XD: Mismatch of reroll's seed between hardware and program about gc-pokemon-rng-manipulation-assistant

[aldelaro5]

Ok so after talking with @ryziken over discord, a lot has been found about this issue and absolutely everything about it is confusing.

Let's just focus on the case number 1 mentioned in the issue: 6516269F. this has the program give the correct target on the frame number 479 after 102 rerolls. However, what was found instead is the same frame number was hit after 94 rerolls and not 102 despite the fact the program seemed to report the correct information tested with Dolphin.

Things gets much more complicated once we look at the recording of the procedure done on this case:

https://www.twitch.tv/videos/403880031

This videos first shows the seed finding procedure which went well and then a series of exactly 102 rerolls, the rerolls starts when the pacing of the cancel -> go back are getting fast. After many checks, it turns out the first 48 teams exactly matches what dolphin reports, but on the 49th one, this happens:

As you can see on the lower left, the Entei's HP stat doesn't match, but everything else is correct. This would normally point to a problem with the EV generation function, but the problem is that the behavior between hardware and Dolphin mismatch which suggest something is wrong with Dolphin, or something happened with Ryziken's game that made it skew the result a bit, it's hard to tell at this point.

I tried as suggested by @JMC47 to use the interpreter after updating my version of Dolphin to the latest master as well as many different accuracy settings and none of them made it change behavior: it was always generating 382. At this point, further hardware testing is required to know if Dolphin was wrong or some special circumstances I am not aware of happened before the procedure started.

This would be a very weird issue because as far as Dolphin is concerned, the program is reproducing everything correctly. Also, it goes without saying, but on the 50th reroll and onward, the teams are radically different which is what ultimately caused the mismatch at the end, but it's very strange why they would happen to be offset by a certain amount of reroll and be very close to the target frame.

Just in case it helps, here's what Ryziken had to say before this was recorded:

powered on the console, let it play through the pre-title screen cutscene while I set the stream up, and then just skipped to title screen and went straight to vs mode

In theory, this shouldn't change anything: no RNG calls are done on any of these events except the usual ones that happens from boot to the pokemon company logo.

The seed before the erroneous team generation is E4B74C47. Also, we know the revision of the game between me and Ryziken are the same.

[aldelaro5]

Update, this is footage from my Wii as I applied a gecko code that would set the seed to E4B74C47 when pressing R:

So this proves Dolphin isn't the problem, but then what is it?

....I am very confused so I'll try to modify the code to set the seed to 6516269F and do everything @ryziken claims to have done.

The code btw:

28444B2C 00000020
044E8610 XXXXXXXX

Replace the X by the desired seed. Now, the seed will be set when you press the R button so I pressed it when my cursor was on "ultimate".

EDIT: tried again, but this time, I used the seed returned by the finder and actually did the 102 rerolls while checking the 49th one. I also performed the same steps @ryziken said to have done and again, I cannot reproduce what is in the vod, I get 382 on the 49th and the last team matches what the program tells me.