I just noticed a few things wrong with Yuuno's run
4 years ago
Portugal

All of the screen transitions are gone? It looks like the game is skipping all screen transitions i.e. fade to black, fade-in, etc. is this a glitch in the emulator? The game's not supposed to look like that. With the transitions disabled, the game is naturally advancing much faster than it should between rounds.

With the help of a friend, we noticed a few more oddities with the run.

The CPU is quite lenient throughout the run and this can be noted in the first match, where Sean is effectively doing nothing, which makes me think the difficulty is set to 1. Throughout the run, there are other examples of the CPU doing nothing or being overly docile and this is incredibly lucky even for lvl3. You can run a short experiment yourself, load up Sean at level 1, do the whole match, then do the same at level 3, you'll see what I mean. Basically, you cannot get this lucky with a single character, let alone an entire run.

Even Dudley is frequently just walking forward and doing nothing for seconds, only a lvl1 CPU would do that, while lvl3 would do Uppercuts the second you're in range.

It's also far too convenient that his setupless jump-ins pretty much never got blocked.

At 2:32, Akuma's portrait/screen transition plays twice. It clearly looks like it was messed up during editing. Akuma's avatar also moves back (to the left) for a few frames. This weird "glitch" occurs later after the match against Dudley (3:46), so I'm unclear whether this is simply a glitch, but that doesn't explain why all screen transitions are seemingly disabled.

The INSERT COIN/JOIN IN text also vanishes for a few frames sometimes. Sometimes it lingers longer than it should. This simply should never happen. The text should keep blinking at the same interval regardless of anything and should only disappear during transitions (which, as clearly shown, were disabled anyway).

At least once, the characters vanished from the screen. https://steamuserimages-a.akamaihd.net/ugc/786375888692101066/9E3C14205C478561030F8B9A8979A974ADF1A524/

Basically, the run looks fake as hell, and we're also talking about a player who fails Shoryuken inputs frequently and even started the SUV bonus game by high jumping when attempting to do KKZ.

There may be more oddities, but the fact that the run is as glitched as it is (and that the person playing is fairly inept) makes it hard to tell which "glitches" shown come from the screen transitions being disabled, and whether anything in the video was edited.

Regardless, the inacurately and overly docile CPU is the biggest hint (seriously, you have to be lucky in every round, the whole time for this to happen even at lvl1) but the simple fact the game doesn't look the way it should makes me and my friend believe it should be taken down from the leaderboards.

New South Wales, Australia

On default, Level 1-7 The CPU IS generally very passive and un-responsive. That's actually pretty normal. I can get free 92's on Level 1 Sean using different patterns.

Jump In punches/kicks DONT get blocked, its normal. Check my run, I used the same strategy.

At 2:32 his portrait doesn't play twice, its the same transition.. the "black bars" shows up late due to the visual glitch in his emulator (as covered in the other post).

The INSERT COIN message, I have not noticed. But I would imagine it has something to do with the same visual glitch responsible for the missing black bars transition.

The Characters vanishing from screen was because it was transitioning from round 1 to round 2, this is what it looks like if the black bars dont appear .. this is what the black bars hide its why screen transitions exist in video games, to hide re-drawing a new scene. It re-drew the scene with background first, then foreground objects (on the next frame).

You make the assertion that the run is fake, but you have also not specified "how it was fake", just a few things that bother you about it.

You're wrong about the passive AI, I know this from experience. Are you running the same revision?

On Japan 990512 Default AI doesnt ramp up the difficulty until fight 8/9 .. in which case mostly 9 - the RYU fight (for Akuma) is the biggest pain in the run. There is some RNG to getting good perfect rounds, but the AI is in fact a punching bag right up until fight 8/9. I was able to reproduce the Ryu strat he used but I found a better one with the jump-over projectiles to use for my run. As for Gill, his wake up Shoryuken tech is the best method I could find and its the same strat I used.

Edited by the author 4 years ago
New South Wales, Australia

That last part:

"Basically, the run looks fake as hell, and we're also talking about a player who fails Shoryuken inputs frequently and even started the SUV bonus game by high jumping when attempting to do KKZ."

How do you even know he fails at Shoryuken inputs frequently? ... Do you know him?

Portugal

Let me answer your question with another question: do YOU know him? Can you prove for a fact that everything I pointed out is legit? Furthermore, how can you accept a glitched video? This is the first time I've ever seen any footage of the game glitched as in this video, but the one time I do, it's coincidentally in the world record speedrun?

I don't feel like I need to prove my experience with the game any further, everything I said could've only come from experience and even if the glitched transitions didn't give any advantage, it doesn't matter, it doesn't look normal, and the CPU behavior is unlike any that I've ever seen. Even in your run, #2 on the leaderboards, you can actually see, clear as day, the blatant difference in CPU behavior. Your game looks correct. His does not, and I can only assume the glitchy graphics are proof of something fishy going on as well.

As for spamming shoryukens on Gill, this is a known strat, everyone's known about it for years. But the entire run is just too lucky for lvl3 cpu, regardless of revision.

But let's say you're right. Let's say I'm wrong about the cpu behavior (which I'm not).

Why is the footage glitched? Is it a well-known glitch? Is it common? Does it occur in any known version of the game, be it arcade, consoles or modern ports? If any or all of those questions are answered with "no" then the video shouldn't be allowed.

New South Wales, Australia

You don't feel you need to prove your experience any further .. lol .. you come here .. 0 runs to your name and we're supposed to take you at your word. If you feel like you don't need to prove your experience with the game, then I feel like I don't have to listen to your opinion on the matter.

You make a claim that he fails Shoryuken inputs which appears to be false because he nailed every single one of his wake-up Shoryus on Gill and Ryu. You make claims that his AI is too passive which is again false, hell on my runs its even MORE passive lol. How is the AI any different in my run? .. I exploited the jump in punch/kicks BETTER than he did, but the the jump in Punch/Kick exploit works ... proven by my run.

It's simple, you get yourself mid distance to the opponent, jump in and do a late punch or kick.. it connects like 90% of the time up to fight 8 and you just combo and capitalize on it. You need to fade some block RNG sometimes but it works.

The glitch is from the emulator clearly, and it is visual only and doesn't effect gameplay or speed as PROVEN by my analysis in the other thread. You can't sit here and claim its fake one moment and then just claim that the glitch makes it invalid the next, which is it? .. is it fake? or should it not be on the board because of a glitch? because those are 2 different discussions to be had.

Edited by the author 4 years ago
Portugal

I tried to be reasonable, but if you'd rather use personal attacks rather than stick to refuting my points, then I'd rather not respond to your insults. It's fair you don't want to accept my experience with the game. you don't know me and I haven't submitted any runs. I don't plan to, either. I made my point as best I could strictly with words.

As for my comment on his failed SRK inputs, I'll admit I may have exaggerated a bit, but it's true he failed to follow up bnb's with srk numerous times, already in the first match. I didn't say it to insult the player, I raised it as a fact, though the way I worded it may have sounded offensive and exaggerated.

Anyway, I'm glad that you've agreed to disallow runs with any visual modifications or glitches. It's particularly strange, though, this should've been an obvious rule from the start. Any modifications whatsoever, cosmetic or not, will always put the legitimacy of any run.

I consider the possibility that maybe, just maybe, I'm imagining an overly lenient CPU where there isn't, but even after what you've said (I don't doubt your experience), I still can't shake the feeling Yuuno's run has something wrong going on besides the glitchiness.

Edited by the author 4 years ago
New South Wales, Australia

I apologize. I admit, I did get a little heated.

I will say that I am very confident that the CPU is in fact just as exploitable as it is in the current WR holders run. I'm not sure exactly what you've tested and how you've tested it. It may only be with Akuma, it may only be on JAP 990512 No CD. It also does depend on which opponent you are facing, theres a few that you specifically want to avoid because they are harder to exploit (Oro, Yun/Yang, Ken, Remi come to mind).

But in general, Yuuno's run simply exploited jump in punches/kicks to chain into shoryuken/tatsu/hadoken as the main strategy. I took this and changed it to something much simpler (jump in punch > MK > sweep - for the most part, but with some variation sometimes). If done right you can get it very consistently .. maybe not 100% .. but upwards of 80-90%, sometimes u do derp out on round 1/2 and reset, but if you persist and grind enough you can get it throughout the entire run (up to fight 8, as mentioned previously).

That's the assessment I made when I originally investigated, which also prompted me to challenge the WR myself. I got really close with relative ease (2 sessions of about 3-4 hours grind each) and was satisified that the strats were doable and the run was reasonable. Don't get me wrong, I'll be back to beat it soon .. (hell all this comotion definitely lit the fire once again). It's just that I had satisfied my own confidence in the run to leave it be for a while.

Edit: I should also clarify that at this point in time that I had analyzed Yuuno's video extensively already and debunked the claim that the missing transitions were "evidence of splicing" as seen in my video comparison which clearly shows no time was cut and no frames were lost. This in conjuction with me studying the strats he used lead me to conclude that the run was legitimate.

My issue with removing the time comes from the basis that: A. I am confident it is not cheated for aforementioned reasons. and B. Removing it on such a light technicality just seems really dirty. Especially since it gives me the "fake WR" as a result. The board is still young, the current WR run is destined to be obselete in the not too distant future anyway. Having the new rule in place should keep things clean moving forward and hopefully keep everyone happy.

We may not see eye to eye on some priorities but I do recognize that as a moderator its my duty to uphold the communities interest and try to protect the integrity of the leaderboard. I just think that a lot of people tend to underestimate the damage that is done by removing a WR time from the leaderboard needlessly, and thus its something I do not take lightly.

Anyway, again .. I apologize for the remarks I made in my previous post, and hope we can move forward from here.

Edited by the author 4 years ago
Piedmont, Italy

Can't we at least strongly suggest (if not enforce) the usage of WolfMAME and inp files for the emulated arcade versions? They are not, like everything else, 100% foolproof, but they are very hard to tamper with, surely beyond the average user capabilities, and they are used since forever in the arcade community to validate scores.

Edited by the author 4 years ago
NerdyNester, Peanutfan22, and Tenka like this
New South Wales, Australia

@Jck This is something we are definitely considering. At the very least, it's something every Arcade runner should be doing by default if they want to really protect their runs from scrutiny (not every moderator out there is as lenient as we are lol).

If we did enforce it, we might employ it as we have done over at the Mortal Kombat boards, where it is only required under a certain time threshold, so as to not put too much unduly pressure on new inexperienced runners. Sometimes newer runners can often struggle to get their head around the INP recording process.

Frankly though, I don't know if we are even at a point where the run is optimized enough where we can determine an appropriate threshold.

At the very least, I do highly recommend everyone try to get familiar with WolfMame and the INP recordings, incase it becomes a thing we enforce in the future.

NerdyNester, Peanutfan22, and Jck like this
New South Wales, Australia

Just to add to this, here is the guide I wrote for recording a WolfMame INP for Mortal Kombat Arcade: https://www.speedrun.com/mk1/guide/euoqg

NerdyNester and Peanutfan22 like this
United States

We should just be using 30th anniversary as the standard. Since it's a direct port and emulation actually increases the speed of the game slightly. I'm gonna miss my color mods :(

New South Wales, Australia

30th Anniversary has its own category (under misc) as set up by Nerdy years ago when he first set up the categories. I haven't really investigated and compared differences between the versions though, I'm unaware of how different it actually is (maybe it should be considered same as arcade?)

I don't believe you are entirely correct about emulators though. They don't increase the speed of the game, they increase the framerate. OG Arcade hardware runs at 53 fps (roughly) and emulators bump that up to 60 fps. However the way they do this is not to simply redraw every frame faster ... instead they actually draw more frames to compensate. This produces technically more "smooth animation" albeit not really noticeable to the naked eye, but it keeps the game speed intact and accurate to the original hardware.

At least this has always been the case with other arcade games emulated.. I was unaware of there being any specific inaccuracy in SF3.

Unless there is something specifically wrong with SF3's emulation I don't see us banning emulation, sorry .. but if you pay attention to the above posts about WolfMame. It's a very good alternative to curbing fears of illegitimacy.

Edited by the author 4 years ago
United States

I might have to do a side by side comparison test with wolfmame and 30th just to see. But after playing for 10 years I notice a big difference between emulation, and actual arcade/ports. It could be also in my head but I swear this was the consensus on fightcade

New South Wales, Australia

Well .. I did a bit of searching and found this:

According to this comparison Retroarch with FBA core (should be the same thing Fightcade runs on) is actually almost dead accurate. Online Edition does appear to be slower .. so that may be where the difference lies - it could be that the port iself is innacurate to real hardware.

ShmupMame is also innacurate, but actually slower - not faster. However ShmupMame is notorious for having speed issues, nobody uses it in speedrunning AFAIK. The developers for the main MAME cores are usually pretty anal about speed accuracy, so I can't really imagine it being innacurate, unless the Emulation project was left incomplete or something .. but I doubt that is going to be the case for something as high profile as SF3.

Edit: I just noticed that he had to specify a specific setting in Retroarch. Hmm this may actually require some further investigation after all. I'll do some more thorough digging when I get home from work later.

Edited by the author 4 years ago
NerdyNester and Tekrat like this
United States

30th Anniversary really might be the closest thing we have to an arcade port. It is different from OE which actually had some gameplay changes included. Unless we all want to get some CPS3, supergun boards and go ham on real hardware.

Tenka and NerdyNester like this
United States

Found this article http://shoryuken.com/2018/05/29/the-street-fighter-30th-anniversary-collection-delivers-30-years-worth-of-arcade-history-fantastic-art-and-spectacular-gameplay/ and few other sites that say 30th is close to arcade perfection. The only way we could get perfection is if we had actual boards. So I feel emulation and 30th is the closest we are going to get. gonna talk with other mods but most likely will be removing the 30th category and merging it with the arcade board. Any runs will be transfer as well. In a perfect world we all could have cps3 boards and doing runs but sadly that will never be a thing except for the lucky few who can access it.

Edited by the author 4 years ago
Tenka and Tekrat like this
United States

@NerdyNester I would like for it to be mixed together. I usually play on 30th to begin with, it just feels right to me. Im working on setting up mame right now to test it however.

I mean we can all get the supergun boards if you are willing to fork out $3000 D:

NerdyNester and Tenka like this
New South Wales, Australia

Makes sense to me, if its arcade perfect port, might as well be in with arcade.

New South Wales, Australia

So, after a lot of digging these are my results on the emulator speed issue: https://www.speedrun.com/sf3ts/thread/jaiqw

Please move all discussion over to that thread.

I am locking this thread now since the matter has been discussed/resolved and it has since gone a little off-topic.

Edited by the author 4 years ago
NerdyNester likes this