Thanks to everyone for their input. Since nobody has come forward with any objections, I'll go make the changes now. Thread closed.

I agree with moving it to misc and calling it Any% (Trading), and renaming Any% NoCD to Any%. If nobody objects in 24 hours I'll go ahead and do it.

Making a long delay at the start would do nothing to prevent you from knowing the RNG state. However, it would make resetting to try and hit a deck significantly more inconvenient, especially because to my knowledge there aren't any bands of good decks one could go for to make it nice and easy (whereas a range for easy modes or drop manip is much easier to find, due to how the game does the calculations). At which point it comes down to how determined someone is to cheat in this way, and that's not a technical question.

Another thing to consider is the attack of learning the RNG state mid-run, and creating a Dragon Treasure or whatever manip on the spot. This is absolutely possible, given one can learn the RNG state from the player's draws. Clicking on a tool is inconvenient mid-run, but a brb or doing it during an A TEC are two possible approaches to disguise this. How much one should consider this a problem or not is a subjective matter so I won't go there.

I decided to make a video demonstrating starter deck manip and Seto 1 Beast Fangs manip to make the point.

There are some obvious points to raise so I'll pre-empt them: the menuing is bad and I scan through the deck pretty quickly. Neither of these were an attempt to hit a specific seed to start the duel on, I just did it naturally and didn't bother spending much time on it because I just wanted to get into a Seto duel. I just went with the first duel I got into and turned it into a not egregiously suspicious Beast Fangs drop. You could easily do this with any range of seeds so you wouldn't need to delay going into the duel. Furthermore, you could experiment with the duels you go into to find even less suspicious ways of playing the duel (although I argue most of the plays in that duel can be justified by imperfect play).

I'll also note that this was very easy to find (hell uploading the video turned out to be the bigger hassle) and I only used my manip tool for this, which has been publicly released.

I'll comment on one thing since it's at least a somewhat technical matter. Part of the problem is it is (in my and Gcah's view) impossible to ban enough things to get anywhere close to the point where RNG manip can always become 'obvious'. And in the absence of rule proposals that would achieve that, the claim that you could do it feels very hollow.

Now you could of course hold the view that this doesn't matter. Or you could hold the view that of course it matters. That is beyond the scope of what I intended this thread to be about however, so I'll ask everyone to kindly move that part of the discussion elsewhere.

EDIT: One more thing, RNG manipping in this way would actually be one of the simpler ways to cheat. Changing drop pools and enemy deck pools is completely detectable with current knowledge, and splicing has to be done very carefully to be undetectable (and I do mean more care than just the video editing aspect).

Tyler, people are free to submit runs that fit in a misc category to a non-misc category simultaneously so long as it fits the rules. See for instance Kanek's no THTD run. In future do at least try to have a passing semblance of intelligence, beyond that required to use a keyboard.

I argue for misc primarily because it's a compromise: generally people not in favour of separation are okay with it and there are people pro-separation who also said they would be fine with it. One could always count the people who said they've been up for it as a compromise. Yes it will annoy some people, but that is true of literally every course of action. I don't see how it's possible to really get much further into reasons without basically rehashing the arguments featured in the speparation/no separation discussion so I'm not going to go there in here.

I do feel one danger should be pointed out. If you have a poll on misc./no misc. after the separation discussion has concluded in favour of separation, then at that point any incentive for compromise is lost. So I approve of having this discussion concurrently.

I'll just say I prefer no separation here, for the purposes of tallying, and failing that I would be in favour of having no RNG manip as misc.

On random sorts, yeah I see no objection to banning spamming it. Banning random sorts completely is in my view out of the question (and wasn't what Grig suggested anyway) simply because people often deck edit by doing Numerical -> Random -> Type. It should be noted that one random sort only saves you 40 frames of waiting anyway, so allowing 1 or 2 doesn't really change anything. Defining an appropriate limit of random sorts isn't difficult.

On 3d battles, I have no objections to banning them outside of Nitemare, where as GFC says it doesn't matter because the drop is useless and it will do nothing to save you from BEUD or anything. The absolute most it could theoretically do is change the RNG so Nitemare doesn't fuse MBD, but if you had the knowledge of the current RNG state required to know to do that then you could most likely achieve the same result with dumps or fusions, and if someone can get into that position then there are far bigger things one could do.

For console resets, I will bring up that crashes are possible (at least I think so; has anyone actually had a crash from Dark Hole/DCJ and not just a stall? (EDIT: yes, someone's told me they have)). I don't think this is a reason to not ban console reset, but I bring it up because someone could conceivably have a problem with it.

For idling, I see no reason for allowing idling outside of places where it makes no difference (duels and library screen basically, EDIT: password screen is fine, also Egypt overworld interestingly enough). Of course someone could idle mid-duel and load up say my program to find a drop manip or a draw predictor or whatever. This would be harder to do much with if duel idling was restricted to the drop screen, but how much the existence of this line of attack matters to someone will vary.

I do contest the claim the above makes RNG manip difficult, because of starter deck manip. One could go a bit further and then manip a Beast Fangs from Seto too. Random sorts and 3d battles would not be essential, since you wouldn't be going for something as rare as the ranges Final 6 manip uses or winning MBD from Heishin with a starter deck. Furthermore, if someone wanted to slide manip under the radar they would be far more likely to do something like this anyway, so in my view all the rules do is ban the current strats.

There's been a lot of discussion about the leaderboards again as of late. Part of the discussion has focused on what rules would demarcate the two, although of course this has not been the whole discussion.

Universal claims like 'of course it can be done' and 'it's impossible to define RNG Manip' are not convincing anybody. Therefore I suggest we discuss concrete proposals for rules here, purely from the point of what certain rule proposals fail to prohibit and uninterrupted by the broader conversation.

I would like this discussion to stay on objective matters. Therefore please no rule proposals involving appeals to 'intent', since this cannot be pinned down. Of course it could be decided that rules need not be objective, but that is a separate matter and beyond the scope of this thread.

Lastly this thread is primarily meant to help inform the general conversation, rather than a place where final rules are decided or any such thing.

My comment on the Kaiba Single WR: 'RNG manipulation is glorious.'

Fine, I'll do it myself then.

Not much to say, this program calculates the medals handed out by everyone if you tell it the medals dropped by six or so pairs. http://www.mediafire.com/file/sqmczso97nccqaj/Medals_Predictor.zip

No doubt this will all be in the code, the question is how difficult it would be to read the relevant code and understand it. If you want to figure out how the AI works, I feel a better idea would be to just set up experiments and see what happens.

The first thing would be to use savestates and check if the AI's actions are determined by the RNG at all (e.g., put a savestate first turn on Seto and see if he always moves in the same direction when you do the same thing). In my limited playing around with Seto I found it didn't depend on RNG, but I wasn't specifically checking that so I could be wrong.

Assuming it doesn't depend on RNG, then the next thing to check would be to see what can change their plays. Cheat Engine could help you look at face-down cards, and also edit your and your opponent's decks before either of you draw cards. Hopefully after some of that you'd have some new general observations that would give you an idea of what to test next.

'Kinetic Soldier wouldn't necessarily need a Destiny Draw as you can use Monster Reborn etc to get him off of Manawyddan fab Llyr'

Right, that's what I meant when I said nobody would ever need to bother trying to get it off of Destiny Draw. Lots of people online do say it can happen, but this could just be because it's similar to Arsenal Bug and Woodland Sprite and so people just assume it can happen.

As for the random rares, the equally likely part comes from me looking at the code. With 22 results you don't expect people to generally get perfectly even results unless they do lots and lots and lots of three-in-a-rows. I don't want to put a number on this since I've not done any calculations, but you'd probably have to get to a couple of hundred three-in-a-rows, all recorded, before it gets convincing. E.g., right now I just generated 100 random numbers 1 to 22, and I got 10 10 times. Of course, I could have made a mistake reading the code but that bit of the code was pretty straightforward.

I'm in the middle of reorganising this post to retain only things of interest to someone, say, speedrunning hundo. Go to https://goo.gl/Ybtr6A for more technical details if you'd like to mod this game or are just into that sort of thing.

Passwords The full password list can be got from https://docs.google.com/spreadsheets/d/1zpBNCHcCoU_lidOsSlgQeEnOvB76oJkYLt5nPlhQBYk/. For convenience, the cards without a working password are listed at https://pastebin.com/LPYjSEBp

Fusion table I've put this up at https://docs.google.com/spreadsheets/d/1N1Q5uf3Xbf1KyWx0iI3REoQi7NrqdwSHb0oR06UH-LE/. It's not very convenient to use, but it's there anyway.

Deck Leader GameFAQs user Hozu found some good information on EXP gain, which they posted at https://www.gamefaqs.com/boards/589455-yu-gi-oh-the-duelists-of-the-roses/74521692?page=1/. At https://goo.gl/BfyPt3 you can find all of the leader abilities.

Slots There are 8 cards where lining them up in slots gives a special rare, and they are:

060 Dark Magician -> 087 Dark Magician Girl

061 Illusionist Faceless Mage -> 733 Brain Control

272 Harpie Lady -> 702 Harpie's Feather Duster

539 Thunder Nyan Nyan -> 700 Raigeki

716 Dark-Piercing Light -> 715 Swords of Revealing Light

722 Stain Storm -> 811 Acid Trap Hole

754 Dark Energy -> 699 Dark Hole

837 Ultimate Dragon -> 083 Lord of D.

Any other three-in-a-row will result in one of the following 22 random rares, all equally likely:

000 Blue-Eyes White Dragon

007 Red-Eyes B. Dragon

054 Right Leg of the Forbidden One

060 Dark Magician

075 Skull Knight

092 Gemini Elf

113 Temple of Skulls

115 Fire Reaper

118 Flame Ghost

151 Flame Swordsman

294 Summoned Skull

362 Toon Summoned Skull

408 Jirai Gumo

486 Labyrinth Tank

498 Cannon Soldier

502 Barrel Dragon

629 Dissolverock

635 Barrel Rock

791 Sword of Dragon's Soul

792 Enchanted Javelin

793 Anti-Magic Fragrance

819 Gorgon's Eye

The cards that do not show up as a normal card in slots are listed https://pastebin.com/iScCqmxN

Destiny Draw Your Destiny Draw draw depends on situation, rank, and some RNG (no surprises there). You must have less than 1000 LP, and your rank must meet a minimum requirement depending on your DL, either 1LT or LTC. The requirement for each card is given in the deck leader abilities section.

Think of the possible draws as laying on a grid, where you have 3 columns and 7 rows. The row is chosen by the game at the start of your turn and comes from the situation you're in, but your draw from that row isn't determined until you look at your hand.

The rows are:

row 1 - Mirror Force, Dark Hole, Woodland Sprite

row 2 - Raigeki, Dark Hole, Arsenal Bug

row 3 - Dark Hole, Earthshaker, Earthshaker

row 4 - Riryoku, Raigeki, Raigeki

row 5 - Royal Decree, Harpie's Feather Duster, Royal Decree

row 6 - Mirror Force, Dark Hole, Dian Keto the Cure Master

row 7 - Raigeki, Swords of Revealing Light, Dian Keto the Cure Master

If your rank is at most RADM, then the card chosen from the row is 20% first card/25% second card/55% last card. If you're above RADM, it's 100% for the first card.

The following are the conditions for each row; they are checked in the given order.

Woodland Sprite row: The highest attack monster* on the field • Must be opponent-controlled • Must be a Fiend • Must be less than 3 squares within your DL Your opponent must control at least three Fiends.

Arsenal Bug row: You must have an Insect as your DL. Your opponent must control at least three monsters such that • Their attacks are between 3000 and 3999 inclusive • They do not benefit from Forest (i.e., they are not Beasts, Beast-Warriors, Insects, or Pyros).

Earthshaker row: There must be at least 31 Labyrinth squares. You must control no monsters that can go through Labyrinth. Your opponent must control at least 3 monsters that can go through Labyrinth.

Riryoku row: Your opponent must control the highest attack monster*. The highest attack of cards out of your hand must be lower than said monster's attack. The difference must be lower than half your opponent's LP.

Royal Decree row: You must control the highest attack monster*. Your opponent must have at least 4 traps out.

If none of the above conditions are met and the player has less than 250 LP, then one of the remaining two rows are chosen at random. Else there is no Destiny Draw.

• If there is a tie, highest defense is the tiebreaker. If there is still a tie, the card summoned first wins.

Reincarnation Not every card can be obtained from reincarnation. Those that can't are listed at https://pastebin.com/VH4Yw8Xw

When you reincarnate a card, the game first decides whether to give you a monster or spell. If you reincarnate a monster, it's 80% monster/20% spell. If you reincarnate a spell, it's 60% monster/40% spell.

Next it decides what DC of card to give you. There are two ranges depending on the DC of the card you're reincarnating: the high range that goes from DC + 1 to DC + 10, and the low range that goes from DC - 10 to DC + 1. The game decides which range to pick depending on the ranks of your deck leaders. It takes the maximum rank of the deck leaders of decks A and B as a number (0 for NCO, up to 12 for SD), and the chance it gives you the high range is (8 + 2 x rank)%. The chance of the low range is (92 - 2 x rank)%. (So if you want the low range, make deck C your main deck and don't have a deck leader in decks A or B.) Once the game has decided the range, it just picks a deck cost from the range at random.

Next the game gets the list of monsters or spells (as chosen earlier) with the chosen DC, except for the card you are reincarnating. If this list is empty, the game lowers the DC by 1 and tries again until either the list is not empty or the DC is below 1.

Once that is done, if the list is not empty the game returns a random card from the list. Otherwise the game gives you Fake Trap.

~ Stuff below here still needs to be edited. ~

Strings Okay, this isn't really that insane, contrary to my initial view. It's an LZ77-type thing. You'll want the character map at https://pastebin.com/znxgzEct

So, generally characters on their own are represented as 0x2000 + <code>. I've also seen 0xA000 + <code>, which I used to think was a string terminator but I've since figured it can't be that. Maybe it's some sort of instruction to the game to help it space out the font or something?

So there's a table at 0x2D19D0 in the RAM, or 0x1D82D0 in the SLUS file. This is an array of 4-byte ints telling the game where to jump to for a given string. There is then an array of 2-byte characters at 0x2D49D4 in the RAM, or 0x1DB2D4 in the SLUS file. I'm not sure exactly how the game is keeping track of the string lengths, but I found that by assuming they're just packed beside one another I got reasonable results.

However, they do a trick to try and save room. Let's take the string for BEWD's name, which is Blue-Eyes White Dragon and is string 320. We find that the string is 2057 207E 2087 2077 20A2 205A 208B 2077 2085 2070 206C 207A 207B 2086 42C3 40FD 2084 2073 2079 2081 A080.

This mostly seems reasonable enough. Apart from the strange 42C3 40FD, it spells out Blue-Eyes Whit<???>ragon. So what's going on here? Well, first the second most significant bit is set here, which tells the game these aren't normal characters but are in fact pointing to segments of other strings to save room. The second 2-byte number tells us what string it's from once you and it with 0x3FFF, so we're told to look at string 0xFD = 253. Now, take the first number and right shift it by 6, then and it with 0x7F. This tells you how far you need to jump along that string (in this case, it's 11 characters). Lastly and the first number with 0x3F to get how many characters our substring needs to be (here it's 3).

It turns out that string 253 is Play a Practice Duel, and it's taking the 'e D' bit. The 11 seems weird, but parts of string 253 are compressed so it works out. Yes, this means things can get recursive.

I've decided to check my understanding by running a short script to extract the strings according to this, My results are at https://pastebin.com/m1mcmaJ5 . Any strings skipped over are due to them containing a character not in my character map; odds are this means I need to look deeper for more characters, given the gaps in the area with the cutscene dialogue. Also some of the last strings are suspiciously gibberish,

Starter decks and names The game calculates a number between 0 and 15 from your name and grabs a pool of three starter decks using it. The pools are:

Pool 0 - Fairy King Truesdale, Patrician of Darkness, Twin-Headed Behemoth Pool 1 - Serpentine Princess, Birdface, King Tiger Wanghu Pool 2 - Luminous Soldier, Kryuel, Tactical Warrior Pool 3 - Maiden of the Aqua, Airknight Parshath, King Tiger Wanghu Pool 4 - Robotic Knight, Fairy King Truesdale, The Illusory Gentleman Pool 5 - Serpentine Princess, Airknight Parshath, King Tiger Wanghu Pool 6 - Maiden of the Aqua, Birdface, Wolf Axewielder Pool 7 - Luminous Soldier, Patrician of Darkness, Tactical Warrior Pool 8 - Fairy King Truesdale, Kryuel, Twin-Headed Behemoth Pool 9 - Serpentine Princess, Patrician of Darkness, King Tiger Wanghu Pool 10 - Luminous Soldier, Birdface, Tactical Warrior Pool 11 - Maiden of the Aqua, The Illusory Gentleman, Wolf Axewielder Pool 12 - Airknight Parshath, Maiden of the Aqua, Molten Behemoth Pool 13 - Thunder Nyan Nyan, Serpentine Princess, Luminous Soldier Pool 14 - Fairy King Truesdale, Thunder Nyan Nyan, Tactical Warrior Pool 15 - Kryuel, Injection Fairy Lily, Twin-Headed Behemoth

To calculate the number from a name, you'll need the character map at https://pastebin.com/znxgzEct . All of those numbers are in hex.

Now first, extend the name to 12 characters by padding with a filler character I'll call @. Next, for each character in the name look up the corresponding number in the character map and take the remainder after division by 16. For @, the number is 14. Now add the 12 numbers together, and take the remainder after division by 16.

I'll do an example: CM Punk. The name after padding is CM Punk@@@@@, and converting to numbers we get 8, 2, 0, 5, 7, 0, 13, 14, 14, 14, 14, 14. Add these up and you get 105, which is 6 * 16 + 9 so Pool 9, which gives us Serpentine, Patrician, and Wanghu.

The info the game shows you at the deck selection screen is stored at 0x1A0600 in the SLUS file. Each deck's info is given by a 12 byte block. The first two bytes are the deck leader's ID (if this matches the correct DL then the image will display correctly). The next two bytes are the average summoning level multiplied by 100.

Next two bytes are the attributes, stored in the lower 12 bits. It's four 3 bit numbers packed together, each representing an attribute. 0 is nothing, and the normal attributes are their ID above plus 1. The highest 3 bits is the leftmost attribute, the lowest being the rightmost.

Next two bytes are padding and the final four bytes are the main types or 'races'. It's similar to the attributes, it's four 5 bit numbers packed together with 0 being nothing and the others being their usual ID plus 1. Again, highest 3 is leftmost type and so on. Note that the leftmost type is the type displayed by the DL's name, which is why Fairy King Truesdale is shown with the Insect symbol.

http://datacrystal.romhacking.net/wiki/Yu-Gi-Oh!_Forbidden_Memories:RAM_map

Which brings us on to the draw predictor. The tool I'm working on now will contain all of the functionality of the draw predictor, so there's no reason for me to keep working on it. As a result, I am releasing it. You can download it here: https://drive.google.com/file/d/1Q2CTLME7-Q6yy3nO3aqdZkdBxLU_HL8z/view?usp=sharing. It's Windows only right now, but if you're on another OS and want it then let me know. I'm not looking for feedback on this; I already know plenty of ways it could be improved but there's no need to work on this when I'm building the successor. Yes, it's annoying to use, yes, the error messages never vanish, yada yada I know, spare me. I'm mainly releasing it because it's about damn time the community had the means to do verification. I am however willing to help with issues that could come up when using this for verification; see below.

With that said, a quick tutorial. What you're interested in is the .exe inside the folder, I suggest making a shortcut. To recreate a duel, you need three things: the deck, the pre-duel sort, and the seed. By punching in the deck and the opening hand (and more of the first 20 cards if needed), the program will spit out the range of seed numbers that could give that shuffle. Once you have a seed number, you can use the predictor to figure out what the actual seed is (so for example, seed 0 is 0x55555555). To recreate this duel, you then recreate the deck, go into the duel with the correct pre-duel sort, then as the screen goes to black you pause the emulator and change the seed to the desired value (using say Cheat Engine, or the RAM Watch if you're using Bizhawk). There's a bit more info at the start of my Mergy 3:07 doc: https://docs.google.com/document/d/1y_zXISrseL8at2kVssQJBROfYjS1V3WV8swSu30UMuI/.

There are some caveats to this however: the draw predictor doesn't work past the first 10 million seeds, it doesn't deal with random sort, and this doesn't in any way help you check the starter deck. If you're verifying a run and run into these issues, then get in touch and I'd be happy to help. None of these issues are that difficult to deal with per se, just the draw predictor doesn't handle them.

I appreciate that this tutorial was very short and probably not that helpful. If people find this sufficiently unclear then I'll write a better one, but for now this'll have to do.

New strats first. It turns out I made a small mistake when I was initially looking for decks, and so there are in fact early decks that can get MBD without winning an additional card first. This includes the 14810 deck, but also some very early decks.

There are two main strategies available: go for decks 706/711, or go for deck 14810. TAS speed is about deck 211 or so, to give you an idea of how quick the early decks are.

The advantages over the previous strats are that you do not need to wait for a > 900 card before getting your MBDs, and both strats only involve one "spawn 3d duel" per MBD. A "spawn 3d duel" is when you quit the 3d duel as your monster is spawning in. A very safe time to quit is once the three coloured circles have merged into one white circle. You can save maybe half a second to a second from this, and if you want to then I suggest looking at the seed in Bizhawk and seeing when it stops changing. I also speak of a "quick 3d duel" in the doc, this is just starting a 3d duel and then immediately hammering circle to quit it asap.

So, what are the differences between the two strats? Well, you can look at the three decks here: http://pastebin.com/mUfS08jd/. As you can see, the later deck has Raigeki, Beast Fangs, Sogen, and 2 D/5 T. Meanwhile the earlier decks have Dark Hole, Power of Kaishin, Umi, and 2 D/3 T. Other than that and waiting time for the deck, they're more or less the same: the MBD duels are more or less identical. Edited: it turns out all three decks have three MBD frames to go for, but the 468 duel for the 706/711 decks is slower. There may be faster ways to win on seed 468 with the 706/711 decks, if instead of setting Umi you play a card in defense.

Next, timers. Here is a set of spreadsheets with everything you'll need: https://docs.google.com/spreadsheets/d/1zF3ZuGZrdMOKSG0EDXkvsVWoJQFHW72kllmdZZLPV2g/. If you opt for the 14810 deck and you ran the old strats then obviously no change is needed. If you opt for the 706/711 decks, and you want to aim for 708.5 let's say, then you will want to subtract 235.03 seconds, or 3m55.03s. If you've never done manip runs before, then the sheets given will help you calibrate your timer.

As for the MBD timers, you'll want to aim for 470, so subtract 2.85s.

Now, MBD strats. You'll find them here: https://docs.google.com/document/d/13xqZYvL-twV6BrHp21HF1zEohDgoXMDSH5rCBkcSD5o/. As you can see, they're all very similar. I provide optimal plays for each duel, for the ease of newer players who may not know all the fusions to avoid.

So, what next? The next big thing would probably be final 7 manip, then chaining the MBDs together to save on resets. Needless to say, this is a lot of work and it's not good that the community doesn't have the tools to easily work on this. As a result, I am currently spending my time creating a tool to aid runners in coming up with RNG manip strats. No, I don't know when it will be ready, that depends on how much free time I find and how many technical challenges I meet on the way.

So guys I think this could be the next YGO speedgame. With that in mind, what kind of tech does a runner need to execute to get a top time?