he submit a video demonstrating his technique under controlled conditions. Until then, I believe his runs should be scrutinized further.
I feel like the minimum apples category is outdated and was properly replaced by the category extensions category. Should it be archived or removed?
Should I add a patch (under the resources tab) that instantly presses the continue button for you that could be used for full game?
I'll get straight to the point: Michael Chan cheated his full game world records and I have proof.
Proof One: He Has a History of Cheating
Michael Chan has been known for cheating runs for a while now. The first fun he cheated was his 4:07.95 in this video is this frame which pretty much sums up how this run was cheated:
Proof Two: The Real Evidence
In order to understand why I'm concerned you will have to learn about how the level transitions in Appel work. There are two types of transitions Appel can play between levels. One happens after you enter the level form the level select and one happens when you enter from the previous level by pressing "Continue."
Continue:
Level Select:
Notice how the "Level 4" text fades out in the one that comes from the level select but doesn't on the other one? Well guess what happens in Michael's 3:59 and his 3:50? Yep, they fade out meaning they were spliced. Have a look for yourself:
There should be a 50% category. You would have to get half or one more of the yellow apples if there is an odd amount.
Lukex & I made a version of Appel that has a better timer and I was wondering if it would be better to use for runs.
Here is the link: scratch.mit.edu/projects/888364503/
