Should we use monitors for SSBM tournaments? Depends on what you care about. Should we be informed about the differences between monitors and CRTs? Yes, especially in the midst of people with financial interest in the question. A short review of the empirical evidence about input latency perception and "lagless" monitors in Super Smash Brothers Melee.

With Low Tier City 7 sparking the dreaded monitors-or-no-monitors debate for the hundredth time, I figured it would be a good idea to gather the current empirical evidence relevant to the monitors discussion.

I think there are three fundamental questions to explore in this conversation, and I will attempt to summarize in this document what I consider "the literature" on these topics:

1. Do monitors have lag?
2. If monitors have lag, can humans perceive lag differences of such a small scale?
3. Do small changes in input delay affect gameplay?

I think these three questions are largely independent of each other; even if you can't tell that there is lag on your inputs, your reaction time / timing windows are all moved that small amount and will still affect your play (if lag affects play) regardless of how you feel about it. As such, I will be visiting each question separately below.

I frequently hear claims that high-end gaming monitors are "lagless" or, worse still, "1ms". Rather than listen to claims made on advertisements trying to sell you something, I encourage you to consult the wisdom of our very own world-class community technical expert - Kadano¹.

For this discussion, there are two important "myths" Kadano has debunked through empirical measurement. The first is that LCD lag exists due to analog-digital conversion. This one is pervasive, and to my knowledge is the claim used by EON to market their extremely expensive converters as "truly lagless". This is empirically debunked by Kadano; even the cheap $3 HDMI to VGA adapters are comparably also lagless, clocking in at a whopping 0.0033ms. No measurements exist for the EON adapters, but I'm sure we can take them at their word here, since it doesn't matter much for the chain anyways.

The display lag (which is different from response time, which is what typically gets advertised) on monitors was empirically measured by Kadano to be on the order of 5ms above CRT. I've reproduced his summary of the work below, and there isn't really much I need to add to it. The fact that monitors have this degree of delay is relatively unambiguously true to anybody informed on the topic and has been independently verified by other smart people in the scene such as Fizzi².

As a very important aside, Kadano mentions near the end here that input lag for emulation is much lower than console melee. The reason that this is the case is that native console has lag also! It just applies equally to CRT and LCD, so it's canceled out in measuring it. Once again, this is very well documented by Kadano, who measured the difference in console and dolphin input lag. Console melee has an inherent input lag of roughly 3 frames, and reducing this is what allows netplay to function close to CRT despite requiring signal to be sent between players. There's other wonkiness with emulation (see: delayed audio), but it's important to remember this for later: console play has ~3 frames of lag.

We have established that LCD Monitors, even good ones, unambiguously and demonstrably have somewhere around 5-6 milliseconds of lag. "But so what?" you might say, "it's obvious that 5-6 ms of lag is going to be completely invisible to human perception!"

Well, if there is any one thing I've learned from studying cognitive science, it's that pretty much nothing about human perception is obvious, pretty much in any capacity.

And indeed, I feel that a lot of the people making these sorts of claims of "clear obviousness" are doing so from a place underestimating the utter insanity that is human perception.

Finally, we arrive at the question of whether or not gameplay can be substantially affected by slight changes in input delay (in our case, of about 5ms). For the purposes of this question, I'll be considering the hypothetical scenario of a player who does not know there is lag and performs inputs identically the same on both monitors and CRT. This is pretty obviously not always the case (playing with what you feel as lag is distracting, and the placebo effect is definitely a real thing regardless of whether you can perceive the lag or not), but it simplifies the discussion. It's important to note that this question is independent of the actual perception of the lag, as if you aim for the center of a window and that window moves, your success rate will change even if you think you are aiming at the correct spot.

Importantly, input delay will not really affect your inputs relative to your inputs, since input delay is a global phenomenon. As a rule of thumb, if you could do something with your eyes closed, then input lag will not damage your ability to do it. Your wavedashes, for example, will be the same in both versions, since both inputs will be delayed by the same amount. Input latency affects only (and all) situations where you react to some sort of stimuli before acting - anything for which you use a visual cue, any reaction (e.g. tech chasing), and so on. So, it would be nice if we could model how the probability of your rate of success in these scenarios changes before and after the presence of some added lag.

Thankfully, this has already largely been modeled by Fizzi, who ran through some toy examples back in 2014. This article is frequently touted by monitor enthusiasts as proof that monitors do not affect gameplay. However, while I agree with all of Fizzi's methods and love this article, I drew the opposite conclusions from him using the same examples. Where I think Fizzi's article shines is that it precisely details hypothetical situations which measure exactly how different monitors would be from CRT.

Fizzi provides three examples of inputs being moved a few milliseconds for successful execution of a 2 frame window, and the resulting differences in probability of success. A perfectly centered mean with 95% success rate gets moved to 93.7%, a slightly off-center input with 62.3% success rate goes to 49%/74.2% (depending on direction), and a very laggy monitor (16.66ms) moves even a perfectly centered 95% success rate to 50%.

Fizzi uses this to argue that the difference is fairly small and only really affects players who are not that proficient at tech anyways, but I very much disagree with this assessment! I think that something like powershielding is often worth going for even if you're only 66% successful at it, and the fact that this moves so much is very significant to me. I think this is largely a subjective question, but personally I drew very much different conclusions from this exercise. Small changes to the means of gaussians can significantly move the percentage of the distribution that falls within a specific window for success, which heavily affects situations in which competitors often deliberately aim for the first or last few frames of the window rather than the center (i.e. wakeup, reaction tech chase)

But don't take my word for it - play around with it yourself! Below please find another WebPPL program you can use to model specific situations, and come to your own conclusion about whether monitors would affect gameplay in this toy example of melee. In my examples, the same gaussian that gets 95% -> 93% for fizzi's measurements moves the "techchase" example from 85% to 61% using lag measurements from Kadano's tweets.

var run_scenario = function(framewindow, frametime, aim, sig, lagamount) {
  var input = gaussian({mu: (framewindow*frametime)*aim, sigma:sig});
  var input_withlag = input+lagamount

  var success = (input > 0 && input < frametime*framewindow)
  var success_withlag = (input_withlag > 0 && input_withlag < frametime*framewindow)

  return {'success': success, 'success_withlag': success_withlag};
}

//Kadano lag measurements, perfect center mean: 95 -> 89%
var powershield = function() {
  var framewindow = 2 //window of technique
  var frametime = 16.66 //length of 1 frame
  var aim = 0.5 //where in the window you want to hit, range from 0 to 1
  var sig = 8.4 //stdev
  var lagamount = 6 //amount of lag, in ms

  return run_scenario(framewindow, frametime, aim, sig, lagamount)
}

//Fizzi lag measurements, perfect center mean: 95->93.7%
var powershield_fizzi = function() {
  var framewindow = 2 //window of technique
  var frametime = 16.66 //length of 1 frame
  var aim = 0.5 //where in the window you want to hit, range from 0 to 1
  var sig = 8.4 //stdev
  var lagamount = 2.86 //amount of lag, in ms

  return run_scenario(framewindow, frametime, aim, sig, lagamount)
}

//Kadano lag measurements, aiming for center of first frame on wakeup 85%->96%
var wakeup = function() {
  var framewindow = 5 //window of technique, let's say it's generous if you're late
  var frametime = 16.66 //length of 1 frame
  var aim = (0.5 * (frametime))/(framewindow*frametime) //where in the window you want to hit, range from 0 to 1
  var sig = 8.4 //stdev
  var lagamount = 6 //amount of lag, in ms

  return run_scenario(framewindow, frametime, aim, sig, lagamount)
}

//Kadano lag measurements, aiming for center of last frame of techchase 85% -> 61%
var techchase = function() {
  var framewindow = 4 //window of technique
  var frametime = 16.66 //length of 1 frame
  var aim = ((framewindow*frametime) - (0.5*frametime))/(framewindow*frametime) //where in the window you want to hit, range from 0 to 1
  var sig = 8.4 //stdev
  var lagamount = 6 //amount of lag, in ms

  return run_scenario(framewindow, frametime, aim, sig, lagamount)
}

var distpshield = Infer(
  {method: 'rejection', samples:10000},
  powershield);

var distpshield_fizzi = Infer(
  {method: 'rejection', samples:10000},
  powershield_fizzi);

var distwakeup = Infer(
  {method: 'rejection', samples:10000},
  wakeup);

var disttechchase = Infer(
  {method: 'rejection', samples:10000},
  techchase);

//visualize
print("Powershield")
viz.marginals(distpshield);

print("Powershield (Fizzi measurements)")
viz.marginals(distpshield_fizzi);

print("Wakeup")
viz.marginals(distwakeup);

print("Techchase")
viz.marginals(disttechchase);

It's very possible you don't agree with some part of this toy design; maybe your lag measurements are different, maybe your standard deviations are different, maybe you aim for a different part of the frame to reduce risk at the cost of being late or giving you less time to react - there are a lot of variables you can tune here. The point is that you can just see for yourself how your devised scenario changes from CRT to LCD. You can literally go to http://webppl.org/ and paste the above code into the box and run it with your proposed changes, and check if it makes a difference. Key here, though, is that "makes a difference" is entirely up to you. Is a 3% difference important? Is a 5% difference? How about a 20% difference?

I think changing a major part of the tournament infrastructure for the melee community is a huge decision, and as such I think it's very important that everybody be reasonably well-informed on the topic before a major decision is made (rather than after).

I think there are good, reasonable arguments for introducing monitor melee into the tournament ecosystem. The dwindling CRT supply, logistics and transportation, sponsorships, whatever. While I can't claim to particularly care about any of these topics (I'm not a TO, after all, I just play the game), they are reasonable arguments, and I'm willing to hear out people that claim the drawbacks of monitors are worth the benefits for the greater scene. It's not like the monitors are unplayable (I own one myself, even if I never use it).

What bothers me is when people make repeatedly falsifiable or unsubstantiated claims either to make themselves seem right or to fit whatever narrative they want to sell to people. I've heard any number of insane claims about this topic, everything from "all CRTs have more lag than monitors anyways" to "the only way to get lagless monitor melee is buying our $200 converter". These comments range from inane to outright dishonest, but I hope this document serves as some small resource a few people can use to learn about the topic in a centralized location.

But to ultimately answer the three original questions, here are what I consider to be the answers to them:

1. Do monitors have lag?

Yes, unambiguously and verifiably, monitors have 4-6ms of lag depending on the thresholds you use to measure it.

2. If monitors have lag, can humans perceive lag differences of such a small scale?

There's not that much research on the topic, and none of it is peer-reviewed, but I think it's very plausible that smash players can perceive lag changes between monitors at roughly the threshold in question, even though I think perceptually it is quite literally exactly the same for people who it does not affect.

3. Do small changes in input delay affect gameplay?

This depends on your thresholds for "affect gameplay" but I think the answer is a fairly obvious yes; moving your input windows around definitely affect your success rate, and whether or not you think it moves the success rate enough to make a difference depends on your tunable parameters for "enough" and "a difference".

If you have comments, I'm usually available on twitter @Ambisinister_, and thank you for reading!