Celia Alexis Wagar's CritPoints

Care to do a breakdown on reads in fighting games?

There’s a bunch of different types of reads and types of information you can base reads on. First however, I’m gonna cover reactions, because that’s related.

Human reaction time is about 15 frames for something you’re expecting that you have a specific response planned for. So if you’re blocking low and know your opponent is going to overhead at some point and there’s no other variables, you can see him do that when he finally does it and block high. You can try this with the Millia Blocker game here:
http://www.teyah.net/milliablocker.html

Some people have better reaction speed than that, down to 10 or 13 frames. They’re extremely rare. Every thing you add onto the situation, every factor that makes it more complicated, adds just a little more reaction time. Like, do you need to pick between 2 or more possible choices? Are you not specifically anticipating the action to be reacted to? Do you not have a specific action prepped? If you answer yes to these, then your reaction time will likely take longer. There’s a big difference between waiting for something and responding to it, and something coming up unexpectedly and responding to it. Also, reaction time in response to audio is lower. Reaction time in response to tactile sensations is even lower.

So this means that it will take you about 15 frames minimum in order to press a button once you see something. That’s really slow. If it takes 15 frames for us to realize something and do it, how do people who play rhythm games ever hit 1 or 2 frame windows? How do you beat people who use attacks that are faster than 15 frames? The answer is that even though you physically can’t see that quickly, you can still time your inputs to come out whenever you want them to, just you’re flying blind. In cases like rhythm games, you can see what’s on the screen and extrapolate forward to match the timing of the thing coming in. Our brain also plays a trick on us, it edits our perception to make it seem like we’re perceiving everything instantaneously and acting instantaneously too, so you can have the illusion of seeing something happen, but feel like you’re too slow to stop it. Once you’re conscious of this and have a good feeling of how long that timing period is, you can start thinking of pressing buttons in that blind-spot period of time. In cases like beating a move that’s faster than 15 frames, that’s a matter of pressing the button before they do, which means predicting that they’ll press the button and when.

This is called a read. A read is essentially predicting that your opponent is going to do something. The specific type of read above is called a “Tempo Read”. Tempo Reads are about reading the tempo of the action and acting right before the “beat”. If you know when your opponent will attack, then you can attack first, such as in this example:

Another type of read is what I’ll call an “Option Read”, where you know your opponent is going to do a certain type of option, so you perform an action that beats that action. These reads work a bit like Rock Paper Scissors, and fighting games have a lot of examples of these. The most common is hit standing punch or the more common, poke < whiff punish < deep poke.

So you have a prediction that your opponent is going to do something at a time, and what they’re going to do. These predictions can be more specific or more general. If you don’t have a prediction, if you have no idea what your opponent is going to do, then you’re likely going to lose until you get an idea of what their gameplan is. Conversely, players who don’t know what they’re doing can act unpredictably and be tricky or impossible to read, however these players can be smothered by using tactics that require a specific response, such as fireballs, jumpins or sweeps.

So then there’s the issue of option coverage. Jumping attacks in most fighters will beat most ground attacks that don’t point diagonally upwards. Jumping is popular among new players for this reason (and because it’s faster than walking). So if you think your opponent will attack, you can jump it and hit them frequently. However jumping has the issue that it’s slow, and reactable. So if your opponent is not doing anything (and they’re not a noob), they can see you jumping and easily anti-air you, or at least block high. If you think your opponent is going to jump in, you can watch the skies for when they do, and then do your anti-air option. So in this way, predictions about your opponent’s action can augment your reaction time. A common trick is to buffer a dragon punch if you think your opponent will jump, and only press punch if they actually jump. You’re doing a tempo read, but using your reaction time to avoid the downside of an incorrect prediction. If you know they’ll jump, but don’t know when, then you can just crouch and do heavy punch when they finally do it.

If you think your opponent will do attacks that don’t have much range or which have significant startup, you can stand at the edge of their range and poke them. If you think your opponent will dash up to you or jump in, you can wait with a flash kick charged and use it once they try something. If you think your opponent will walk forward, you can poke with a cancellable move just out of their range and buffer that into a special. Daigo above knew that Dhalsim was going to poke with his arms fullscreen, so he tried dragon punches, which are invincible and will beat any attack, and buffered that into super. If Dhalsim were smarter, then after seeing the first shoryuken whiff, he could have used other moves or mixed up his timing to bait out the second one and punish it.

However instead of doing that, the Dhalsim player kept attacking more or less when his last attack ended, which made him predictable, making it easy for Daigo to time his Shoryuken correctly. A Tempo Read is about reading the tempo of the match. In fighting games, there’s a “beat” to when players typically do things. Players create a new plan of actions every 20-26 frames or so, about every half second. If you’re conscious of this beat, then you can attack right on the beat, before the beat, or after the beat. When I fight another Marth player in Smash, it’s really common to run forward off the platform and Forward Air. Since I’m conscious of the beat and have fast reaction times, usually I do this first. Sometimes however the other player is faster than I am or just as fast as I am.

This brings me to the next topic, if you see what the other player does in a situation, you can remember that, and you can punish it next time it happens. If I notice that the other player is running off the platform and slashing first, if I think they have better reaction time than I do, then I’ll actually delay a little and run off the platform later. I’ll let them miss and hit them afterwards. If I know that someone always dragon punches on wakeup, I can remember that, block, then punish their dragon punch. Everyone has general patterns, and reading people is about having a good short term memory for what patterns you see people do over a match. What options do they prefer? What options can you pick which beat those options? What did they choose last time this situation came up? Based on your knowledge of the situation, what might they also choose? Do you have an option that beats both of those options? Do they switch their options quickly, or do they stick for a long time? Are they due to switch, and what option do they have that will beat the option you’ve been beating them with, what beats their counter-option? Beyond that, there’s generalities about the ways people think, people tend to respond to some situations similarly. If you remember common responses to that situation, then you can generally predict what’ll happen when someone gets put in it.

So apart from paying attention to what your opponent commonly does, how else can you force them to get hit? The thing is, while you’re paying attention to their behavior, they’re also paying attention to yours. You doing things can lead them to try different things, and so you can choose options on purpose to get them to try to counter those options, setting them up for a bigger fall. In 3rd Strike, Alex has a stomping move that can hit his opponent in front or back just as they get up. Beginners will likely try to attack, or jump out, but they’ll learn this is impossible, and they need to block. Since the stomp can hit in back, they also need to learn to block the opposite direction. Once they start blocking the opposite direction, the stomp becomes a mixup. You can hit them in front or back and they need to guess which. However because they’re trying to block in front or back, you can also use Alex’s Spiral DDT, a command grab that only hits standing characters. Since they need to block the stomp high, spiral DDT will hit them regardless of which direction they decide to block. So by setting them up with the stomp, you condition them to block high, making them vulnerable to a new threat that beats both their options. Of course they could just duck until they see a stomp and stand up to block, but then Alex can hit them with either a standing HP or the universal overhead, which can be linked into super, both of which hit high and will beat crouch. Of course, a new player is never going to learn this, so just keep crossup stomping them as they try to press buttons on wakeup instead of trying to go for mixups they don’t understand yet. Remember, it’s not a mixup unless the opponent understands how to counter you.

A similar example is that in Tekken, jabs are +1 and therefore are the basis of your pressure. Since jabs move you forward, you can jab forever, but people can duck underneath jabs to whiff punish them, but most launchers hit mid, which beats a crouch block. So by jabbing, you can condition them to crouch, then hit them with a launcher for a big damage juggle combo. However this takes a tempo read, because you need to predict when they’re actually going to crouch, or you could get punished in the process, trying to launch them when they’re just going to block standing.

One way to solve situations like this is to start counting. How many times does your opponent typically do something before they do a certain option? Do they jab twice before the tick throw? once? No times? There’s a lot of different opportunities to do this during different games. This video explains the idea in more detail:

Another thing about fighting game design is, the moves tend to be designed where you don’t have totally even rock-paper-scissors relationships between things. You have options that beat multiple options, you have options that beat very specific options, you have options that are reactable, unreactable, punishable, plus or minus. And they might deal wildly differing amounts of damage, leading people to favor more damaging options over less damaging ones. You have different situations that change the value of various options, transitioning into a minigame temporarily for that situation. Understanding these relationships of moves that counter each other can also help inform you about your opponent’s next action. Since they’re not all of perfectly equal value like in RPS, you can get a more clear idea of what your opponent might favor in that scenario, improving your odds beyond 33%. Of course, them knowing that you will try this may lead to them picking that last option that will beat your coverage, or simply waiting it out and pushing the situation back to neutral.

A bit of esoterica is that even if the startup of your move is so long that it’s reactable, if the opponent’s move has a recovery that’s longer than your move’s startup, you can still hit them if you make sure their attack misses you. If both of your attacks have a reactable startup, then you can time your attack slightly before theirs, so that the actual time period you both commit to attacking during is double blind and they can’t react. This requires a tempo read for when they’ll press the button of course, and it means if you read wrong, then they can beat your option on reaction.

Worth mentioning is that in top level play, all reads are to some degree tempo reads. Nobody is going to throw out something reactable unless they’re reading that their opponent will try something wrong. People pick options that beat their opponent’s options, but they’re also attacking with the right timing to beat their opponent’s options most of the time too. Attacking first or slightly after to whiff punish is usually much more effective than trying to cleanly counter an opponent’s option. Smart players won’t attack immediately after a move recovers or stun ends. They’ll stagger the timing of their attacks to hit at different times so their opponent can’t predict what they’ll do. Good players are conscious of their opponent’s ability to react and how many actions they can get in during the time their opponent can’t react. Of course, a good player also can account for those types of plans in the time they cannot react too, and thus counter their opponent’s unreactable plans with a good read.

And of course what ties this all together is that humans aren’t actually random, as I’ve covered in a previous article. https://critpoints.wordpress.com/2016/09/14/people-arent-random/ Ask a cryptographer to use a human as their source of entropy and they’ll laugh in your face. Try an exercise, ask a person to write down heads or tails like 50 times randomly, then actually flip a coin and compare the results. It’s usually very easy to tell which one is the human and which is the coin. Humans tend to expect certain behaviors out of randomness that don’t actually exist, like expecting it to stay very close to the average in the short term. Humans tend to abhor doing the same thing more than 3 times in a row when trying to be unpredictable, and this ironically makes them more predictable. A better word to describe humans would be “chaotic”. We’re progressively thinking about what just happened, what we just did, and what we’re going to do. We’re a big feedback loop that produces results based on a large and changing number of semi-deterministic factors. Of course, something that is similar enough to randomness in the context of a game can be effectively random as far as anyone is concerned, but I would argue that predicting the behavior of humans is a special exception.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1044840/ https://www.sciencedaily.com/releases/2015/02/150226132046.htm http://www.fil.ion.ucl.ac.uk/~karl/Predictive%20coding%20an%20account%20of%20the%20mirror%20neuron%20system.pdf

It’s been shown that humans have a limited ability to predict the actions of other humans under observation, believed to be facilitated with Mirror Neurons, Neurons that fire both when you do something, and when you see someone else do something (or even if you think they’re going to do it, but don’t realize you think it yet).

I think it’s a logical assumption that we evolved mechanisms to predict the behaviors of other humans because doing so is beneficial to survival among other humans. Like other mental processes, our conscious mind isn’t allowed direct access to these, but it’s been shown that people can recognize and adapt to patterns without consciously recognizing them. So if you get a hunch about what someone’s going to do, sometimes it’s for a reason. Sirlin has a number of citations for that in this article: http://www.sirlin.net/articles/balancing-multiplayer-games-part-4-intuition