Thelo's quick guide to reaction-based defense

Thelo note: I am reposting this 2009 article of mine here, since some people asked to re-read it. I originally wrote this guide in a SF HD Remix Honda strategy thread, but I think its lessons are fundamental enough to apply to all fighting games.


“For every action, there is an equal and opposite reaction.”

  • Isaac Newton

*** There are two kinds of moves: Prediction and Reaction.

All moves you can do are done either in prediction to what you think your opponent will do, or in reaction to what you’ve actually seen him do. If, with Honda, you jump forward against Ryu, that’s because you think he’ll throw a fireball and you want to punish him with a jumping attack - but at the moment you jump forward, you haven’t actually seen Ryu throw anything yet. That’s a prediction jump. Or maybe against Zangief, you do a fierce Hundred Hand Slaps because you think Zangief will either block or lariat - that’s a prediction HHS, because at the moment you start mashing the fierce button, you don’t know if Zangief will really lariat or block when the HHS will come out.

On the other hand, if you see Ryu jump towards you, and you jab headbutt him out of the air at the last moment, you’ve done a reaction jab headbutt. That’s because at the moment you hit the headbutt input, you already knew Ryu was flying through the air towards you, so you simply punish him for it. Or maybe you saw Guile start to throw a Sonic boom, so you decided to reaction buttslam through it.

Even simple things like blocking and walking can fall under these two categories. Against Chun Li, a reasonable strategy is to walk forward and reaction block her slow fireballs. But against Sagat, whose fireballs are incredibly fast, you can’t do that - you need to walk forward and prediction block his Tiger shots, because you don’t have enough time to react to them.


“The history of man is a graveyard of great cultures that came to catastrophic ends because of their incapacity for planned, rational, voluntary reaction to challenge.”

  • Erich Fromm, philosopher

*** Reaction is better than prediction.

When you act on reaction, you actually know the state of the game, so your move becomes very safe. However, when you act on prediction, you need to guess correctly - and a wrong guess can be punished.

The easiest example to understand is Honda vs Guile, at the start of the round. Both players are charged down-back, ready to do either of their two charge moves, and Guile will likely throw a Sonic boom at some point.

Honda can predict that Guile will throw a Sonic boom in the next half-second, and do a prediction buttslam to beat it. If Guile did throw a Sonic boom, then all is good for Honda and he wins the exchange. But if Guile did nothing, then Guile can easily react to Honda’s buttslam with a Flash kick and win the exchange. Guile’s reaction beat Honda’s prediction.

However, if Honda focuses strongly on Guile, he can instead try to wait, then react to Guile’s Sonic boom. At some point, Guile will likely try to predict that Honda will walk forward or do a HHS in the next half-second, and throw a Sonic boom to beat it. If Honda did try to act, then all is good for Guile and he wins the exchange. But if Honda did nothing, then Honda can easily react to Guile’s Sonic boom with a buttslam and win the exchange. Honda’s reaction beat Guile’s prediction.

When you act on reaction, you know what’s going to happen. When you act in prediction, you don’t. That’s why, when you can do it, it’s better to act on reaction than in prediction.

Of course, you can’t always act on reaction, because you need to have enough time to react to do it. That’s why you need to put yourself in a position where you have plenty of time to react, you need to try to minimize the reaction time you need to react properly, and you need to try to minimize your opponent’s own ability to react. We’ll go over these a bit later.


“The main factor behind success is - self control.”

  • The Rigveda

*** Reaction is better than prediction, no really!

Even if you consciously agree that you should react more than predict, even in a very particular situation like Honda vs Guile where you actively want to react rather than predict, even when you explicitly say to yourself “I must not do a random buttslam, I must not do a random buttslam, I must not do a random buttslam…”, sometimes your instincts will take over and you will do a kneejerk random prediction buttslam when you didn’t actually want to do one.

The truth is that it’s really hard to just sit there, while your nerves are on fire and adrenaline is pumping through your veins, and not do anything but wait for your opponent to do a specific move. Sometimes, no matter how patient and resolved you think you are, you will lose your cool and randomly do an irrational attack. It will happen, and early on, it will happen often.

The only thing you can do is consciously realize that it was a mistake, and resolve yourself to avoid it next time. Visualize that same situation happening again, and visualize the right thing to do in that situation. Instincts are often very useful, but sometimes you need to fight them, and doing moves on reaction is very un-instinctive. You need the unwavering willpower necessary to overcome your instincts in order to properly act on reaction.


“I party but I know my limits.”

  • Dante Hall, football player

*** You only have a finite amount of focus.

Reacting in time to an opponent’s move needs a huge amount of focus. When you ready yourself to quickly react to an opponent’s specific move, you get tunnel vision and it gets harder to react to something else. The simpler the situation is, the better your chances are of reacting correctly. Here are some reaction scenarios in increasing order of difficulty:

  • React when the opponent jumps, wait otherwise, like life advantaged Honda’s jab headbutt vs Zangief. It’s easy to recognize a jump.

  • React when the opponent loses his down charge, wait otherwise, like Honda’s buttslam vs Guile. It’s harder to recognize standing vs crouching, especially on Guile’s crouching fierce move that looks like it’s standing.

  • React with move A if the opponent does move X, react with move B if the opponent does move Y, wait otherwise, like Honda vs T. Hawk’s jump / dive (must jab headbutt) or walk forward (must far sweep). Now you have three possibilities, which is much harder to manage than just two.

  • React with move A if the opponent does move X, react with move B if the opponent does move Y, and react with move C if the opponent does move Z, like Honda vs Ken at round start. Honda wants to buttslam through Ken’s fireball, jab headbutt if Ken jumps forward, and move forward with strong headbutt or normal walking if Ken jumps backwards. This is among the most difficult reaction scenarios.


“Work smarter, not harder.”
-Scrooge McDuck, DuckTales

*** Manage your focus, part 1: Simplifying reactions

What can we do when we’re faced with a tough scenario like Honda vs Ken at round start? Ideally, we want to take some shortcuts to lower the difficulty, and here we actually can, because the reaction times of the three possibilities are different. Here’s how.

[I]Vs Ken, round opening decision tree:

  • IF Ken fireballs THEN buttslam
  • IF Ken jumped forward THEN jab headbutt
  • IF Ken jumped backwards THEN strong headbutt[/I]

First, we can notice that reacting to the Ken fireball is the most difficult of the three possibilities, since it’s the fastest one. If Ken jumps forward, we have about half a second to prepare a jab headbutt, but if he fireballs, we need to buttslam right away. Furthermore, Ken will rarely wait half a second, then fireball - he’ll usually fireball right at the start or not at all for a few seconds.

So what we can do is, for the first half-second, pretend that the only thing we need to react to is the fireball, and devote our full focus to recognizing the fireball and reaction buttslamming through it. Then, after the initial half-second, switch to reacting to either of the two remaining possibilities. This lets us allocate our focus where we need it most.

But that’s not all! Ken jumping forward at match start is actually really rare, and Ken jumping backwards is very common, so we can actually focus almost exclusively on what to do if he jumps backwards. This lets us save another bit of focus!

Finally, if we need even more focus, we have a last trick up our sleeve: overlapping our reactions. In our Ken scenario, we have three possible reactions: buttslam to beat fireball, strong headbutt to beat backwards jump, or jab headbutt to beat forward jump. Jab headbutt is the best counter to forward jump, but it consumes focus for a pretty rare case, so what we can do is to use either of the other two reactions, i.e. strong headbutt or buttslam, to beat Ken’s jump forward. They won’t be quite as effective, but they happen to do an okay job, and we won’t have to worry about hitting the jab button at all. Which to choose?

If we choose to buttslam against Ken’s forward jump, we have the following reduced decision tree:

_0.0s to 0.5s: _
- IF Ken fireballs THEN buttslam

0.5s to 1.0s:
- IF Ken jumped forward THEN buttslam
- IF Ken jumped backwards THEN strong headbutt

If we choose to use strong headbutt instead, we have this even simpler reduced decision tree:

0.0s to 0.5s:
- IF Ken fireballs THEN buttslam

0.5s to 1.0s:
- IF Ken jumped (forward or backwards) THEN strong headbutt

Bingo! This decision tree is much simpler to handle than the initial three-way one, and we can fully focus on each part, rather than split our focus between three options. By managing our focus this way, we can use a much more reliable reaction defense when faced with very tight timing windows.

Remember: when reacting, you must choose to either react to many options slowly (complex reaction tree), or react to few options quickly (simple reaction tree).

*** Manage your focus, part 2: Focus spikes

When trying to react to your opponent, at some moments, your opponent is much, much more likely to act than at others. For instance, at the start of a round, right after a knockdown recovery or a jump, or right at the end of a move’s recovery, many players will immediately act – they are much less likely to stay standing for 0.3 seconds, and then act.

Accordingly, you can concentrate all of your focus in these very short moments where you’re likely to need to react. I call these “focus spikes”, because on a graph of your intensity of focus over time, these moments would show a sharp spike – as mere humans, we really can’t stay at 100% focus for very long and our intensity of focus soon drops back down.

Because using reaction is mentally very exhausting, when you use reaction, you should try to purposefully focus spike in these critical moments where you predict your opponent will act, in order to reduce your reaction time and conserve your mental energies.

Conversely, if you’re trying to defeat your opponent’s reaction, try to avoid acting during your opponent’s focus spikes. For instance, doing “jump backwards, wait 0.3 seconds, fireball” can be all it takes to avoid the focus spike and catch your opponent off-guard. Introduce small delays in your attack sequences and you can become much harder to react to.


“The most successful people are those who are good at Plan B.”

  • James Yorke, mathematician

*** Compensating for reaction time

Some moves can be reacted to, but you need lightning quick reactions to do so. If your current plan is reaction-based, they are your worst enemies: for instance, Zangief’s Running bear grab, Cammy’s Hooligan throw, Blanka’s forward hop or whiffed horizontal roll into bite, Bison’s Headstomp. If you’re in top shape, you can try to fight them head-on by just reacting really fast to them, but it’s dangerous. What to do?

One possibility is to realize that each of these moves happens to have a counter that’s mostly safe and doesn’t consume much focus. We can use these counters that look like attacks (but actually are merely a defense), while continuing to actually focus on reacting to our opponent’s other possible moves. Most of the time, they are safe whiffed normals.

As an example, for Honda:

Against Zangief’s Running bear grab, Cammy’s Hooligan throw, Blanka’s forward hop or whiffed horizontal roll into bite:
- Repeated low jab (while holding down-back)

Against Bison’s headstomp:
- Either repeated backjump jab or just walking backwards (high block)

Again, we must be careful, though, of not focusing too much on executing these counters - the whole idea is to pay little attention to their execution, while paying a lot of attention to the opponent and staying ready to do the real reaction. They’re also slightly more vulnerable to other actions, since you do commit some time to executing them and they can be prediction-beaten by some risky moves. But against an opponent who likes to use the hard-to-react-to moves, it’s an effective tool.

**

Another way of improving reaction time, at the cost of some flexibility, is “buffering” the reaction motion. For instance, let’s say you have a very short time to react to a jump by doing f, d, df+P. This is a long input sequence, so you might not have enough time to complete it in time to react.

What you can try in this scenario is, when you think that your opponent will jump (before you actually see any jump), input f, d, df. If, during the time you were inputting this motion, you saw the jump, then and only then you hit P.

This way, you effectively shorten your required reaction time, because you don’t have to do the entire motion after your reaction – only the last punch input.

Charge characters can also try to use this, especially for their supers, but they have a higher price to pay if they predict wrong: they lose their charge. Nevertheless, it remains a useful and threatening tool.


“Everything is in a state of flux, including the status quo.”

  • Robert Byrne, champion billiards player

*** When to use reaction

By this point, you might be thinking that I’m telling you to do everything as reaction. Of course, it doesn’t work that way. Every move in Street Fighter can be done as a prediction, but only a very limited subset of scenarios, involving a very limited subset of moves, can actually be resolved using reaction. What’s more, even the scenarios where it is possible to use reaction are never really safe – many things can still go wrong. Nevertheless, reaction is a useful tool, when used at the right times.

You will often have to decide when to try reaction and when to try prediction. Here are some things that favor reaction:

  • You have life advantage
  • You are far away from the opponent
  • The opponent has very few attacking options
  • You can react to most of your opponent’s actions with a low number of reaction moves
  • You have a lot of available time to react
  • Your reaction move(s) are fast to input (f+P is faster than f, d, df+P, which is faster than hcb, hcb+P)
  • It’s easy to guess which attacking option your opponent will next use, and you can react to it
  • You are familiar with the scenario being played out
  • You have a solid hit rate on the execution of your reaction move
  • You are well-rested, physically comfortable, and using a good and familiar controller
  • You are playing offline, or with minimal lag
  • The opponent is hotheaded, tends to attack at every opportunity, and/or rarely uses fakes and feints.
  • You have not used a particular reaction setup much in this match yet (i.e. your opponent doesn’t really expect you to react to his attack)

The more of these are true, the more you should try to use reaction. For instance, if you are playing Honda against a jump-happy T. Hawk, with full life against his quarter life and 10 seconds remaining on the clock, sitting in the comfort of your couch in a lazy Sunday afternoon, you would do well to try reaction jab headbutts.

However, if you are playing your new character Fei Long against an unfamiliar Honda opponent on a laggy LCD screen, you are down in life, hungry, and playing in a noisy environment on a borrowed stick at 3 am, then it should be wiser to try some prediction Rekka-ken punches.

Even when in a situation of life disadvantage, if the timer is still high, don’t rush in needlessly - let at least some time pass to see if your opponent will hit the lose button and mindlessly give you an opportunity with a careless prediction attack.


“You have to learn the rules of the game. And then you have to play better than anyone else.”

  • Albert Einstein, physicist

*** How to beat reaction

Reaction is a powerful tool, but you can’t always rely on it. In particular, there will be times where your opponent tries to use reaction and you must crack his shell. To do that, remember the two things that reaction needs: time and focus. By robbing your reactive opponent of either, you can prevent him from reacting properly and force him to switch to riskier tactics.

The easiest way to defeat reaction is simply with attacks that are so fast that you cannot properly react to them.

  • In Honda vs Guile, Guile’s reaction plan is to hold down-back, ready to Flash kick. To defeat this plan, Honda can get to midrange and do strong or fierce Hundred Hand Slaps, which come out too fast for Guile to react to and chip the blocking Guile of some life.

  • In Honda vs Honda, it’s possible to be at fullscreen, waiting for the enemy Honda’s fierce headbutt, and react to it with your own jab headbutt. However, it’s extremely difficult to react to a fierce headbutt when the enemy Honda starts at midrange instead, since there is much less time to react. In Honda mirrors, the best place for reaction Honda to be is therefore at fullscreen, and the best place for prediction Honda to be is at midrange.

  • In Fei Long vs Honda, Fei Long can be at midrange and do prediction Rekka punches, which are too fast for Honda to react to and can be safe on block - Honda needs to prediction jump or buttslam to counter them.

**

An alternate way to defeat reaction is by exploiting its narrow, “tunnel vision” focus. As we saw earlier, reaction works best against a very limited set of obvious moves. If we do some unexpected move on an opponent who is fully focusing on his reaction decision tree, even if it looks kind of dumb, it just might work. The best example of that is to simply walk up to the opponent and throw him. Whether you fail or succeed, you’ve just shaken up his reaction plan and he must now add the threat of walk-up throw (or whatever your unexpected move was) to his reaction decision tree, which will make it harder for him to react to whatever you will do next.

Be VERY careful not to overuse this! The strength of this tactic is precisely its surprise factor, so don’t dilute its power through repetition. Even doing an unexpected move once is enough to shake your opponent. The threat is often stronger than the execution, so keep your real options varied and your “surprise” options rare.

**

A third way to defeat reaction is through fakes and feints. These look like a reactable attack, and can bait a reaction while remaining safe from that reaction. For instance, in Ryu vs Honda, Honda may want to react to Ryu’s fireball by jumping or buttslamming through it. Honda will focus intensely on Ryu, ready to react as soon as Ryu throws a fireball. Ryu can then do a fake fireball instead, which will trigger Honda’s reaction jump or buttslam since it looks like a real fireball. The fake fireball recovers quickly, allowing Ryu to punish Honda easily.

This is a very obvious example, but many others exist. Guile can fake losing his down charge with a crouch fierce punch, Deejay can fake a fireball with a standing fierce punch. Bison can fake a meaty or tick throw attempt by moving very close to his knocked down opponent to bait a reversal, then stepping back at the last moment. T. Hawk can fake a jump-in on Honda by whiffing a jump jab just in front of Honda, then DP his headbutt.

To a certain extent, all characters can also make it harder to react to their moves by generally whiffing a lot of safe attacks. By whiffing a lot of normals or safe short hurricane kicks, for instance, Ryu can make it much harder for his opponent to react to his fireballs, since their signal is lost in the flurry of noisy motion that Ryu makes. It’s also hard to keep a reaction focus for very long, especially when our nerves are constantly tested like this (Is this a real fireball? How about this?.And now this one? Etc.)


“As far as the laws of mathematics refer to reality, they are not certain; and as far as they are certain, they do not refer to reality”

  • Albert Einstein, physicist

*** Twenty frames

It’s now time to venture into pretty controversial territory, and make a bold claim: your effective reaction time in Street Fighter hovers somewhere around twenty frames, or about 0.3 seconds.

I already hear you guys howling in the back, throwing tomatoes, and giving a ton of counter-examples, including a bunch of things I just said that affected reaction time a lot, and heavy variation between people and scenarios, so I’ll preemptively give in and say that yes, it varies a lot and it’s definitely not stable at twenty frames. Here’s an online test that tries to measure “go/no-go” reaction times, you can see the large variations by yourself and try to measure your own.

20 frames is still an okay ballpark figure. It means that you probably shouldn’t really try to react to things faster than 20 frames (predict or avoid them instead), that you have a good shot at reacting to things slower than 30 frames (those situations are reliable for reaction, seek them), and that things in between are more or less risky to try to react to, depending on the situation. See the “When to use reaction” section for a bunch of things that can increase or decrease your actual reaction time.

As a quick reference, you have about 20 frames to react to fullscreen Honda’s fierce headbutt.


“It’s natural for a sumo wrestler to be the world’s strongest.”

  • Edmond Honda, Ōzeki

*** What is the best way to use reaction?

Play as Honda.

9 Likes

The post so nice I’m liking it twice!

1 Like

I think it’s important to understand the flipside of this. A hyper-emphasis on reaction leads to a plateau of skill imo; a local maximum.

Think of it this way: it’s obvious that reaction is better than prediction in situations where reaction gives a better result. That’s by definition of “gives better result.” But it’s not better than prediction in cases where it leads to worse results. The entire matchup of Fei Long vs Guile in SF HD Remix is a good example of this. As Fei Long, if you stick to reactions, then the overall EV of the matchup is against you, and you’d think it’s an unfavorable matchup. But if you gamble constantly, the EV is in your favor.

I played this matchup a lot with a Guile player and destroyed him really bad. Then we switched characters so I could try to show him how my tricks worked. He sucked at the Fei Long side too though. It wasn’t because he was a bad player though, he wasn’t. His years of Guile play had made him rely entirely on reaction-based thinking. He didn’t have it in him to constantly be guessing with Fei Long, even though it’s correct to do so because it’s +EV. So instead, he tried to maximize the number of times he could do moves on reaction with Fei Long, which just limits his options to the point that Guile controls the match. You can’t really “play it safe” and stick to reaction moves when playing a rushdown character against one of the most reaction-based defensive characters in the game. It doesn’t really make sense there.

I’d go so far as to say that being truly good involves exploiting AS MUCH of the +EV as possible from prediction play, rather than leaving all that advantage on the table, unused. The progression goes something like this:

  1. worst player. Flops around, does random things. Maybe gambles with predicting things, but with bad risk / reward, and at times there was no reason to risk anything at all.
  2. better player. Find all the cases where you can react to things and do it. This is straight up better than…not reacting to things when you could have. Some gambles are no longer necessary now, which is helpful. This stage of development offers a huge boost to your win rate because it eliminates a bunch of your stupid mistakes and allows you to capitalize on mistakes from the opponent.
  3. even better player. Increase your win ratio further by adding +EV guesses everywhere you can. By definition, they are better than not doing them UNLESS you’re in a case where you don’t even need +EV because you’re already winning and time is almost out or whatever.

If anyone still needs intuition on this, imagine your character can run up and do a high / low mixup that leads to 100% damage on hit, and if you get hit out of your attempt, you take 5% damage. You’d want to shift basically all gameplay to pure guessing if your character had that. It would be overwhelmingly powerful to constantly force that guess. You’d want to react to things you CAN react to still, yes, but every moment you spent fishing for that stuff is probably worse time spent than abusing your crazily powerful guessing game.

So anyway yeah learning about reaction-based stuff is a good first step. I’d advise learning the second step too, which aside from being +EV by definition, is also where the truly interesting gameplay lies.

8 Likes

[quote=“Sirlin, post:3, topic:96, full:true”]

I played this matchup a lot with a Guile player and destroyed him really bad. Then we switched characters so I could try to show him how my tricks worked. He sucked at the Fei Long side too though. It wasn’t because he was a bad player though, he wasn’t. His years of Guile play had made him rely entirely on reaction-based thinking. He didn’t have it in him to constantly be guessing with Fei Long, even though it’s correct to do so because it’s +EV. So instead, he tried to maximize the number of times he could do moves on reaction with Fei Long, which just limits his options to the point that Guile controls the match. You can’t really “play it safe” and stick to reaction moves when playing a rushdown character against one of the most reaction-based defensive characters in the game. It doesn’t really make sense there. [/quote]

This analysis reminds me a lot of when I was learning to fight against Rook with Valerie in Fantasy Strike; Rook’s C Throw is a powerful tool because it’s both predictive AND reactive in its usage, and as Valerie I just wanted to try to play in a way that negates the C Throw as much as possible.

This lead to a keepaway playstyle where I tried to use Valerie’s superior poke range to keep Rook away from me, but every correct guess on my end was 1/8th of Rook’s health, and every time he was right I lost 2/5ths or more. It’s strictly because I was trying to play as safe as possible and not focus on the guessing games (the blocked BBC mixup) that I made it fundamentally more difficult to win the matchup, even though Sirlin argues it’s a Valerie favored fight. Irony here of Valerie kind of being the Fei Long analogue in this game, too :smiley:

Sirlin said, “just go in, it’s +EV” and once I started engaging Rook directly and trying to bait his moves, I started getting the 3 damage combos I needed to beat him.

Nonetheless, when I play Setsuki vs. Rook, it feels a lot better to me because I have very safe pressure with kunais and flying fox so I get to be much more aggressive and make fewer guesses. Whether the matchup is better or worse for Sets I’m not sure yet, but it just goes to show that the better Val players are going to be the ones who are better able to make you guess unfavorably and get critical damage out of those exchanges.

EDIT: Also, keep in mind that most of my SF career was playing Guile, so it might stand to reason that I fell into the trap of trying to be a purely reactive player, and unwilling to branch out from the “set play” mindset.

6 Likes