Reacting to what your opponent does and behaving accordingly is one of the most important skills in Tekken.
For example, when you block a -9 move, you want to respond with a strong mixup; whereas when you block a -1 move, you still have an advantage to press, but you have to contend more with challenges and movement. When a throw connects you need to respond with the right break. When you block a string, you want to duck, sidestep, low parry, interrupt, etc. as appropriate.
Many moves are balanced around it being possible for the opponent to see them before they're active and guard accordingly. However, it being possible doesn't mean it's practical, nor does it mean the move is useless. A move can be reactable in some contexts but not others, and there's never a point at which a move is 100% reactable. Similarly, the threat of a reactable move can occupy enough of the opponent's focus to let other threats pass by unanswered.
Types of reaction times
There are three main types of reaction time experiments:
- Simple reaction time experiments study how fast the subject can react to a single stimulus, such as a light going from off to on.
- Recognition reaction time experiments study how fast the subject can react to a single stimulus, with the addition of some distracting stimulus that they should not react to, such as a screen showing a coloured dot, and only reacting if it's red.
- Choice reaction time experiments study how fast the subject can react to multiple stimuli with a different response for each one, such as pressing the right number on a numpad when it appears on the screen.
Average simple reaction times for college-aged persons are ~190ms for visual stimuli and ~150ms for audio stimuli. This number changes based on a number of factors including age, gender, fatigue, distractions, and psychoactive substances. For age, it peaks at around mid 20's, increases slowly until the 50's and 60's, then increases very quickly. Critically, this cannot be improved via training.
Recognition and choice reaction times are proportional to simple reaction times. Recognition times are longer than simple, and choice times are even longer.
Choice reactions are proportional to log(n), where n is the number of valid choices—not the number of stimuli. This can be substantially improved with practice, but not to the point where it's equivalent to simple reaction times, except when there's only 1 choice.
When is something reactable?
The main factors affecting how reactable a move is are:
- The player's simple reaction time
- How much input lag there is
- How many different choices there are to make
- Whether the cue being reacted to is a visual or audio cue
- When the cue becomes distinct
- How distinct the cue is and how much training the player has with it
Given the variety of factors, there's no clear point at which a move becomes reactable. In addition, even when all of these factors are accounted for, there is still a lot of variation in how quick the response is, i.e., a move can only be said to be reactable with some probability, never 100% of the time—everyone gets hit by Snake Edges sometimes—so a more precise question is, “When is a move reactable enough that it can't be relied on?”
As a rough estimation, if we assume that the cue is visual and that the player's simple reaction time to these is ~190ms, that there is roughly 60ms of input lag (on PC Tekken 7, ~74ms on PS4, under ideal circumstances), that the cue becomes distinct at ~100ms (Tekken animations are tweened, so the first few frames are not necessarily distinct), that there is only one choice being made (blocking low if the player sees a low move), and that the player is well-trained, then the cue should be reactable at ~350ms or ~21 frames.
However, keep in mind that this is very rough estimation. Some moves have more or less distinct animations; being well-trained against all ~100 moves that 1 of 50 characters can do is not practical; this is an average speed and will be slower than that at least half the time; this only applies to offline play with no ping between opponents. Most importantly, this estimate is only applicable if someone is only looking for lows. If they also want to react to frame advantage, throws, whiffs, stances, strings, or unblockables then their reactions won't be this fast.
If an animation is distinct early in its animation and a player is trained, talented, and stimulated, moves as fast as ~18 frames could be reacted to.
If a move has an audio cue it could be reactable a few frames earlier—not just because audio reaction times are faster, but because animation tweening isn't a concern. However, audio cues are not always played immediately, so reacting to an audio cue is not always better than reacting to a visual one.
In all cases, there is never a point at which something can be 100% reacted to. If a move can be reacted to only 20% of the time, there is some sense in which this is reactable, but it would be misleading to say that, for example, Asuka's d+1+2 is the same as a Snake Edge. Figuring out how useful a move is against skilled opponents and how it influences a robust strategy requires deeper analysis than merely identifying whether or not it's reactable.
As a broad rule, use these categories depending on the move's startup:
- i18–19 — Theoretically reactable (0–50%)
- i20–22 — Borderline reactable (10–70%)
- i23–25 — Reactable (20–90%)
- i26+ — Easily reactable (40–99%)
Numbers in brackets are estimated chance of being blocked by trained, talented opponent. Only applies in situations where the opponent is only looking to make a single choice on reaction (usually crouching to block a low). Can vary a lot depending on the cue.
Input lag refers to the time it takes for an input device's actuation to appear to affect the game. This is affected by a number of factors, including the controller, monitor, computer, game, operating system, and network quality (for online matches). Input lag is, of course, critical to reacting.
Holistic measurements of the total input lag, such as those done by Nigel Woodall, are good for getting a simple measurement under ideal circumstances. However, understanding all the puzzle pieces that make up input lag is useful to diagnose and address where the bottlenecks in a particular setup are. The most significant things under a player's control are the controller, monitor, and to some extent the network quality. The game and operating system can also be configured to reduce input lag, with the most notable effect coming from turning off V-sync.
During online matches, Tekken 7 tries to keep the input delay the same as offline matches with rollback netcode, enabling matches between players with a stable ~80ms round-trip time connection to have the same input lag as an offline match. This involves up to the first 5 frames of a move's startup being skipped by a rollback. Because most cues aren't distinct at that point anyway, this doesn't significantly affect on how reactable they are.
Dealing with an opponent reacting to your lows
If all your lows are getting blocked or parried and you feel like your opponent is doing it on reaction, the first thing to do is make sure it isn't a fuzzy guard or hard read first. Lows are usually slower than mids, so varying your timing is essential to creating a real mixup.
If the opponent is genuinely reacting, however, the most robust strategy is to increase the total number of choices the opponent must react with. In other words, do something that isn't a mid/low mixup. For example, use throws, stance transitions, unblockables, or strings that must be stepped or interrupted—threats that require answers other than guarding. Once these threats occupy the opponent's focus, they'll have more trouble reacting to the low.
It's advisable but not robust to mixup a move with another one with a similar cue, such as Lee's d/b+3+4 with 1+2, b+4, or u/f+3, because increasing the number of stimuli has little effect on a trained subject. Unless the cues are identical, it's still possible to tell them apart. Instead, the robust approach is to increase the number of choices the defender has to make.
A lot of reactable mixups come out of a stance transition with a strong frame advantage. For example, Alisa can mixup FC.d/f+1+2 with WS.1+2 after b+1 hits, but FC.d/f+1+2 is i21 and has a distinct audio cue. In this situation, the defender is focused on the mixup within a very narrow time frame (immediately after being hit) but afterwards will start worrying about other things, so the mixup can go from fake to real just by adding a little delay.
- Kosinski, R. J. (2008). "A literature review on reaction time". Clemson University.
- Hick, W. E. (1952). On the Rate of Gain of Information. Quarterly Journal of Experimental Psychology, 4(1), 11–26. doi:10.1080/17470215208416600,
- Nigel Woodall (2019). INPUT LAG DATABASE.
- Ricky "Infil" Pusch (2019). Explaining how fighting games use delay-based and rollback netcode.
- Whether Tekken 7 has rollback and how many frames it'll rollback still lacks a verifiable source. Harada said before the Version 4.00 patch, which improved the netcode, that it could rollback up to 3 frames and that they're working to increase this. (“No, in fact there is rollback (everyone talks rollback without knowing it). But the difference is whether the rollback frame is 1 or 6 (TK7 is 3). Increasing this number is more ideal, But causes problems with 3D animation. We’ve already working on this.”)