Rock, Paper, Scissors, Theory, Peach

[Sycorax]

My friend @ Awstintacious came up with an idea for Peach shield pressure, which I've tested out. I'm not entirely sure it's new, but it's his original thought. The idea is doing an FC Nair on an opponent's shield after having already done an FC aerial on their shield. For example, Peach lands an FC Fair, then, as soon as her lag is over, she does an FC Nair. In our opinion, it is a powerful option that opens up possibilities. It overhauls the traditional Rock-Paper-Scissors game Peach has with FC'd aerials on shield. Second Nair covers a lot of options. Here's the breakdown.

--- Standard Peach Shield Pressure ---
I'll first start with the original Peach shield pressure interaction. As we all know, Peach's Fair and Uair are maximally +4 on shield. Bair is can be set up easily to be +4, and Nair is maximally +3 (see this thread). This frame advantage allows Peach to do a lot of options on an opponent's shield because she has lots of frames to start a move before the opponent can act. Her first actionable frame can come up to 4 frames before the opponent's.

When I talk about first actionable frames, I am referring to the frame of input, not the first frame the move comes out on. For example, Peach is actionable on the fourth frame of Landing. She has 4 frames of landing lag on all of her FC aerials. On the fourth frame of her landing lag, she can input down on the c-stick to start dsmash on the next frame. Here's a table:
+-----------+---+---+---+---+---+---+---+---+---+
| Frame ... | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 |
+-----------+---+---+---+---+---+---+---+---+---+
| Animation | L | L | L | L | D | D | D | D | D |
+-----------+---+---+---+---+---+---+---+---+---+
L = landing animation, D = dsmash animation.
On the 4th frame, dsmash is input. On the 5th it starts. On the 9th, it hits. In the case of a +4 aerial, Peach is actionable on frame 4 and the opponent is actionable on frame 8, in the table above. Whether we talk about which frame a move starts on or which frame the move is input on doesn't matter. It only shifts the frames forward or back one frame. If I say, "Input dsmash on frame 4," that's the same as saying, "Start dsmash on frame 5." IASA frames refer to the start of the next animation. Peach's jab, which is IASA frame 16, can be interrupted by an input on frame 15.

So Peach can get up to a 4 frame head start when she hits an opponent's shield. This gives her lots of time to start her moves before the opponent can act. There are lots of good options for Peach to choose, and there is counter-play to each of Peach's choices. In this tight window of interaction, there is no time to spend on reaction. The options of each player need to be pre-calculated and timed. You have entered an intense battle arena of Rock-Paper-Scissors and Yomi Layers. Common options for Peach after an FC'd aerial on shield are jab, grab, dsmash, and wait. Common responses include buffered roll, wait in shield, grab, and attack OoS.

Here is a matrix of the options and who wins:
+-----------+------+------+------+--------+
| Peach\Opp | Roll | Wait | Grab | Attack |
+-----------+------+------+------+--------+
| Jab ..... | P* . | O .. | P .. | P .... |
| Grab .... | P/O**| P .. | P .. | P~ ... |
| Dsmash .. | P*** | O+ . | P .. | P .... |
| Wait .... | P .. | ~_~ .| O .. | O .... |
+-----------+------+------+------+--------+
P=Peach wins, O=Opponent wins
*Peach needs to input the jab between being +1 and -1 on shield; if more than +1, she can delay to avoid hitting shield.
**Peach needs to be +4 exactly to land grab here
***Peach needs to be +2 or more to land dsmash
+Can survive most of the time if shield is big enough and with shield DI
~Some OoS options are technically fast enough, but are difficult and few people actually go for them. No attack options cover a grab after +4 advantage.
~_~ Stare into each other's soul

To understand why each option beats or loses to what, you must understand the frames of each option. Rolls are vulnerable for the first 3 frames. Relevant grabs take 7 frames (some take more). Options OoS vary; the fastest are usually 5 or 6 frames (Fox's shine OoS is 4). Frames on attacks can be looked up here. How frame positive Peach's aerial is will open her up to more of the opponent's options. It would behoove Peach players to know what each sort of frame positivity looks like and to practice getting fast aerials on shield.

This constitutes the standard RPS of Peach shield pressure:

• If Peach thinks the opponent will roll, she should time a jab, dsmash, or wait.
• If Peach thinks the opponent will wait, she should grab.
• If Peach thinks the opponent will grab, she should jab, grab, or dsmash (most reward goes to grab probably).
• If Peach thinks the opponent will attack OoS, she should jab or dsmash.

From the opponent's perspective:

• If the opponent thinks Peach will jab, they should wait and shield DI away.
• If the opponent thinks Peach will grab, they should buffer a roll or go for an attack OoS if they are technical enough.
• If the opponent thinks Peach will dsmash, they should wait and shield DI. Even small shields can survive this because the dsmash will tend to only hit once.
• If the opponent thinks Peach will wait, they should grab or attack OoS.

For the opponent, waiting is a really powerful option especially since jab and dsmash are the most popular options. If the opponent waits, they are free to react to Peach's wait with a grab. Roll works pretty well, too, especially if the Peach isn't very frame positive.

For Peach, jab, dsmash, and grab are all effective options that cover loads of options and lose to at least one. Jab and dsmash (jab espeically) tend to be the favored options chosen by players currently.

Before I move on, I'd like to explain something about why jab is actually a terrible option for Peach despite being the most popular. She doesn't even need it to cover all of the opponent's options. She can mix up between grab and dsmash.

--- Jab ---
Jab is either not safe or Peach doesn't get much out of it. There is easy counterplay to it, and it can be done on reaction (kinda). Peach can either space the jab or not; that is, she can land far away from your shield with her FC aerial so that she is out of your grab range, or land close within your grab range.

• If Peach does not space the jab, you can grab her every time. Do this by holding down on the analog stick, holding shield, and mashing Z or A to grab. This inputs ASDI down and grab. I suggest timing the input of the first grab of the mash to be the first actionable frame out of shieldstun. There are two scenarios.

1. She hits you with jab. Perhaps you were trying to do something out of shield, such as grab or roll. If she hits you with a jab, you can ASDI down to the ground and grab before she is out of lag from her jab, even though you got hit.
2. She hits your shield. Peach's first jab is -10 on shield, and her second jab is -11. This is plenty of time to grab.

• If Peach spaces the jab, she is still -10 or -11 on shield. You have 10 or 11 frames to do whatever, WD, jump, usmash OoS with Ness to go for the Yo-Yo glitch, etc. If you see her land her FC aerial far away on your shield, don't go for the ASDI+grab spam because you will most likely not be in range to grab.

I've replaced grab in the matrix with ASDI down+grab. The opponent now has counter-play, besides wait, to Peach's jab.
+-----------+------+------+-----------+--------+
| Peach\Opp | Roll | Wait | ASDI+Grab | Attack |
+-----------+------+------+-----------+--------+
| Jab ..... | P* . | O .. | O ....... | P .... |
| Grab .... | P/O**| P .. | P ....... | P~ ... |
| Dsmash .. | P*** | O+ . | P ....... | P .... |
| Wait .... | P .. | ~_~ .| O ....... | O .... |
+-----------+------+------+-----------+--------+
*Peach needs to input the jab between being +1 and -1 on shield; if more than +1, she can delay to avoid hitting shield.
**Peach needs to be +4 exactly to land grab here
***Peach needs to be +2 or more to land dsmash
+Can survive most of the time if shield is big enough and with shield DI
~Some OoS options are technically fast enough, but are difficult and few people actually go for them. No attack options cover a grab after +4 advantage.
~_~ Stare into each other's soul

If the opponent wants to roll because [reasons], they can buffer the roll and then immediately start ASDIing down and mashing grab. If they get the roll, good for them! If they get jabbed out of their roll, they get a grab. They can stop the mash once they see the roll succeed and do whatever out of roll. It's an option select almost. If Peach dsmashes, the opponent will eat a meaty dsmash.

The favorable options for the opponent become wait and roll. They can ASDI down+grab if they're expecting a jab. If not, do whatever they feel appropriate OoS. The favorable options for Peach become grab or dsmash. However, there are some situations of low frame positivity where grab and dsmash don't cover roll sufficiently. To cover roll, Peach needs to wait, but wait loses to almost all of the opponent's options. Using an FC Nair after her first aerial adds a new twist.

--- Second Nair on Shield ---
Now let's get to the new stuff! I call it second Nair on shield because it's a Nair and it comes second. It doesn't necessarily mean that a Nair was the first aerial used on shield. The idea is to, after landing an FC aerial on shield, dash and then use an instant FC Nair. It needs to be frame perfect to be legitimate (but opponents, most likely, won't be prepared for it so you have some wiggle room). Second Nair is technical, but doing an FC Nair after a previous FC aerial provides great advantages. First of all, it'll probably scare the **** out the opponent. Second of all, it covers roll and wait, and pseudo-covers grab and attack.

Here's the frame-by-frame breakdown of the inputs. It takes 11 frames to get out an FC Nair hitbox (1 frame of Dash, 5 frames of Kneebend, 1 frame of JumpF, 1 frame of Float, and 3 frames of start up on Nair). There is a frame of Dash in there to keep you close to your opponent when they slide back in shieldstun.
+-------+--------+-----------+
| Frame | Input .| Animation |
+-------+--------+-----------+
| ... 1 | Right .| Landing . | Right or left on the analog stick to dash.
| ... 2 | X+Down | Dash .... | You can actually input down as late as frame 8 to buffer the float.
| ... 3 | hold . | Kneebend .| Hold your jump and down inputs.
| ... 4 | hold . | Kneebend .|
| ... 5 | hold . | Kneebend .| Opponent becomes actionable on this frame.
| ... 6 | hold . | Kneebend .|
| ... 7 | hold . | Kneebend .|
| ... 8 | hold . | JumpF ... |
| ... 9 | A .... | Float ... | You need to let go of down.
| .. 10 | Down . | Nair .... |
| .. 11 | Down . | Nair .... |
| .. 12 | Down . | Nair hits |
+-------+--------+-----------+
It's the same timing as a normal instant FC Nair. You see Armada do them as hand-warmers; you can do them too. There is just a dash input right before it. You don't need to press down right after the dash either; you have some frames of leniency to get your analog stick to the down position to float. After a +4 FC aerial on shield, this gives your opponent 7 frames (+4 - 11 = -7) to act. An FC Nair on shield will maximally be +3 giving your opponent 8 frames to act. For this reason, I wouldn't suggest going for second Nair unless you think you're really frame positive.

With these 7 frames in mind, how does second Nair change the matrix? It may seem the same as dsmash but 11 frames instead of 5. Second Nair covers the same options as Dsmash and some others. Dsmash can cover roll, it for-sure covers grab OoS, attack OoS, and ASDI+grab; however, dsmash loses to waiting in shield. Second Nair covers the wait, and sacrifices coverage of attack OoS. Here is a breakdown of how Double Nair covers options:

• It covers roll not by hitting your opponent but by being a fast option. If your opponent opts to roll they will indeed go through the roll animation because you will have missed the 3 frames of vulnerability at the beginning of the roll. However, you will land and be actionable soon enough after the FC Nair to react to the roll and punish. After a +4 aerial, you will be actionable on the 11th frame of the opponent's roll. You will be actionable one frame later for each frame fewer of advantage you have. Most rolls take 31 frames, if not more, giving you 19 frames or more to do something.
• It pseudo-covers wait in shield by resetting the situation with Peach having +3 advantage. This is still advantageous for Peach, though, because the opponent now has a smaller shield, and a dsmash after the second Nair is more likely to shield poke. This means that after second Nair, dsmash covers all options of the opponent. Only shields that are very close to full right before the first aerial are likely to survive this dsmash after the second Nair.
• It loses to grab barely. Your opponent has 7 frames to get out a move and grabs come out in 7 frames (or more if the opponent's character sucks). The opponent needs to input the grab on their first actionable frame, the last frame of their shieldstun.
• It covers most attacks OoS. A few characters have OoS options that are 7 frames or fewer. Some notable ones are Fox's and Falco's shine, Sheik's Nair, Marth's UpB, and Samus's UpB. The timing is tight and most opponents will not be used to the sheildstun to achieve the correct timing. And against characters with bad OoS options, they are SOL. It is important to note that this attack OoS can't be done on reaction to Peach doing the second Nair. The opponent needs to know or feel the timing to act instantly out of shieldstun from the first aerial.
  • Many opponents may be reluctant to try to attack OoS and rightfully so. If they guess wrong, they could be grabbed or dsmashed. If Fox or Falco try to shine, they could eat a meaty dsmash.

Here is a new decision matrix including the 2nd Nair:
+-----------+------+-------+-----------+--------+
| Peach\Opp | Roll | Wait. | ASDI+Grab | Attack |
+-----------+------+-------+-----------+--------+
| Jab ..... | P* . | O ... | O ....... | P .... |
| Grab .... | P/O**| P ... | P ....... | P~ ... |
| Dsmash .. | P*** | O+ .. | P ....... | P .... |
| Wait .... | P .. | ~_~ . | O ....... | O .... |
| 2nd Nair .| P .. | Reset | P~ ...... | P~ ... |
+-----------+------+-------+-----------+--------+
*Peach needs to input the jab between being +1 and -1 on shield; if more than +1, she can delay to avoid hitting shield.
**Peach needs to be +4 exactly to land grab here
***Peach needs to be +2 or more
+Can survive most of the time if shield is big enough and with shield DI
~Some OoS options are technically fast enough, but are difficult and few people actually go for them. No attack options cover a grab after +4 advantage.
~_~ Stare into each other's soul

Staleness matters for the shield pressure. Specifically, if your aerial does less than 13.43 damage, then you will lose a frame of shieldstun, and thus, have one frame fewer of advantage on shield. If your aerial does less than 11.195 damage, you will lose two frames of shieldstun. Unless you've been extremely trigger-happy and built up a lot of staleness, you can expect a staled aerial to do 1 fewer frame of shieldstun. Nair will do under 13.43 after one recent hit. Fair will do under 13.43 after two hits, generally. You can play around with Strongbad's Shieldstun Calculator to get a feel for how staleness works. It's incredibly hard to keep track of it in game. Staled Nair will be +2. Staled other aerial will be +3 maximally.

For all OoS options performed by the opponent, it is important to understand what sort of frame advantage Peach has from differently timed aerials and float heights. Only FC Fair and Uair can achieve +4 near to the ground (and Bair can achieve it if Peach floats for at least 3 frames and lets go of float when she starts the bair, see here). FC Fair can achieve +4 from almost any height because of it's long startup; it's the best :3. FC Nair is going to be maximally +3 on shield.

--- Conclusions ---
The final RPS looks like this:
+-----------+------+-------+-----------+--------+
| Peach\Opp | Roll | Wait. | ASDI+Grab | Attack |
+-----------+------+-------+-----------+--------+
| Jab ..... | P* . | O ... | O ....... | P .... |
| Grab .... | P/O**| P ... | P ....... | P~ ... |
| Dsmash .. | P*** | O+ .. | P ....... | P .... |
| Wait .... | P .. | ~_~ . | O ....... | O .... |
| 2nd Nair .| P .. | Reset | P~ ...... | P~ ... |
+-----------+------+-------+-----------+--------+
*Peach needs to input the jab between being +1 and -1 on shield; if more than +1, she can delay to avoid hitting shield.
**Peach needs to be +4 exactly to land grab here
***Peach needs to be +2 or more
+Can survive most of the time if shield is big enough and with shield DI
~Some OoS options are technically fast enough, but are difficult and few people actually go for them. No attack options cover a grab after +4 advantage.
~_~ Stare into each other's soul

The opponent likes to roll or wait still. They can throw in ASDI down+grab against low frame positivity, jabs, or second Nair. They can also attack OoS, but that loses to all other options :/

Peach still likes to go for grab and dsmash, but she can do second Nair to cover all options except against the most competent opponents. Even against competent opponents, it's still a mixup. The thing is opponents need to go for poor options, which lose to the standard grab/dsmash mixup, in order to cover second Nair.

The year is 20XX and that doesn't only apply to Fox and Falco. Second Nair is mildly technical by spacies' standards; however, it is completely doable with practice. Furthermore, counteracting it requires a fair amount of skill or luck on the part of the opponent. This means it can cover ALL OF YOUR OPPONENT'S OPTIONS.

 

[rje457]

THIS IS AMAZING, I definitely want to apply this to my play. Thank you so much for sharing this!

 

[Awstintacious]

Nobody else thinks it's possible huh? Glad to hear it peches.

 

[Scoopy]

I already kinda did this on shield, but I really wasn't sure if it was a good option or not. Thanks for posting.

 

[red stone]

This is very impressive. if this is safe, then this will legitimately be seen in future peach play for sure

 

[Quetzalcoatl]

Very well written. Jab being overrated post FC aerial is a bit of a wake up call. I feel like its a relic of the past that has stuck with me personally at least.

I am keen to try out follow up Nair instead. I could imagine the amount of shield destruction that takes place with the sequence of: aerial turnip throw-> fc fair-> fc nair-> dsmash

 

[Sycorax]

Something I never considered in writing this was buffered jump OoS. The 3---and maybe 4 and 5---frame jumpsquat characts may be able to make use of this. Roll has 3 frames of vulnerability for most characters so any option that covers roll would cover a 3 frame jumpsquat similarly. Also, it takes some time to actually move high enough into the air to avoid grab, dsmash, or whatever. I think buffered jump should only be a viable option when the Peach doesn't land a move that is very frame positive.

 

[JBW2]

I will go back an read the entire post again. Based on my observations, I believe that this is indeed more advantageous than not.
I think that the matrix is too myopic. I think there are more factors that determine the success of a move etc. I'm assuming we mean Spacies for the most part here.

For example, I think you have underrated the Jab. I do agree that a bad jab can be punished severely, but I also think there's RPS within Jab which you've talked about.

1) the opponent can ASDI + grab 2) Nothing 3) attack 4)) roll 5) regular grab

Jab goes the next step compared to grabbing, because you can choose whether to jab twice

1) peach is screwed 2) jab again, dsmash, reset neutral, grab, anything really 3) jabbing may catch the attack 4) chase 5) jab again

In other words, there is only one bad option on the surface. The problem is if she jabs twice and they do nothing 2) peach is screwed

I really do like the possibilities for FC NAIR. The only issue is spacing and proper FC. There's more room for error here. Depending on a character and situation, a roll may not mean Peach wins. For example, rolling to the center stage. If you are programmed to FC FAIR FC NAIR, I'm sure that you would instinctively have something to cover most options like a down smash or grab. Rolling is harder because it's the only true mobile one. Peach at the edge versus a spacies towards the middle of the stage is bad news.

All in all, I like the idea. The RPS definitely changes based upon match up. I think it's a new ball park for some of Peach's harder match up's i.e Shiek (amazing shield), Fox/Falco (extreme mobility; myriad options), Jigglypuff (must I say anything).

Thoughts

 

[Sycorax]

I think that the matrix is too myopic. I think there are more factors that determine the success of a move etc.

Could you elaborate? I can guess what you mean, but I'm not sure.

...but I also think there's RPS within Jab which you've talked about.

1) the opponent can ASDI + grab 2) Nothing 3) attack 4)) roll 5) regular grab

Jab goes the next step compared to grabbing, because you can choose whether to jab twice

1) peach is screwed 2) jab again, dsmash, reset neutral, grab, anything really 3) jabbing may catch the attack 4) chase 5) jab again

In other words, there is only one bad option on the surface. The problem is if she jabs twice and they do nothing 2) peach is screwed

I have no idea what you're trying to say here. What is the "RPS within Jab"? Are you talking about mixing up between jab and double jab? Or using jab then something else? Jab is too negative on shield and on hit to get anything "true" off of. You can only hope it staggers your opponent because they weren't prepared or expecting it.

Depending on a character and situation, a roll may not mean Peach wins.

Doing a second FC Nair while the opponent rolls gives Peach 19 frames of advantage and plenty of time to react. Yeah the opponent may roll to center stage, but a prepared Peach should be able to follow the roll with a grab or dash attack in my mind (I haven't actually ever done this second nair technique in game).

 

[JBW2]

Sorry for the ambiguity. I will explain each. I think you understand my views of the RPS being myopic. I mean the option you choose, even if in Peach's favor, can be bad stage positioning-wise, depending on stage and her opponent. I don't know if I need to elaborate on that too much. I'll give an example. Peach against Sheik. Since Sheik's shield is really good, it changes the RPS dynamic. Typically going for a grab, run away -> run toward sheik-> attack, or a spaced down tilt is always in Peach's favor, especially towards the edge of the stage.


With RPS within Jab. I mean all three really. I'm not against it being a bad option. I'm just explaining the reasoning of why people would use it regardless of that fact. I also think that it's an archaic method that should be used sparingly.

And for the 3rd thing (tech chasing a roll), I think you're partially wrong here. Even if your able to react, I don't think peach is that fast to catch an opponent based off of reaction all of the time, especially against Fox, Sheik, Falco, and Marth who have larger rolls. You may find yourself running into an attack. If you watch Armada, you'll realize that yes his tech chase is reaction, but a lot of it is setting up to react. You really have to guess a lot with Peach against characters like Fox, because a milisec can be the difference from a tech chase or just getting punished for approaching. Peach is not fast. I think that OOS options and grabbing will become troublesome down the line as the community gets better.

Again, I am not knocking your research. I actually agree with it and find it very interesting! I just think that the game isn't as simple, which I'm sure you'd argue as well. Hopefully, Peach can evolve in the meta. There are still some things yet to be exploited, like her double jump cancel.


I hope I've cleared some things up.

 

[Vestboy_Myst]

^ I'm confused... or maybe just myopic

in any case, several peach players do fcfair>fcnair as a mixup on shield already. im not sure about the small dash between the two aerials or how frame perfectly they've been up until now, but its been around. here is armada's comment on the fc nair / grab mixup after fc fair on shield:

"You can buffer roll OOS before the grab and also jump before the grab if your name is Fox (for example). The thing that also is important to remember is that in certain scenarios you have to remember HOW the move is used in actual matches and not only how "it could be used"

Peach doing a fair will more often than not hit with the fair pretty high on the shield and not at the very bortom. Which of course makes a difference on how much stun they will have by the time you reach the ground. Just wanted to be a bit clear with what I meant in case we have some new players here"

He points out a key flaw: its difficult to time/space fc fair on shield for maximum advantage in the heat of battle (if your opponent isn't just sitting in shield). most of the time, even at top level, you'll be landing on shield safely, but not perfectly enough to guarantee the +4 from fair.

on top of that, wanting to add a perfect dash>fc nair right after to maintain +3 makes this technique even harder. it may be the best option if executed precisely in certain scenarios, but for the majority of the time it won't totally usurp the traditional mixups. its not in the TAS realm and certainty possible to do, just a bit situational.

@ Sycorax is there another reason for the tiny dash between fc fair / fc nair in your described shield pressure, besides closing any distance they might have slid?

 

[Sycorax]

@ J JBW2 I still don't understand fully. You're doing a really poor job of making your point.

Sorry for the ambiguity. I will explain each. I think you understand my views of the RPS being myopic.
...
Again, I am not knocking your research. I actually agree with it and find it very interesting! I just think that the game isn't as simple, which I'm sure you'd argue as well.

I actually don't think I'm being myopic. I'm not trying to say the game is simple and "all you need to do is this." Of course there are situations where this won't work perfectly. There are several assumptions going into what I wrote about. The biggest one is the frame advantage. I'm saying that if you can create these frame advantages, then you get these options and here is how they interact.

I would think anyone reading this would be able to understand when this does and doesn't apply. Melee is a game of situational awareness. Situations and assumptions are basically synonymous in this context. If you can create the situation I talked about, then you can do what I wrote. If you can recognize or create certain advantageous situations, then that's is good.

I mean the option you choose, even if in Peach's favor, can be bad stage positioning-wise, depending on stage and her opponent. I don't know if I need to elaborate on that too much. I'll give an example. Peach against Sheik. Since Sheik's shield is really good, it changes the RPS dynamic. Typically going for a grab, run away -> run toward sheik-> attack, or a spaced down tilt is always in Peach's favor, especially towards the edge of the stage.

You said that you'd provide an example of how an option in this situation could be "bad stage positioning-wise" and then proceed to do the opposite, providing examples of good stage positioning options against Sheik. I actually can't think of how Peach could lose stage positioning unless the opponent goes for a roll. Roll loses to most of Peach's options though in this case of high frame advantage. Roll becomes significantly better as the frame advantage goes down, but then you should be able to realize when you do an aerial with low advantage, imo, and choose your followup accordingly.

And for the 3rd thing (tech chasing a roll), I think you're partially wrong here. Even if your able to react, I don't think peach is that fast to catch an opponent based off of reaction all of the time, especially against Fox, Sheik, Falco, and Marth who have larger rolls. You may find yourself running into an attack.

It's not a tech chase. First off, if they roll behind you, they are still plenty close for you to follow them; I think grab and dash attack are good for that. This leaves roll away. Roll away is pretty hard to follow. Against Sheik (who has a very long roll), if you do the dash->second nair then Peach technically has enough time to land a strong dash attack before Sheik is actionable. Peach has to be fast with her inputs though. Covering roll away might not be humanly possible. If the roll is cut off by a stage or platform edge though, roll away becomes worse.

I don't know what to make of the rest of that paragraph of yours. It's all over the place. You start at tech chasing, then move to what Armada does, then talk about all situations involving reactions and Fox, before finishing with OoS options.


@ Vestboy_Myst
Thanks for the input and quote. Do people really already do the FC Nair? I'd love to see a video if you know one. I've never noticed it before.

In my testing the grab works against a buffered jump (against Fox) as long as the aerial is +0 or better.

I agree though that having +4 aerials is a bit idealistic. One thing is, though, that since her Fair takes so long to start up, she can get really close to the ground before it actually comes out. The trade off is that it doesn't cover her decent so an opponent could hit her on her way down.

The reason for the dash is to close the distance. Otherwise, against most characters, they get pushed too far away to threaten grab, and dsmash would only hit once or something like that.

 

[Vestboy_Myst]

I agree though that having +4 aerials is a bit idealistic. One thing is, though, that since her Fair takes so long to start up, she can get really close to the ground before it actually comes out. The trade off is that it doesn't cover her decent so an opponent could hit her on her way down.

thats part of the reason why her fair is so good from float. the ~16 startup frames happen during the fastfall and the hitbox comes out right as you reach ground level. in addition, due to the fc you can hit near the top-half of their shield and still have time to escape (~15 frames of shield stun), albeit not with significant advantage to continue move-based pressure.

late fair runs the small risk of getting stuffed by nair oos or avoided with wd oos / roll since its slightly more telegraphed. early fair covers her descent a BIT, but more importantly is a good tool to punish an overexertion asap or to 'lock them down' in shield quickly so you can exploit their oos habits.

the smartest way to win at RPS is to know what the enemy will do next, not by trying to cover everything they could possibly do. the former is both easier and more realistic, just takes a bit of observation and intuition

 

[Vestboy_Myst]

thought some more about when to use fc nair after fair pressure. assuming the nair connects (peach 'wins' the matrix) and that you cant reliably maintain +3 (ie in realistic not ideal scenarios) mu is something to consider

against fastfallers at low/mid percents you are going to get a lot more value out of a grab or jab>grab leading directly to the uthrows and all the % that comes with them. nair could be worth going for to get them offstage but also risks resetting neutral, so stage and KB awareness is another factor. at high %s where all you need is a trade, that turnip>fcfair>fcnair>dsmash pressure @ Quetzalcoatl described sounds very scary and would require smart angling to avoid a poke/lost stock.

since +4 is going to be very rare in practice and spacies have shine oos this may be better suited for an opponent like sheik/marth. i like the idea of a successful nair creating distance for safe turnip pulls. grounded nair also leads to some nice follow ups on sheik.

as for slippery characters like luigi/ICs that one frame of dash is not going to be very helpful

 

[DavemanCozy]

@ J JBW2 I think you're giving jab too much credit. It can be useful as an interruption if opponents aren't expecting it and 2nd jab does enough KB to get tumble animation on spacies and Sheik at mid % so it could lead to something if they don't tech, but otherwise I don't see why it deserves any more credit than what the matrix shows. Jab can not only be poor on shield and ASDI'd, but even crouch cancelling beats her jab. Either way, I find jab too situational.

This post was well written and very informative. Never thought of using a second n-air after hitting an fc aerial on shield, I've been finding it especially useful if you really need to knock an opponent offstage and you're near the ledge. I noticed too that a habit I had was to go for grab -> f-throw near the ledge after an fc f-air on shield, so my opponents who played against me mostly went for roll to avoid the grab or shine (if they were a spacie) to knock me away and reset. Second N-air covers both of those options well and still does the job of tossing my opponent offstage when I want them there.

 

[JBW2]

OK. I concede. You guys seem to understand the game better than me. The more I try to explain what I was thinking considering the fact I don't regularly keep up with this post, is confusing me as well. I see what you guys are saying. Ultimately, it's a learning experience, so I don't want it to seem as though i think I know everything. So, from what I can see after skimming, which I've usually do, we all

AGREE that:

1) Jab is very situational
2) The matrix is a great analysis
3) FC NAIR after FC FAIR could be the future of peach pressure, because of shield stun, etc.
4) I should learn how to articulate my thoughts/know what I'm talking about fully before expressing my opinions


DON'T AGREE

I'm not giving jab credit. Just explaining that the game mix-up USED TO go further with jab. It was used, because people didn't get the game. Now, people are getting more educated so the use isn't as safe. All responses have been you can crouch and ASDI, etc. Up to about Upper-mid level players, this means nothing. However, that will definitely change in the future, and so will the matrix too (reference to the portions where Peach could lose if an opponent was frame perfect). My understanding was that it's like Fox's up throw up smash on a spacie. It's the go to option, but the game has evolved punish wise.


BASICALLY

I think I should have used a different word then myopic, because I wasn't trying to insult the post. I was trying to say in the best way I could what Vestboy tagged with what Armada said. I was really just trying to say that.


When I have time, I will definitely go back and read the post in its entirety, instead of glancing at it. My analysis was only based on that. My apologies.

 

[Quetzalcoatl]

I was going to make a new thread about this but it was too relevant to this topic:

Picture this:

Peach comes in on Sheik with a FC fair, nicely spaced but close enough that either player can grab the other.
Instead of Down smashing, grabbing or jabbing (…) Peach starts to.. Charge forward smash?

Sheik attempts to shield grab but the lean back on the forward smash evades the grab every time and Peach releases the forward smash after recognising the grab attempt.

Sheik attempts to hold shield.. and Peach continues to hold the charge while the shield depletes. Sheik either takes the Fsmash on a rapidly shrinking shield or rolls away.

Sheik attempts to side step, but Peach easily holds the charged smash until the side step finishes.

Sheik attempts to aerial out of shield, however is cleanly defeated every time by the charged forward smash thanks to the disjointed hitbox.

I briefly tested this with a friend and it was surprisingly effective. Charging the smash allows you to counter any out of shield option at the appropriate time, the only thing it would traditionally lose to is grab, which it auto-evades. Rolling simply resets to a neutral state as Fsmash has a pretty short wind down.

Of course this would be mixed up with your usually FC routine but its definitely worth exploring more. Usefulness against certain characters probably differs too.

 

[N JA]

Peach comes in on Sheik with a FC fair, nicely spaced but close enough that either player can grab the other.
Instead of Down smashing, grabbing or jabbing (…) Peach starts to.. Charge forward smash?

This is wonderful. I love the theorycrafting that's been going on with lesser used moves since Peach has unexplored mixups and can use all the help she can get. How exactly do you beat Sheik nair oos if you're charging the smash, wouldn't you get hit before the release?

 

[Sycorax]

Lol, that fsmash thing sounds wonky, but just barely good enough to work once. Spacing would definitely be really important.

 

[CAUP]

I've been thinking about a forward thrown turnip for shield pressure. It bounces inside the opponents shield for a while and if they do literally anything they will get hit. Thoughts?

 

[blue cheez]

I'm a fairly technical peach. I have very fast, consistent FC nairs. (I can FC-nair out of sheild, beat sheildgrab with FCnair +grab, etc)
I cannot, for the life of me, beat a shieldgrab (in 20xx) with FCd nair after FC'd fair.
Are you sure this is actually possible with a controller? Has anyone been able to replicate it?

 

[Sycorax]

I'm a fairly technical peach. I have very fast, consistent FC nairs. (I can FC-nair out of sheild, beat sheildgrab with FCnair +grab, etc)
I cannot, for the life of me, beat a shieldgrab (in 20xx) with FCd nair after FC'd fair.
Are you sure this is actually possible with a controller? Has anyone been able to replicate it?

I'm sure it an option afforded by the mechanics of Melee and the GameCube controller. Whether it is human possible. I'm not 100% sure. I'm just 98% sure. You just fall with an FC fair, and then do an FC nair right after it. Both of those are not hard by themselves so the only added difficulty of doing them in succession is making sure to be quick about it.

Just to make sure, you're not trying to dash between the FC fair and the FC nair right? That would make it impossible to beat grab with. Also, FC nair is +3 on shield, so you can't beat grab if the first aerial was an FC nair.

Here's a video I recorded showing it's possible. Sorry the frame advance is so jerky. You can play it in slow motion and pause to see individual frames.

 

[blue cheez]

Do you think there is an easy way to check if I'm getting +4 on a FC fair?
I'm trying to confirm that this stuff is possible with a controller (in 20XX 3.02) and it's difficult for me to troubleshoot what I'm doing wrong.
So for example I don't know if I'm not +4 OR if I'm not doing a perfect nair timing immediatly following the +4.

Also, FC nair is +3 on shield, so you can't beat grab if the first aerial was an FC nair.

I thought the point of (FC Fair -> Nair) was to be an option that simultaneously beats Opponent Roll while not losing to Opponent Wait.
But you have on the chart that if you're frame perfect (+4) grab beats ALL options already. So if I'm going to go for frame-perfect techskill that only works when I'm +4, why on earth would I go for a nair INSTEAD of a grab?

 

[Sycorax]

Do you think there is an easy way to check if I'm getting +4 on a FC fair?
I'm trying to confirm that this stuff is possible with a controller (in 20XX 3.02) and it's difficult for me to troubleshoot what I'm doing wrong.
So for example I don't know if I'm not +4 OR if I'm not doing a perfect nair timing immediatly following the +4.

I think the best way is frame counter. For me, since I've done so much testing with it, it's really obvious whether I hit with frame 16, or frame 17 of fair. It's also obvious if I hit with a later hit than frame 17, but I don't know how to distinguish those individual frames. So like, I could set the frame counter to record float fair frames. Then I could do FC fairs on shield. If I hit the shield with frame 16 and the counter reads 16, then I know I landed the frame after Fair hit. If I hit the shield with frame 17 and the counter reads 17, then I know I landed the frame after fair hit.
You could also set frame counters for relevant FC nair animations and make sure they match up with what I outlined in the OP. I would suggest counting Landing, empty float, and float nair. Those seem to be the most important things to count.
FYI, when using frame counter, float nair is different from regular nair. Same for other aerials.

I thought the point of (FC Fair -> Nair) was to be an option that simultaneously beats Opponent Roll while not losing to Opponent Wait.
But you have on the chart that if you're frame perfect (+4) grab beats ALL options already. So if I'm going to go for frame-perfect techskill that only works when I'm +4, why on earth would I go for a nair INSTEAD of a grab?

That's a good point. Grab even beats frame percent shine oos from Fox @_@ Something I didn't consider when making this though was spot dodge. It's generally a stupid option against Peach but if your opponent is playing a character with a fast spot dodge, e.g. Fox, Falco, Sheik, and they find out you like to grab a lot after FC aerials, grab will lose to buffered spot dodge. FC nair beats that. There's also the psychological factor of quick FC nair pressure being scary as ****.

 

[blue cheez]

I have some experience in game theory, so I'm interested in using this information to figure out the optimal mixup for peach.

I plan on doing a big write-up introducing game theory and how it can be used in melee. And I think I will use some of this peach data as a starting point.

The math itself is very easy, but there are a lot of conceptual ideas to mess up - so I want to make a nice clear writeup so there isn't any confusion.

In order to do the most standard gametheory math successfully - we need to a numeric that tells us how "good" each outcome is. It doesn't have to be perfect - and just needs to be a good rule of thumb. So in these situations, the most obvious numeric to use is the total change in percent each player receives. By doing so we are seeking to find the mixup that optimizes percentage (which is a sensible thing to do when both players are below kill perecent).
Sorry if this is a little bit confusing. I'm being intentionally vague here, because there are quite a bit of details I'm brushing under the rug. The details don't mess up our math, but would require a lot of time to explain. so just trust me for now if you're confused and when my big writeup comes out, you'll understand.

So basically, I want a matrix like the ones the OP created but with "percent damage" in each of the different elements of the matrix.

As a simple example to make the math more clear - let's consider an unrealistic case where peach only has grab or downsmash as options, and fox can only either wait or grab:
Peach/Fox Grab Wait
Grab +80% + 80%
Downsmash +30% -40%

The plus in +(x%) indicates that Fox is taking damage, while the minus in -(x%) indicates Peach taking damage.
What this matrix says is the following:
If both Peach and Fox grab, Peach deals 80 damage (which might make sense if fox is around 30% on FD -allowing for a chaingrab)
If peach grabs, and fox waits, the same thing happens - Peach deals 80 damage (nothing stops peach from grabbing).
If peach uses downsmash and fox grabs, Peach deals 30%
If peach uses downsmash and fox waits, Fox deals 40% (he points his sheild down and punishes the lag of dsmash)
Here we see that if Peach is playing optimally, a mixup is not necessary - as a "pure strategy" of grabbing 100% of the time will always deal the most damage for her. Then since we know Peach always deals 80% - it doesn't even matter what Fox's decision is, as he will always take 80% if Peach is playing optimally.


So, as you can see from this example, there are a lot of assumptions you must add for a "percent matrix" to work. But this is FINE, as long as you are careful in your assumptions. You just need a matrix for EACH particular situation.

Here's an example payout matrix that I came up with for (+4 frame advantage) with Fox at ~40% on FD. I did not spend much time working out this matrix perfectly (so it's probably very wrong) - but did it as a general idea of what I'm looking for.

With a matrix like this we can work out the optimal mixup for peach in this particular situation. If you're curious, the "optimal mixup" with these numbers says that Peach should grab 33% of the time, jab 32% of the time, downsmash 0% of the time, and wait 35% of the time. Fox should grab 16% of the time, wait 18% of the time, and sidestep 66% of the time. If peach plays at this optimal strategy then she will deal 14% more than fox will on average. Each of these "optimal mix ups" are called the "Nash Equilibrium" - and if either player chooses something else OTHER than this optimal mixup - then the other player has the option of leaving their nash equilibrium to capitalize on their unsafe mixup (but consequently the capitalizer then risks his opponent capitalizing on them).


So I'm interested in talking and working out matrices like this for the most common situations for peach. Additionally, I have only been able to grab a buffered roll with consistency of 1/10 (btw OP, the best way I could test for +4 frame advantage irl was just try to grab a buffered roll out of a FCd fair. But I haven't tried your frame counter idea yet.)
Honestly for now I think +4 frame advantage charts aren't worth our time until players can (at least somewhat) consistently do them.

I would like to see +1, +2, and +3 frame advantage charts for Fox on FD at and below chaingrab percent.

Additionally, in the meantime, if you are too impatient waiting for me to write up my melee gametheory lesson - here are some incomplete resources you can read in the meantime:
1. I attached an old, incomplete introduction to game theory (the goal of the writeup was to discuss the creation of a "best mixup" in the pokemon sense - but the intoduction can still be applied to melee.)
2. Here is a link to the nash equilibrium calculator I used to get those percentages for the "optimal mixup."
http://banach.lse.ac.uk/
Choose zero-sum and just paste in the values for your matrix to find the "optimal mixup."

 

[CAUP]

Very interesting, PLEASE write that write up

 

[Sycorax]

I have some experience in game theory, so I'm interested in using this information to figure out the optimal mixup for peach.

I plan on doing a big write-up introducing game theory and how it can be used in melee. And I think I will use some of this peach data as a starting point.

The math itself is very easy, but there are a lot of conceptual ideas to mess up - so I want to make a nice clear writeup so there isn't any confusion.

In order to do the most standard gametheory math successfully - we need to a numeric that tells us how "good" each outcome is. It doesn't have to be perfect - and just needs to be a good rule of thumb. So in these situations, the most obvious numeric to use is the total change in percent each player receives. By doing so we are seeking to find the mixup that optimizes percentage (which is a sensible thing to do when both players are below kill perecent).
Sorry if this is a little bit confusing. I'm being intentionally vague here, because there are quite a bit of details I'm brushing under the rug. The details don't mess up our math, but would require a lot of time to explain. so just trust me for now if you're confused and when my big writeup comes out, you'll understand.

So basically, I want a matrix like the ones the OP created but with "percent damage" in each of the different elements of the matrix.

As a simple example to make the math more clear - let's consider an unrealistic case where peach only has grab or downsmash as options, and fox can only either wait or grab:
Peach/Fox Grab Wait
Grab +80% + 80%
Downsmash +30% -40%

The plus in +(x%) indicates that Fox is taking damage, while the minus in -(x%) indicates Peach taking damage.
What this matrix says is the following:
If both Peach and Fox grab, Peach deals 80 damage (which might make sense if fox is around 30% on FD -allowing for a chaingrab)
If peach grabs, and fox waits, the same thing happens - Peach deals 80 damage (nothing stops peach from grabbing).
If peach uses downsmash and fox grabs, Peach deals 30%
If peach uses downsmash and fox waits, Fox deals 40% (he points his sheild down and punishes the lag of dsmash)
Here we see that if Peach is playing optimally, a mixup is not necessary - as a "pure strategy" of grabbing 100% of the time will always deal the most damage for her. Then since we know Peach always deals 80% - it doesn't even matter what Fox's decision is, as he will always take 80% if Peach is playing optimally.


So, as you can see from this example, there are a lot of assumptions you must add for a "percent matrix" to work. But this is FINE, as long as you are careful in your assumptions. You just need a matrix for EACH particular situation.

Here's an example payout matrix that I came up with for (+4 frame advantage) with Fox at ~40% on FD. I did not spend much time working out this matrix perfectly (so it's probably very wrong) - but did it as a general idea of what I'm looking for.

View attachment 90703


With a matrix like this we can work out the optimal mixup for peach in this particular situation. If you're curious, the "optimal mixup" with these numbers says that Peach should grab 33% of the time, jab 32% of the time, downsmash 0% of the time, and wait 35% of the time. Fox should grab 16% of the time, wait 18% of the time, and sidestep 66% of the time. If peach plays at this optimal strategy then she will deal 14% more than fox will on average. Each of these "optimal mix ups" are called the "Nash Equilibrium" - and if either player chooses something else OTHER than this optimal mixup - then the other player has the option of leaving their nash equilibrium to capitalize on their unsafe mixup (but consequently the capitalizer then risks his opponent capitalizing on them).


So I'm interested in talking and working out matrices like this for the most common situations for peach. Additionally, I have only been able to grab a buffered roll with consistency of 1/10 (btw OP, the best way I could test for +4 frame advantage irl was just try to grab a buffered roll out of a FCd fair. But I haven't tried your frame counter idea yet.)
Honestly for now I think +4 frame advantage charts are worth our time until players can consistently do them.

I would like to see +1, +2, and +3 frame advantage charts for Fox on FD at and below chaingrab percent.

Additionally, in the meantime, if you are too impatient waiting for me to write up my melee gametheory lesson - here are some incomplete resources you can read in the meantime:
1. I attached an old, incomplete introduction to game theory (the goal of the writeup was to discuss the creation of a "best mixup" in the pokemon sense - but the intoduction can still be applied to melee.)
2. Here is a link to the nash equilibrium calculator I used to get those percentages for the "optimal mixup."
http://banach.lse.ac.uk/
Choose zero-sum and just paste in the values for your matrix to find the "optimal mixup."

Haha! That's cool as ****, dude. I took a micro economics class about two years ago so I have an introductory understanding of game theory. A project like the would be deeply interesting to say the least, but I'm not sure how useful it would be. I can also think of some possible hindrances.

For one, I'm not sure how appropriate it is to use percentages as the measure for how good an option is. Size of the punish varies so much by the opponent's percent and the stage that using percent seems misleading. If some sort of subjective relative return value was used instead, would it still work? What I mean is something like "Grab is generally twice as rewarding as dsmash and three times as rewarding as jab so the rewards for those should be 3, 1.5, 1." Or something like that, would the process still work? Those values aren't accurate to real life, but just an example.

Second, would recognizing that the opponent Fox favors certain options sub-optimally change the payoff? With the theories you presented, I don't think it wouldn't matter. However, in real life, it makes sense that you could leverage a greater punish by biasing your own weights depending on the observed weights of the opponent. Donkeyspace basically. This may be beyond the scope of what you have in mind, and I don't know the theory or math to support it. I feel like it would be some combination of game theory and Bayesian statistics.

Either way, I'm down to help you out with the project. If there's something you need from me, let me know.

 

[Vestboy_Myst]

I have some experience in game theory, so I'm interested in using this information to figure out the optimal mixup for peach.

I plan on doing a big write-up introducing game theory and how it can be used in melee. And I think I will use some of this peach data as a starting point.

The math itself is very easy, but there are a lot of conceptual ideas to mess up - so I want to make a nice clear writeup so there isn't any confusion.

In order to do the most standard gametheory math successfully - we need to a numeric that tells us how "good" each outcome is. It doesn't have to be perfect - and just needs to be a good rule of thumb. So in these situations, the most obvious numeric to use is the total change in percent each player receives. By doing so we are seeking to find the mixup that optimizes percentage (which is a sensible thing to do when both players are below kill perecent).
Sorry if this is a little bit confusing. I'm being intentionally vague here, because there are quite a bit of details I'm brushing under the rug. The details don't mess up our math, but would require a lot of time to explain. so just trust me for now if you're confused and when my big writeup comes out, you'll understand.

So basically, I want a matrix like the ones the OP created but with "percent damage" in each of the different elements of the matrix.

As a simple example to make the math more clear - let's consider an unrealistic case where peach only has grab or downsmash as options, and fox can only either wait or grab:
Peach/Fox Grab Wait
Grab +80% + 80%
Downsmash +30% -40%

The plus in +(x%) indicates that Fox is taking damage, while the minus in -(x%) indicates Peach taking damage.
What this matrix says is the following:
If both Peach and Fox grab, Peach deals 80 damage (which might make sense if fox is around 30% on FD -allowing for a chaingrab)
If peach grabs, and fox waits, the same thing happens - Peach deals 80 damage (nothing stops peach from grabbing).
If peach uses downsmash and fox grabs, Peach deals 30%
If peach uses downsmash and fox waits, Fox deals 40% (he points his sheild down and punishes the lag of dsmash)
Here we see that if Peach is playing optimally, a mixup is not necessary - as a "pure strategy" of grabbing 100% of the time will always deal the most damage for her. Then since we know Peach always deals 80% - it doesn't even matter what Fox's decision is, as he will always take 80% if Peach is playing optimally.


So, as you can see from this example, there are a lot of assumptions you must add for a "percent matrix" to work. But this is FINE, as long as you are careful in your assumptions. You just need a matrix for EACH particular situation.

Here's an example payout matrix that I came up with for (+4 frame advantage) with Fox at ~40% on FD. I did not spend much time working out this matrix perfectly (so it's probably very wrong) - but did it as a general idea of what I'm looking for.

View attachment 90703


With a matrix like this we can work out the optimal mixup for peach in this particular situation. If you're curious, the "optimal mixup" with these numbers says that Peach should grab 33% of the time, jab 32% of the time, downsmash 0% of the time, and wait 35% of the time. Fox should grab 16% of the time, wait 18% of the time, and sidestep 66% of the time. If peach plays at this optimal strategy then she will deal 14% more than fox will on average. Each of these "optimal mix ups" are called the "Nash Equilibrium" - and if either player chooses something else OTHER than this optimal mixup - then the other player has the option of leaving their nash equilibrium to capitalize on their unsafe mixup (but consequently the capitalizer then risks his opponent capitalizing on them).


So I'm interested in talking and working out matrices like this for the most common situations for peach. Additionally, I have only been able to grab a buffered roll with consistency of 1/10 (btw OP, the best way I could test for +4 frame advantage irl was just try to grab a buffered roll out of a FCd fair. But I haven't tried your frame counter idea yet.)
Honestly for now I think +4 frame advantage charts are worth our time until players can consistently do them.

I would like to see +1, +2, and +3 frame advantage charts for Fox on FD at and below chaingrab percent.

Additionally, in the meantime, if you are too impatient waiting for me to write up my melee gametheory lesson - here are some incomplete resources you can read in the meantime:
1. I attached an old, incomplete introduction to game theory (the goal of the writeup was to discuss the creation of a "best mixup" in the pokemon sense - but the intoduction can still be applied to melee.)
2. Here is a link to the nash equilibrium calculator I used to get those percentages for the "optimal mixup."
http://banach.lse.ac.uk/
Choose zero-sum and just paste in the values for your matrix to find the "optimal mixup."

very neat, and exciting that you are starting with peach. i took some courses in game theory as an undergrad so even if i can't help, i'd love to see your work on this. can you think of any areas you could use some basic assistance in?

 

[blue cheez]

For one, I'm not sure how appropriate it is to use percentages as the measure for how good an option is. Size of the punish varies so much by the opponent's percent and the stage that using percent seems misleading. If some sort of subjective relative return value was used instead, would it still work? What I mean is something like "Grab is generally twice as rewarding as dsmash and three times as rewarding as jab so the rewards for those should be 3, 1.5, 1." Or something like that, would the process still work? Those values aren't accurate to real life, but just an example.

Short-answer:
Probably will be necessary if we go really, really deep into this stuff. But way too hard for now. Also bad for understanding what's going on.

Long-answer:
Using a heuristic (heuristic = "rule of thumb for what's good") for "payoff" (payoff = the numbers in the matrix) other than percent is something that we could consider - but this is a very hard thing to do and you lose a lot of intuition in doing so. Additionally (I believe) it's only necessary for "patching" multiple mixup games together. So for example, when (both) players are at a high percent, percentage doesn't matter much - and all that really matters is securing the stock. But instead of using percent, we might be able to get away with just using "stocks" as our heuristic payoff. But now that we have a good idea of how good certain situations are at high percents with our stock-heuristic matrix - at low-percent, we can then swap-out our percent-heuristic with a hybrid heuristic that weights percent in terms of how "good" it is based on how helpful it is for the stock-heuristic.
There's more to be said here, but I'll talk about it in the writeup.

Second, would recognizing that the opponent Fox favors certain options sub-optimally change the payoff? With the theories you presented, I don't think it wouldn't matter.

This is an interesting question - but would likely require a proof to give a satisfying answer. Doing a suboptimal punish WILL change the nash equilibrium, yes. Can you intentionally pick worse punishes so that it messes up the standard nash equilibrium? YES. Can this create situations where you could do better on average by picking worse punishes. I really, really doubt it. I would have to do a proof to prove it - but I think that the average gains you can get by exploiting your opponent being off Nash will always be worse because you had to hurt your own payoffs to do so. And I expect that the losses in hurting your payoffs will always hurt you more than it hurts your opponent for being off Nash. It's certainly worth looking into - because it would blow my ****ing mind if there were situations where playing inarguably worse actually helps you.

However, in real life, it makes sense that you could leverage a greater punish by biasing your own weights depending on the observed weights of the opponent. Donkeyspace basically. This may be beyond the scope of what you have in mind, and I don't know the theory or math to support it. I feel like it would be some combination of game theory and Bayesian statistics.

If you know with 100% certainty what your opponents mixup is (and it's not at Nash equilibrium), you can easily capitalize on it by going off your Nash equilibrium. I will go into detail in the writeup about different techniques for this - but the most important thing to note is that this will ALWAYS be risky (because once you're off Nash, your opponent can also capitalize on you). You are assuming that your opponent is not trying to trick you. In the melee community this is sometimes referred to as "conditioning" (Although you can also argue conditioning is a little diffferent...we can talk about this later).

If you don't know with 100% certainty what your opponents mixup is.....yes you'd use some sort of Bayesian maximum likelihood algorithm. It might be interesting to see if there are easy techniques for humans to be smart Bayesian thinkers in a live setting. But I'm honestly a skeptic that any hardcore math would be useful for a human in a live setting. (But maybe we can use math to come up with an easy "trick" to use in a live setting.) But I think the Bayesian thinking is really an afterthought. Something to be considered much, much, much, much later. (Realistically I think it'll only be useful if we start thinking about developing a MELEE AI that is forced to have a human reaction time.)

Either way, I'm down to help you out with the project. If there's something you need from me, let me know.

I would like to see +1, +2, and +3 frame advantage charts for Fox on FD at and below chaingrab percent. (BTW, anyone reading can help out with this!) Simply using "percent-damage" as our heuristic is the best starting point. (Using a more advanced heuristic is interesting, but that's something that should be saved for a much, much later time.)

There are a couple of ways of making these charts:

The first way is to theorycraft it. So figure out exactly the optimal punish for a given situation. An easy example I used earlier is peach's chaingrab. Assume both players get the optimal punish - or don't assume that, and make it clear what you're assumptions are. (So for example maybe fox can buffer shine OoS but he can't buffer waveshine OoS). I'm not an expert in all of the details here - (I don't know exactly what percent peach's chaingrab works at, or when you can combo upthrow into downsmash at low percent, or how much damage a downsmash will do to a fastfaller with good/bad DI etc.) so it would be very helpful to me if other players could work out these matrices.
There is one annoying problem theorycrafting these percentages - and that is when mixups lead to further mixups. It's not that bad when the second mixup is simple. For example, let's say that we're not on FD but on untransformed pokemon stadium under a platform. If peach gets a grab on fox, how much percentage is the grab worth under a platform? Here we can theorycraft what peach's best mixup is for techchasing and then we'd say the grab is worth the average value of the percent damage obtained by doing the mixup. And then we can use that average value as the value for grab in our original, first payoff matrix.


(*/Advanced Digression -You should probably ignore this *)
But when mixups lead into the same exact mixup.......I suspect this is extraordinarily hard to analytically solve for - but pretty easy to get an accurate guess. (I'll go into detail here on how you'd calculate this numerically, but you can just ignore this since this is fairly advanced stuff - and it's very easy to contribute to this project without understanding this part.) The way you'd get a good guess is by doing it perturbatively. Find the total expected payoff of the Nash Equilibrium mixup assuming the value of the reset is zero (arbitrary guess. It may or may not be a good guess. Will discuss later.) Now plug replace the zero with the newly found "total expected payoff" for the reset situation in the matrix. Recalculate the Nash Equilibrium mixup. Take the newly, newly found "total expected payoff" and replace with the old, newly found "total expected payoff." Recalculate the Nash Equilibrium mixup. ...Rinse and Repeat and hopefully this will converge. It might converge depending on our initial guess - but then we just play around with our initial guesses and see how it converges depending on the initial guess. It might not converge (then idk what to do, but let's worry about that when/if that ever happens).
(*/End of Advanced Digression - You should've ignored this^ *)

The second way (to make these payoff matrix charts) is to experimentally verify them. Watch videos of Armada and calculate how much (on average) he obtains by getting a grab on FD. It will be tempting to FIRST check what Armada's mixup options are - but be very careful here - as this is what we will do absolutely LAST (if you just copy what Armada does, you'll never learn why he's doing it - OR even if what he's doing is truly optimal. That is, it might be that he has such a good combo game that it doesn't even matter if his mixups are suboptimal.)
In my opinion, it's useful to watch and study Armada, but it's also helpful to observe lower-level peach's play. In addition to looking at how much Armada gets off of a grab on battlefield under a platform, it'd also be helpful to see how much the average high-level (top-200) peach gets off of a grab on battlefield under a platform (or any particular"punish" situation).


I'm hoping that this game theory stuff takes off. Because if it does, we can coordinate with the people working on hardware that makes live input-recordings of games. We might be able to get them to tweak their hardware to make it easy for us collect the data we need. So if Armada plays a game with this hardware installed we could know how much average damage he does on against fox on FD after hitting fox's sheild with +3 frame advantage (without needing us to watch a recording of the match and gather the data by hand).
Additionally, we could use the data to see if Armada is actually playing at our theoretical nash equilibrium. And if he's NOT then either he's not playing optimally, or his brain is considering something more similar to the hybrid-heuristic I discussed earlier. (Much, much, much later) We might be able to then use top players to help us "tune" our hybrid-huristic. An interesting thing to note if anyone starts looking at mixups of top players. If armada is doing an option LESS than you expect, that (may) mean that that option is actually much BETTER than we expect. In general you "hide" your powerful options, and you spam your counter to their counter of your most power option. So if somehow downsmash always leads to a kill for Armada, then against an aware opponent Armada should spend more time trying to counter THEIR counter to his downsmash and less time downsmashing.


Anyway, I'll go more indepth with the introduction to this stuff in my writeup. I hope to get it down this coming weekend - but quality>timeliness so no promises.

 

[blue cheez]

I am extremely happy that there is any interest in this. Having a single person to talk these ideas out with is extremely helpful in itself. Having multiple people is fantastic.
If you are at all interested in this stuff, I strongly encourage you to contribute. To the people already familiar with game theory, read my previous posts and you might already be able to contribute a significant amount.
If you are not that familiar with game theory, do not worry! There are already things in my previous post you can help out with without having any experience in game theory. AND, you do not even need to know a lot of math to learn game theory. You only need to know algebra to be able to learn game theory - and my write up (I hope) will be sufficient for you to learn what's necessary for this stuff to be helpful for your melee game (and for you to contribute).

Here are some examples of VERY helpful things people can do if they're not that familiar/comfortable with gametheory:

• try working out the percent damage dealt for different situations/characters

I want a matrix for many different situations, but it's a lot of work to figure out what these are. Here is an example I was thinking about:

As you can see I don't really know what happens in a lot of situations. (And even the places where there are numbers, I'm not sure how right these are.)
As an example of something I don't know in the chart: Does grab beat jump with +1 frame advantage? Maybe there's enough time for fox to jump over the grabhitbox. I would have to test it. There are a lot of things like this that would take a lot of time for me to do alone. But with a group it would be pretty easy!
Additionally, I think it's important for us to collect information that's useful for midlevel Peaches as well. It's very hard to consistently get +4 frame advantage with Fair, so it's important to collect data on situations where the peach isn't perfect but still has some frame advantage (using +4 frame advantage is hard as ****, try grabbing a buffered roll in 20XX with peach, it took FOREVER for me to even have 1/10 consistency).


And as described in the previous post, there are a number of ways we could fill out these charts. You can theorycraft it, you can find the values by watching top players, or you can do a mix of both.

• try collecting data of armada [or other top peach players] and record how much damage they do in certain situations. So find out how much damage armada does after a grab when fox is at 0% under a battlefield platform. )

We can theorycraft what each of those elements in the matrix should be. But we can also watch Armada and find the average of what happens in each situation.


Thanks for reading, and I look forward to working with everyone! I think a lot of this stuff will be very useful for pushing melee to its limits.

 

[Sycorax]

This is an interesting question - but would likely require a proof to give a satisfying answer. Doing a suboptimal punish WILL change the nash equilibrium, yes. Can you intentionally pick worse punishes so that it messes up the standard nash equilibrium? YES. Can this create situations where you could do better on average by picking worse punishes. I really, really doubt it. I would have to do a proof to prove it - but I think that the average gains you can get by exploiting your opponent being off Nash will always be worse because you had to hurt your own payoffs to do so. And I expect that the losses in hurting your payoffs will always hurt you more than it hurts your opponent for being off Nash. It's certainly worth looking into - because it would blow my ****ing mind if there were situations where playing inarguably worse actually helps you.

A typo caused a bit of misunderstanding. That second quotes should have said "...I don't think it would matter." Your response to the third quote makes me think we understand each other. The simplest example of what I meant would be if you're playing RPS with someone and you know they'll always throw rock. You should adjust your strategy to be to always throw paper.

I would like to see +1, +2, and +3 frame advantage charts for Fox on FD at and below chaingrab percent.

To be clear, you want items in the chart to be percentages, right?

Another thought on this, if we wanted to go all the way with this, we'd need a chart at each of these advantages, for each character, at like 10 different possible percents, on each stage. 4*12*10*6=2880.... That's a lot of ****ing theorycrafting and/or video data mining.

here are a couple of ways of making these charts:

The first way is to theorycraft it. So figure out exactly the optimal punish for a given situation.

The second way (to make these payoff matrix charts) is to experimentally verify them. Watch videos of Armada and calculate how much (on average) he obtains by getting a grab on FD.

I like the second way more than the first in terms of validity, but not so much in terms of amount and kind of work X( Either way, there is a big problem with defining a punish. It's a very big grey area in defining how much you earn off a punish. An example would be something like: Armada lands a dsmash after the FC aerial and gets a big true combo for it. The combo ends with the spacie player in the corner of the stage, and when trying to fight his way out, eats two more hits. The true combo ends with the spacie player in a distinct positional disadvantage and Armada quickly and directly punishing them for it. I realize that the percentage values are only there to give a rough idea of the strength of the punish, but it's still incredibly fuzzy where you draw the line between a hit being part of a punish and a hit that is the start of something new.
Another problem is a situation like this: Peach gets FC fair, dsmash, the space goes flying off stage, Armada nairs them and then hogs ledge. The space player took only like 27%, but lost the stock. How do you rate the strength of that punish?

I'm hoping that this game theory stuff takes off. Because if it does, we can coordinate with the people working on hardware that makes live input-recordings of games. We might be able to get them to tweak their hardware to make it easy for us collect the data we need. So if Armada plays a game with this hardware installed we could know how much average damage he does on against fox on FD after hitting fox's sheild with +3 frame advantage (without needing us to watch a recording of the match and gather the data by hand).
Additionally, we could use the data to see if Armada is actually playing at our theoretical nash equilibrium. And if he's NOT then either he's not playing optimally, or his brain is considering something more similar to the hybrid-heuristic I discussed earlier. (Much, much, much later) We might be able to then use top players to help us "tune" our hybrid-huristic. An interesting thing to note if anyone starts looking at mixups of top players. If armada is doing an option LESS than you expect, that (may) mean that that option is actually much BETTER than we expect. In general you "hide" your powerful options, and you spam your counter to their counter of your most power option. So if somehow downsmash always leads to a kill for Armada, then against an aware opponent Armada should spend more time trying to counter THEIR counter to his downsmash and less time downsmashing.

That would be soooo cool @_@

 

[blue cheez]

A typo caused a bit of misunderstanding. That second quotes should have said "...I don't think it would matter." Your response to the third quote makes me think we understand each other. The simplest example of what I meant would be if you're playing RPS with someone and you know they'll always throw rock. You should adjust your strategy to be to always throw paper.

I thought you were asking a much more subtle question. I think I understand what you were asking now, but I don't think we're exactly on the same page. I expect any confusion you have will be resolved by my future writeup. In general it'd be more efficient for me to explain everything correctly once than to have to clarify myself to a whole bunch of individuals, but let's try to briefly resolve this:

A Nash Equilibrium does this unintuitive thing where it makes your opponents decisions "useless." In a game of rock-paper-scissors, the Nash Equilibrium is a mixed strategy of 1/3, 1/3, 1/3. If you pick each randomly with equal weight, you will win exactly 1/3 of the time. And you will win 1/3rd of the time no matter what weird strategy your opponent has. You've managed to pick a strategy that cannot be capitalized on. If he picks rock 100% of the time - if you are at Nash equilibrium you will win 1/3rd of the time. If he also does (1/3, 1/3, 1/3) you will still win 1/3rd of the time.
A nash equilibrium is "optimal" in that it is the "safest." It prevents your opponent from capitalizing on your decision using "mindgames". But it is not "optimal" in the sense that it deals the most damage over any strategy.
Here's an example:
If your opponent picks 100% rock, he is not at Nash Equilibrium. But the only way to capitalize on this is to go off Nash Equilibrium and try to capitalize on his dumb strategy. The most optimal strategy to counter 100% rock is 100% paper. But as soon as you decide on 100% paper, you are not doing the (1/3,1/3,1/3) Nash Equilibrium and you're unsafe - ie you risk you opponent turning the tables and capitalizing on your dumb strategy. So it may look like someone is doing 100% rock, but they may be trying to mindgame you into going off your nash equilibrium - hoping you will switch to something like 100% paper - so that they can switch to scissors.

The simplest example of what I meant would be if you're playing RPS with someone and you know they'll always throw rock. You should adjust your strategy to be to always throw paper.

Hopefully now it's clear that you should not necessarily always throw paper. You should only do so if you are absolutely certain your opponent is playing sub-optimally. In always throwing paper, you subject yourself to the risk that being at Nash Equilibrium prevents.

To be clear, you want items in the chart to be percentages, right?

I want the items in the chart to be percentages.

Another thought on this, if we wanted to go all the way with this, we'd need a chart at each of these advantages, for each character, at like 10 different possible percents, on each stage. 4*12*10*6=2880.... That's a lot of ****ing theorycrafting and/or video data mining.

Technically, yes, there are a lot of combinations. But I'm fairly confident that most of those situations are identical to each other.
I don't think this will turn into memorizing thousands of mixed strategies for every situation. I think we are instead collecting a lot of situations to learn about what is and isn't important in a particular situation. We're going hardcore into the details of a couple of important situations so that we can learn how we can improve. It's mostly so that we can get an accurate intuition once we change things up slightly. So you learned the proper NE (Nash Equilibrium) against Fox on FD for slight frame advantage. Well, I bet that NE will be pretty similar to Falco/Falcon in the middle of dreamland.

And we can look once at the differences between Fox and Falco's matrices - and see if there's a big difference (I doubt it). Or we can look at the differences between being in the middle of dreamland and being on FD (I bet they're very similar). Bottom line is I don't think you need to sweat the small differences - it's better to get an idea of the whole big picture.

I would say the most important things to analyze are:
Frame advantage [0, +1, +2, +3] (4 states)
Platform locations [no platforms, platform above] (2 states)
Fastfaller (2 states)
Chaingrab/combo percent (2 states)

So 32 possibilities. And even then I think a lot of them will be redundant. I'm not sure, but I think there's not a very big difference between +1 and +2 frame advantage in terms of what additional options it gives you - so (if that we true), we'd be down to 16 possibilities. And I really think that 16 is overstating how many of these will be interesting. I'm only expecting in-the-end you'll only need know understand a couple of nash equilibriums.

I like the second way more than the first in terms of validity, but not so much in terms of amount and kind of work X( Either way, there is a big problem with defining a punish. It's a very big grey area in defining how much you earn off a punish. An example would be something like: Armada lands a dsmash after the FC aerial and gets a big true combo for it. The combo ends with the spacie player in the corner of the stage, and when trying to fight his way out, eats two more hits. The true combo ends with the spacie player in a distinct positional disadvantage and Armada quickly and directly punishing them for it. I realize that the percentage values are only there to give a rough idea of the strength of the punish, but it's still incredibly fuzzy where you draw the line between a hit being part of a punish and a hit that is the start of something new.
Another problem is a situation like this: Peach gets FC fair, dsmash, the space goes flying off stage, Armada nairs them and then hogs ledge. The space player took only like 27%, but lost the stock. How do you rate the strength of that punish?

I agree this is tricky. But we can do a couple of "duct tape fixes" to get it to mostly work. I want to say that our convention should be that ANY damage that is done is recorded until the neutral game is reset. My intuition is that if it puts you in a situation that actually gives peach an advantage, then we will see it when we calculate the average. The one tricky part that I am struggling with is dealing with trading. If peach deals 40% while fox deals 20% before it's reset to neutral, is that +20%? That's clearly very different from peach only dealing 20%. I think for now when people data record, it would be helpful make a note of these things - but we probably won't use that information for a while. We might be able to handle this information by making the name no-longer zero sum....but I'd have to think about it more.

Peach gets FC fair, dsmash, the space goes flying off stage, Armada nairs them and then hogs ledge. The space player took only like 27%, but lost the stock. How do you rate the strength of that punish?

So, here's a possible fix. It's not perfect but I think it'll do the job for now: anything that results in a kill recieves the "max-cap" of percent. (What percent the cap is fairly arbitrary, but I'd say it'd make sense for it to be a little above kill percent.) So in your example, since downsmash resulted in an eventual kill - it counts as 150%. (And it's averaged with all the other Armada Fair->Downsmash situations) So if half the time Fair-> Downsmash results in a kill, and half time time they escape the downsmash safely and reset neutral - the value of Fair-Downsmash will be 75% (which to me seems about right as a first approximation).

The only way to make progress at first is to keep things as simple as possible. So our goal at first should be to come up with a couple of situations that are easily generalized (like a spacie at chaingrab percent on FD) that are useful in a lot of situations.

 

[Sycorax]

B blue cheez , This is a cool project, but I don't think I'm willing to help. I recently had a "straw that broke the camel's back" kind of moment. I do a lot of develop mode testing like was done for this thread, but it has made me realize how poorly designed, and frankly stupid, a game Melee is. I'm fed up with it. Because of that, I don't have the motivation to contribute to your project. Best of luck.

 

[blue cheez]

B blue cheez , This is a cool project, but I don't think I'm willing to help. I recently had a "straw that broke the camel's back" kind of moment. I do a lot of develop mode testing like was done for this thread, but it has made me realize how poorly designed, and frankly stupid, a game Melee is. I'm fed up with it. Because of that, I don't have the motivation to contribute to your project. Best of luck.

I think I overreacted a little bit after hearing enthusiasm from others. If you don't have the motivation to work on this stuff - that's fine. But I think I might have overwhelmed you with a lot of what has been bubbling in my head for a long time.

I think the most interesting stuff in the "Rock-Paper-Scissors" of Peach is as simple as taking a couple of good guesses, writing them in a matrix, and throwing them into a calculator. Admittedly this is not a prefect solution and a lot of heuristics must be used - but I think that's what makes melee interesting. If every decision was as simple as the same 4 by 4 nash equilibrium, then I think it wouldn't be as interesting of a game.

 

[reverie2]

This was an amazing read.

Just had a couples questions after peach lands a non-spaced aerial on your shield and follows up with double jab:

• She hits you with jab. Perhaps you were trying to do something out of shield, such as grab or roll. If she hits you with a jab, you can ASDI down to the ground and grab before she is out of lag from her jab, even though you got hit.

• She hits your shield. Peach's first jab is -10 on shield, and her second jab is -11. This is plenty of time to grab.

1. So the first option you listed would mean that your grab connects before the second jab hits right? And since you can't really react because it comes out so fast, ASDI down should be input preemptively while you attempt to buffer roll, and if you hold control stick down while spamming the grab button, you'll get hit by first jab and grab will connect before 2nd jab hits?

2. What if between the first and right before 2nd jab starts, you ASDI down spam grab. Then you will tank the 2nd grab, and restart your 7 frame grab animation after the 2nd jab hits you, and your grab will connect right?

3. Will ASDI down grab work also against characters like sheik/fox if they do a quick non-spaced aerial on shield and spam jab into flurry kick/punch asap? If so... would inputting ASDI down and spamming grab in either before the first jab OR between the first and second jabs both be able to grab the opponent?

Thanks

 

[Sycorax]

1. So the first option you listed would mean that your grab connects before the second jab hits right?

It depends which character is jabbing you. Lots of characters' second jabs come out too quickly to ASDI down grab in between the first and second. However, I'm pretty sure you can grab any character before their third jab/rapid jab. So what you can and should do is, if you expect a jab, hold down and mash grab. It doesn't matter how many times they jab because you can grab them between their second and third/rapid jab and you can grab them if they decide to jab only one or two times. There is too much lag on jabs for them to avoid the grab.

And since you can't really react because it comes out so fast, ASDI down should be input preemptively while you attempt to buffer roll, and if you hold control stick down while spamming the grab button, you'll get hit by first jab and grab will connect before 2nd jab hits?

It's hard to buffer roll and ASDI down at the same time. Technically they require two different inputs on the cstick. What you could do is press cstick in a diagonal notch. This would buffer the roll and ASDI diagonal down. ASDI diagonal down would not be as strong as ASDI straight down, but it might be enough to cancel the hitstun on most jabs if they are weak enough. It will be enough to cancel the stun on jabs like Peach's first jab or Falcon's first jab.

Just to clarify, the second jab should hit you but it doesn't matter for reasons stated above. At several points, I used words like "might" and "should" to cover the fact that there might be a character I'm forgetting about. This ASDI down technique works on all the relevant characters, but I could be forgetting some quirk of one of the mid or low tiers.

2. What if between the first and right before 2nd jab starts, you ASDI down spam grab. Then you will tank the 2nd grab, and restart your 7 frame grab animation after the 2nd jab hits you, and your grab will connect right?

Yes.

3. Will ASDI down grab work also against characters like sheik/fox if they do a quick non-spaced aerial on shield and spam jab into flurry kick/punch asap? If so... would inputting ASDI down and spamming grab in either before the first jab OR between the first and second jabs both be able to grab the opponent?

You can spam ASDI down grab any time before the third/rapid jabs. If they hit you with the first jab and stop, you'll grab them. If they hit you with the first jab and continue to the second then stop, you will grab them. If they continue to the rapid jab, you will grab them before they can start the rapid jab. If they start the rapid jab on you, then your SOL. You'll have to DI out of it and punish them another way.

Side note: you can ASDI down and dsmash between Sheik's first and second jab because it's so slow.

One more thing worth mentioning: Fox, Falco, and Sheik have some top tier privilege in that their jabs are rather strong doing 4 damage AND they don't have set knockback like most characters' jabs. This allows them to do silly things like jab usmash, or jab fair, or jab bair. It also means that starting ata 136%, you can't properly ASDI down their jabs. You will get a tumble collision with the ground (green flash) instead of a landing collision which you're quickly actionable out of.