A frame number answers one question: whose turn is it now? Everything below comes out of that.

This is the doing half of frame data. Reading the frame data pages covers what each column on this site means and where the numbers came from, and the glossary defines the words. Mastering Tekken: fundamentals covers the skills around this one, including movement, throws and stage position. Inputs here use western notation; the notation guide covers the other two modes.

It runs in three tiers. The first is what the numbers count, the second is what you do with them while a round is running, and the third is strings, where frame data starts running out and has to say so.

Every number quoted is a real move from this site's own frame data, named to its owner. Where something is a rule of thumb instead of a rule, it says so, and it says what the rule of thumb is standing in for.

Beginner: what the numbers count

One frame is a sixtieth of a second

Tekken 8 runs at 60fps. Every figure on this site counts those frames. +5 is five sixtieths of a second and i10 is a sixth of one. They are tiny, and they settle almost every exchange in the game.

Startup: the i number

i10 means the move's first damaging frame lands ten frames after the input. Lower is faster.

Every character's standing 1 is i10, except Jack-8's, which is i12. The punishment tiers below are built on that jab, because it is the speed the whole roster has from a standing block. A handful of rows do publish something faster: back-turned jabs at i8, a few stance moves, and a few neutral moves whose startup is published as a range that opens below i10. None of them changes where the tier boundaries sit.

On block, on hit: two numbers for the same move

Both count the same thing, the frame gap between the two players when everyone is free to act again. They differ because getting hit locks you up for longer than blocking does.

Jin's 1 is +1 on block and +8 on hit. Same jab, same startup, and the situation afterwards is seven frames apart. Landing a poke and having it blocked are two different results, not two degrees of one.

Positive means the attacker moves first. Negative means the defender does. A move being minus is not a criticism of the move; it is a statement that the turn has changed hands.

Two conventions to know before you read a row, both covered in reading the frame data pages: a range like -31~-29 is quoted here at the least punishable end, and a g after the number (+17g) means they can guard through your recovery, which is not pressure at all.

The arithmetic, which is just addition

At −N, they act N frames before you. So your iM move arrives on frame N + M of their clock, and theirs arrives on its own startup.

Block Kazuya's 4 at −9 and jab. Your jab lands on frame 19 of his clock. Anything he throws that lands after frame 19 loses to it. His own jab lands on frame 10, so that one wins, and so does everything else fast enough to get in first.

At +N the same sum runs your way. Block-punish arithmetic, interrupt arithmetic and frame traps are all this one calculation pointed in different directions.

i10 is the punish floor, and there are four bands

Because the fastest thing you have out of a standing block is that i10 jab, a blocked move is only punishable if it leaves them stuck for at least ten frames. That gives four categories, and they are the four the quiz asks you for:

Launch punishable, −15 or worse. A launcher usually fits, so a mistake here costs a full combo. Kazuya's Spinning Demon f,n,d,DF+4 is an i16 low at −23. Whether your launcher fits is a fact about your character, not about the band.
Non-launch punishable, −10 to −14. Something connects, but not the expensive thing. Jin's uf+4 is −13.
Safe, 0 to −9. Nothing is guaranteed. Kazuya's 4 at −9 is the last frame of this band.
Safe and plus, +1 or more. They keep the initiative. Bryan's b+1 is a mid at +4.

Zero belongs in the safe band, not the plus one. At 0 nobody has been given anything, so calling it plus would promise a turn the frames do not.

Every character page carries that ladder as a table, headed "What Dragunov punishes you with", or whoever you opened. It is built at i10, i12, i13 and i15. Read it on an opponent's page and it is a list of what your mistakes cost. Open your own character and the same table is your punishment ladder.

What "safe" does not mean

Safe means "no guaranteed punish on block". It does not mean the move was a good idea.

A high that is −5 on block is still a high. Duck it instead of blocking it and it whiffs entirely, which costs them a great deal more than five frames. A safe low can be low-parried. A safe mid can be stepped if it is linear. And a punish that is fast enough on frames can still whiff, because some moves shove you out of range as you block them. The quiz refuses to ask you for a punisher at all unless that pushback is recorded for the move, and that is the caution to carry into a match.

Advanced: using the numbers in a round

Plus frames, and what they do not buy

At +4 with Bryan's b+1, your jab arrives on frame 10 and theirs on frame 14. Every button they press loses, and loses as a counter-hit. That is the whole value of plus frames, and it is a lot.

Here is what it is not.

It is not a jail. Being plus does not force them to keep blocking. Only a string documented as jailing does that, and jailing is an authored property of the move, not a consequence of the number. If they can act, they can duck.

It is not step-proof. A sidestep can be buffered out of blockstun, so they can be moving before you are. Whether your follow-up catches them depends on that move's tracking, not on your advantage. wavu's tracking measurements stop at +6 for that reason: past it the step comes out so early that the attacker realigns anyway.

It is not armour-proof. A power crush absorbs the jab and keeps swinging. The answer to armour is a throw or a low, not a faster button.

And it is not always plus. A g on the number, as in +17g, means the opponent can guard through your recovery. Those tend to be the largest plus values in a character's kit, and they are free pressure for nobody.

Minus frames: press, hold, or duck

At minus your options rank by how much you know about what is coming.

Block. Always available, costs only the turn. This is the correct default and it is undervalued at every rank.

Duck, if the next thing is a high. The largest single reward on defence. A ducked high leaves them whiffing and hands you a while-standing punish, and on many characters that punish launches.

Step, if the next thing is linear. See below, because this is the option most often taught wrong.

Backdash. It holds standing guard for its whole duration, so it covers highs and mids and loses to lows. It also cannot be fully buffered, and wavu is blunt about the cost: try to backdash at −9, mistime it by one frame, and an i10 attack catches you. That is the reason people throw jab strings at +9.

Press, if the arithmetic says you win. At −9 your jab arrives on frame 19, which beats a slow launcher and loses to their jab. Pressing at minus without doing that sum is how most counter-hit damage in the game gets scored.

Interrupting is a punish with the guarantee removed

A block punish and an interrupt are the same subtraction. The difference is that a punish is guaranteed by the frames and an interrupt is a bet on which move they picked.

Your fastest move is the same i10 jab, so the sum is short. At −5, the jab arrives on frame 15 of their clock and beats anything landing later than that. At −9 it arrives on 19. At −13 it arrives on 23, which is slow enough that most things they could throw have already hit you.

Two consequences follow. Interrupting gets harder the more minus you are, in a way that is linear and easy to feel in a match. And when the bet is wrong, you were the one in startup, so you take a counter-hit, which is the most expensive version of being wrong that the game offers. Look at the "Launches on counter-hit" list on their character page before deciding a spot is worth contesting.

Whiff punishment is the same problem with distance added

When a move misses, the attacker is stuck in it. Expand any move row on a character page and it shows a recovery figure alongside the startup: Bryan's snake edge df+3 recovers in 33 frames. That is a long time to be standing still in front of someone.

The part the frame table cannot answer is whether you can reach them. Range is published on 421 rows across twelve characters, so most of the roster has no reach figure at all, and pushback is recorded only in the quiz's own hand-checked facts rather than in the tables. The window length is on the page; how far away you were standing when it opened is not. That second half comes out of practice mode.

Stepping, and why "step at −5" is a starting point

The rule of thumb goes: at around −5 or better you can sidestep the next thing. It is a reasonable place to start. It is not a law, and the difference matters, because the failure case here is eating a launcher.

What the frames actually buy you is time. At −5 you are five frames behind, and a sidestep can be buffered out of blockstun, so you can be moving early. The frames decide whether you get to start the step.

What decides whether the step works is the move they choose, and that is a different fact entirely. wavu's Sidestep page puts it plainly: there is no hard-coded crush mechanic for sidesteps the way there is for highs and lows, so evasion is decided by hitbox and hurtbox shapes. A sidestep has no guaranteed component at all, at any frame advantage.

Four things beat the step regardless of your number:

Homing. It realigns during startup and catches a step in virtually any situation. Yoshimitsu's f+3 is a homing high at −9; Bryan's df+3 is a homing low. Homing is not the same as safe, though. A homing high still ducks, so a character carrying a stack of homing highs is hard to duck, not hard to step.
Tracking on the side you chose. Tracking is directional, and which side is weak is not in our data. The tracking score exists on 153 of the corpus's 6,433 rows, from sixteen characters, and it is measured from a standing start only. wavu states outright that the method does not work for moves inside strings.
A delay. Changing the timing of an attack ruins a step that was timed for the old one. There is no crush property to fall back on when that happens.
Realignment. A wavedashing attacker realigns as they move, which gives otherwise linear moves something close to tracking.

The safer version of the habit is a step block: step, then press back to regain guard. wavu's figure for it is roughly twelve frames of sidestep before the block, which is enough to make a lot of moves whiff while still covering the ones that track. That text predates Tekken 8. Treat the twelve as a shape, not a measurement.

So the honest form of the rule: at −5 you have the frames to start a step, and whether it evades is a per-move fact you learn from the lab or from a patch note. The frame table will not tell you, and it does not pretend to.

Expert: strings, gaps and the edges of the data

A string is several frame situations in a row

Each hit in a string has its own on-block number and its own arrival time. That arrival time is counted from the previous hit, not from neutral: ,i16 means sixteen frames after the last hit's active frames. Those rows are marked rel on this site and never sort as if they were startups.

Between any two hits, you are in one of three places.

1. Jailed: you are blocking, and that is all

If a continuation jails, blockstun holds you until it arrives. You cannot duck, step or press. Kazuya's 1,2 reads "Jail from 1st block with 4f delay", so the jail survives up to four frames of delay on his part.

Jailing forces a block. It does not make the damage guaranteed, and the two get confused constantly.

The real problem with jailing is what is missing. It is recorded on about one continuation in eight. A string with no jail note has usually not been checked, which is a different thing from cleared, and it is why the string panel on every character page prints that character's own silence percentage instead of a roster average.

2. A published gap: press the button

wavu sometimes publishes the exact speed of move that beats a continuation, and those notes are the best-supported thing in this whole topic. Jin's 2,1,4~4 carries "Interrupt with i16 from 2nd block". Block the 2,1 and anything of yours at i16 or faster beats the low that follows, which on a lot of characters includes a launcher.

Roughly two thirds of the published gaps need something faster than i10, and nothing you can reach out of blockstun is that fast. "There is a gap here" and "you can use the gap here" are separate sentences. The character pages list only the usable ones, on the strength of that split.

3. Nothing recorded: silence, not clearance

Most branch points are in this third category. No jail note, no interrupt figure, no sidestep note. That means nobody measured it.

Parryable gaps

The universal low parry catches lows and nothing else, so a gap is parryable when the hit arriving is a low and you are not jailed into blocking it.

Jin's 2,1,4~4 is the shape. 2,1 is a mid at −3 that jails from the first hit, so you have to block that far. The 4~4 after it is a low with a published i16 window, and a low is what a low parry catches.

Given the choice, take the launch. Tekken 8's low parry flips them into a tornado for a scaled, grey-life punish where Tekken 7 gave a full combo, so parrying a gap you could have launched costs you damage. The parry is the answer when your launcher does not fit the window.

Fuzzy guard: one input covering two branches

When a string branches into a fast high and a slower mid, a brief duck can make the high whiff and still be standing again before the mid arrives. The community name for it is fuzzy guard, also called a flash duck or a fuzzy duck.

It works because crouch state in Tekken follows whether down is currently held. It is not a timed animation you have to wait out. The 6-frame and 11-frame crouch figures people quote are stance milestones. They govern which moves you get, not whether a high can hit you.

The worked example is Lidia's df+1. The high branch df+1,2 arrives at ,i16~17 and the mid df+1,3 at ,i23, so the mid is seven frames later. Lidia cannot fix that by delaying: the mid publishes "Move can be delayed 5f", and delaying the slower option only widens your window. The branch that closes a fuzzy window is always the high one.

Two limits on all of this, and both are the honest kind.

Nobody knows how big the gap has to be. The crouch to standing-block transition has never been measured for Tekken 8. wavu treats it as immediate, and that assertion is a Tekken 7-era edit on a page shared between the two games. Tekken 8 then shipped a fix for "certain high attacks could still hit at specific timing even when crouch was input", so the model was not reliably true until recently and nobody has re-labbed it since. This site publishes the gap size and refuses to filter on a threshold it cannot support.

A missing delay note usually means nobody checked. Leroy's db+1 is the same shape with a three-frame gap, and its high branch db+1,3 is annotated just "delayable", with no number attached. The window can be closed and the data cannot say by how much.

The version with no timing at all

The strongest thing the corpus supports is simpler than any of that. Where every branch from a point is high and the hit before them is also high, ducking beats the entire string with no window to hit and no dependence on jail notes, because a defender who ducked was never in blockstun for a jail to hold.

Bryan's 1 is the example. The jab is a high, and 1,2 and 1,4 are both high. Duck the jab and none of it can reach you.

Each character page carries these under a panel headed "What you get to do about Bryan's strings", or whoever you opened. The groups inside it are ordered by how well the data supports them, not by how good they sound.

Crouching, standing back up, and the number that does not exist

Crouch state is whether highs whiff and lows can be blocked, and it follows the held direction. Crouch stance is which moves you have access to: half crouch after six frames of holding down, full crouch after eleven, and while-standing lasting about six frames on the way back up out of full crouch. Conflating the two is what produces the belief that ducking takes eleven frames to work.

Crouch cancel is getting from crouch back to standing without passing through while-standing, which is how you reach an ordinary standing punisher. The cheapest version is an up-sidestep at one frame, fully bufferable; dashes cost two and the final input cannot be buffered. That cost is real and wavu gives a case: Lidia's i13 R.f,F+2+3 punishes −14 from standing but only −15 from crouch.

Inside a string's blockstun you can already crush highs by holding down, and in that state db is the reliable input. Plain d is not.

What nobody has is the number of frames between releasing down and being able to block a mid. wavu asserts it is zero, from that same Tekken 7-era text. Nothing on this site puts a figure on it, and you should distrust any guide that does.

Under online delay

None of the numbers change. A frame is a frame, −13 is −13, and every sum above holds.

What changes is which half of your defence still functions. Split it in two:

Arithmetic survives. Punish tiers, interrupt windows, whether your i15 launcher fits a −15, whose turn it is. You worked these out before the round started and they are the same numbers.

Reaction degrades. Blocking a low on sight, whiff punishing something you did not expect, breaking a throw, ducking a specific ender. All of it depends on seeing the animation in time, and online the picture you are reacting to comes through a network.

The reactability thresholds this site uses for lows come from guide authors' own numbers: at i19 and faster a low is too fast to see coming, from i20 to i23 some players react to it, and at i24 and slower it is blockable on reaction. The middle band is the one that goes first online. wavu's reaction model puts a trained player's floor around i18 to i19 for a single choice with everything in their favour, and offline there is already a floor of two to three frames of input processing before your controller and screen add theirs.

Nothing in this site's data measures what a connection adds on top of that, so there is no number here to give you. The practical answer is to move work out of reaction and into preparation: decide before the set what you will do when they throw the thing, and pick punishes you can perform without thinking about them. The quiz exists to make the recall automatic, and the trainer exists to make the input automatic.

What frames never tell you

Four gaps, and no frame table is going to fill them.

Range and pushback. Barely recorded. A punish that fits on paper and whiffs in a match is usually this.
Hitbox shape. Absent entirely, and it is what sidestep evasion actually comes down to.
Which side to step. Not in the data, and mostly not in the patch notes either.
The opponent. A −13 you never punish is a −13 they will keep throwing.

Everything above is general. The character matchups page and the priority board are the parts that are about you, and they are what decides which matchup is worth spending this on first.

Using frame data: beginner to expert