Guards Revisited

Prologue: I wrote this with my conclusions, but I will write it here as well in case you don’t read that far: Please replicate this experiment. If you are in a position in your club where you can decide what activities your fencers do, please take some time to do this. It is fun, helpful for your fencers, and will make an interesting comparison to our results. 

In July 2022, I published an article on SwordSTEM about an experiment we did at my club Bucks in which we tried to figure out whether any of Liechtenauer’s four cardinal guards is particularly advantageous in fencing. At the end of that article, I wrote that further research would include doing this for all matchups between each guard in order to further refine the data. Well now we’ve done that, and I want to talk about it. 

The Experiment

The previous experiment consisted of playing a game in which one fencer starts in a specific guard and the other is free to do whatever they want. A set of 10 clean (non-double) exchanges was done, and the number of successes from each side was recorded. Doubles were thrown out and not counted. This was fairly low-effort to do, consisting of 4 total runs which we finished over the course of three practices.

The new experiment was similar, but instead of each guard against an opponent doing whatever they wanted, both sides started in a prescribed guard. The fencers did not have to stay in the guard the whole time, they just needed to engage in the guard, and return to the guard if there was a separation. So one could, for example, chamber and cut from above when starting in pflug. We did this with all four cardinal guards (ox, pflug, alber, vom tag/shoulder guard) plus sprechfenster. Aside from many more runs, we made a couple other changes to the experiment setup:

  1. We collected data on doubles – instead of throwing out doubles and continuing until 10 clean exchanges were done, we did 10 exchanges total, counting wins for A, wins for B, and doubles.
    1. We counted “same action” as the timing for doubles, that is the strike had to be in progress when the first strike hit in order for it to count as a double. Afterblows were not taken into account.
  2. Overhead was not prescribed for vom tag – for that guard, participants were allowed to use overhead or dominant side shoulder variants. This was in an attempt to make the data more relevant to actual fencing, since the overhead variant is a lot less common than shoulder.

The experiment consisted of 15 runs completed over the course of 8 practices. We included runs in which the same guard matched against itself, and in those cases we only recorded clean hits and doubles. As with the previous experiment, each set of partners fenced each other using both combinations of guards, in order to account for skill disparity. 

We ended up with a percentage of exchanges ending in a double, and exchanges ending in wins for both sides. For wins on both sides, I thought it would be easier to read if we excluded doubles, so that way you can see if a side has more or less than a 50% win rate, which gives you a more meaningful picture of whether a guard has an edge over another. The resulting chart with those numbers, plus win percentage including doubles, and the raw numbers that we got when we did the reps, is included at the end of this article.

For readability, I separated it into three different charts, one showing win percentage not including doubles, one showing total wins per exchange, and one showing doubles. Colors were added so you can see matchup favorability at a glance. The cutoff for each color is arbitrary, those numbers just felt right to me. Percentage of exchanges won excluding doubles is a good metric for comparing two guards head to head, because it shows advantage at a glance; if it’s over 50%, then there is an advantage. Wins per exchange is a good metric for comparing overall guard performance, because you can see how likely it is that fencing from each guard will result in a clean exchange. Doubles shows how likely each guard is to double.

Observations

Going into this experiment, there were not really any specific null hypotheses (which makes it not an experiment? I don’t know, but it doesn’t matter to me), but in hindsight there were some things that were surprising to me. I’ll list a couple things that I noticed.

Alber Supremacy 

While it did not win overall, I was surprised at how well alber did. Low guards (non-point forward, hands low) tend to be less used than most guards other than ox, so the general expectation would be that alber would not fare as well. However, it did quite well, it was the only guard other than pflug to only have one unfavorable matchup. It did come in third on average, but pflug and sprechfenster are both inflated by their destructive power against ox. 

OxPflugAlberVT/ShoulderSprechfensterAverage (mean)
Alber51.70%51.90%/50%48.40%50.50%

In fact, there is a general trend on the chart of guards where the sword is lower beating guards where the sword is higher. In order of height, lowest to highest we have alber, pflug, sprechfenster, ox, and shoulder. Here is a chart that shows all of the matchups that follow the low-beats-high trend:

OxPflugAlberVT/ShoulderSprechfenster
Ox/YesYesYesYes
PflugYes/YesYesYes
AlberYesYes/NeutralNo
VT/ShoulderYesYesNeutral/No
SprechfensterYesYesNoNo/

Following this low-beats-high idea could also explain the small deviations that were found on the original 2022 experiment, in which guards went up against a fencer doing whatever they wanted. The result was ox and pflug were perfectly equal, alber was slightly better than equal, and vom tag (overhead variant) was slightly worse. Assuming the opponent is using a fairly even distribution of guards, pflug and ox will see some higher and some lower, alber will see everything equal or higher, and vom tag will see everything equal or lower.

One way you can look at alber is that it is the guard in which you are most likely to land a hit on your opponent, be it with a clean hit or a double. To find this number, we can add the percentage of exchanges ending in a double to the percentage of wins with doubles included, and find the mean across all guards. By doing this, we get the following numbers: Ox: 58%; Pflug: 61%; Alber: 63%; Vom Tag/Shoulder: 60.93%; Sprechfenster: 59.33%. So if you really need to hit your opponent and you don’t care if it’s clean or a double, Alber is probably the best choice.  

Sprechfenster: Noblest and Best?

Sprechfenster was notably absent from the previous experiment, but we decided to include it this time, both for the sake of completion, and also to see if it really is the noblest and best defense at the sword as the RDL glosses claim. For this guard, I was mainly prescriptive about standing with the non-dominant foot forward, and making sure there is some space between the fencer’s hips and their hands. It doesn’t have to be fully stretched out, that is a terrible position, but it shouldn’t be retracted, otherwise there’s too much overlap with pflug. My general expectation going into this was that sprechfenster would actually do the best, because putting your point in the opponent’s face is a good idea, and I feel like I’m personally comfortable there. In reality it came up pretty even, it has two good and two bad matchups, and is only positive on average because of how much it absolutely dumpstered ox. 

OxPflugAlberVT/ShoulderSprechfensterAverage (mean)
Sprechfenster56.00%48.10%51.60%48.20%/50.98%

It follows the low-beats-high trend for pflug and ox, but notably bucks it for alber and shoulder. I don’t know why that is exactly, maybe having a point in your face is more intimidating when your sword is low and you can’t see it than when it is high and in front of you. 

Despite its average showing in terms of matchup, where sprechfenster really shines is in rate of doubles. It has great matchups against all other guards, and even despite the inexplicably high double rate against itself, boasts the lowest average double rate. 

OxPflugAlberVT/ShoulderSprechfensterAverage (mean)
Sprechfenster17.30%13.60%18.30%18.80%25.80%18.76%

If I had to speculate on a reason for this, it would be because the blade in sprechfenster demands interaction. It is less comfortable to go straight in for a deep target without first either feinting to draw a movement, or engage the blade in some way. The exception would be non-committal shallow attacks, which are enabled because the hands are extended out in front, and I wouldn’t expect them to result in doubles, unless you were also in a position in which your hands are extended out in front. The jump in doubles when both fencers are in sprechfenster was quite surprising, but if I had to come up with a reason why, that would be it. 

In light of this, we can take another look at the line from the glosses about sprechfenster being the noblest and best:

“You have previously heard how you shall assume the four guards in front of the man with the sword, now you shall know that the speaking window is a guard in which you may well securely stand, and the guard is the long point, it is the noblest and best defense at the sword. Whoever can correctly fence from it, he forces their opponent with it, so that they must allow themselves to be struck against their will, and may additionally not come to strikes.” 

(Lew Gloss, Codex Ⅰ.6.4º.3, translation by me)

While it clearly didn’t win every exchange, the idea of forcing your opponent to act in a certain way stands out to me. If the low double rate actually came as a result of a need to engage the sword, then you are indeed compelling your opponent to act in a way that engages the sword, and maybe that’s what they mean by noblest and best. Though I suppose you might be able to say that about any guard, as each position affords different actions from the opponent, so who knows. 

However, after looking at all of this, I realize there is one stat that I overlooked: percentage of wins with doubles included. While wins without doubles gives a clearer picture for comparing one guard to another in a matchup, wins with doubles gives a better measure of actual overall wins per exchange. When I average all of the wins with doubles included, here are the results: Ox: 36.84%; Pflug: 42.03%; Alber: 38.06%; Vom Tag/Shoulder: 38.68%; Sprechfenster: 42.33%. Therefore, in terms of overall exchanges including doubles, sprechfenster came out on top with the most percentage of wins. Noblest and best. 

Similarity of Retracted Point Forward Guards

Ox and pflug, both guards with the point forward and hands retracted, faired similarly against shoulder and alber. Alber won 51.7% and 51.9% of the time against ox and pflug respectively, and shoulder won 48.6% and 48.5% against the same guards.

OxPflug
Alber51.70%51.90%
VT/Shoulder48.60%48.50%

This stuck out to me while perusing the charts after color coding them. My takeaway from this is that retracted point forward guards interact similarly to non-point forward guards, and it is advantageous to approach them from below. 

Interestingly, “hands retracted” seems to be a key element of this, because the effect was reversed for sprechfenster

OxPflugSprechfenster
Alber51.70%51.90%48.40%
VT/Shoulder48.60%48.50%51.80%

Here, shoulder closely resembles the alber values for ox and pflug, and the opposite for alber. For some reason, it seems that having the arms extended makes it more advantageous to approach from above than from below when using a non-point forward guard. 

For point forward guards, the pattern seems to hold true – pflug is lower than sprechfenster and beats it, and both pflug and sprechfenster beat ox.

OxPflugSprechfenster
Ox/43.50%44.00%
Pflug56.50%/51.90%
Sprechfenster56.00%48.10%/

I think the takeaway from this is, most of the time it’s better to be lower than your opponent, but not always. This is of course not taking into account doubles.

Ox: No Big Surprises

In the previous experiment in which the guards went up against someone who could do whatever they want, the null hypothesis was that ox would do badly, and it was a surprise finding when it turned out to be equal. For ox, I was prescriptive about the hands being generally high and the point being generally forward, within those constraints you could do mostly whatever you want, and you can do both left or right side. When it comes to this experiment, ox overall fared as expected. It had the lowest overall win rate, only seeing an advantage against shoulder guard. 

OxPflugAlberVT/ShoulderSprechfensterAverage (mean)
Ox/43.50%48.30%51.30%44.00%46.78%

This in a way corroborates my previous observation, that it is advantageous to engage from below against a retracted point forward guard. In the case of ox, everything is below it except for shoulder guard. 

Does this mean I will never use it in fencing? No, it just helps specify the use case. I can use it as a tool to make someone less comfortable engaging from a high guard. If I am not liking how someone is using shoulder, I can go into ox, and they’ll either change to something I’m more comfortable with, or they’ll allow me to play at an advantage. 

How to Double

Moving to the doubling data, there are a couple of things that we can note here. First, if you want to double, your best bet is probably alber. Alber had by far the highest double rate, only quelled by sprechfenster and, to a lesser extent, itself. The general preference in HEMA is to try not to double, but in a competition you may find yourself in a situation in which doubling is advantageous. This, combined with its overall positive win rate, makes it a good choice.

Percent of Exchanges Ending in Doubles
OxPflugAlberVT/ShoulderSprechfensterAverage (mean)
Alber25%26%19.20%28.80%18.30%23%

Maybe you are fencing someone who does not know about the matchup chart, and you are in a position where you can double and they cannot, and they decide to use a low guard. In this case, shoulder or vom tag is the move, the matchup that had so many doubles I gave it its own category. My speculation for why this happened is that the swords are so far away from each other, and there is no way to force a blade interaction, especially if you are the person in alber. 

If you are on the receiving end of this, and your opponent has taken alber because they don’t care about doubling, then sprechfenster is the clear choice. It not only suppresses the doubling aspect of alber, but also has a favorable matchup. Sprechfenster is generally a good choice if you are in a position in which you want to avoid doubles, as it does have the lowest overall double rate, blemished only by its matchup with itself. 

OxPflugAlberVT/ShoulderSprechfensterAverage (mean)
Sprechfenster17.30%13.60%18.30%18.80%25.80%18.76%

Pflug is the best by some metrics (but I don’t have to like it)

There is no denying how well pflug fared looking at the charts, both in terms of success and doubles. 

OxPflugAlberVT/ShoulderSprechfensterAverage (mean)
Pflug56.50%/48.10%51.50%51.90%52.00%

In terms of success, pflug boasts the highest average, its only losing matchup being alber, the only guard lower than it. This did come as a bit of a surprise to me, because I thought pflug was anecdotally less used in general in our class than other guards, though maybe I’m projecting since I personally don’t like to use it. 

In light of this, it may be no coincidence that pflug most closely resembles the common position that fencers tend to take in tournaments, IE dominant foot forward, point forward, sword somewhere between extended and retracted. In this experiment we were allowed to use right or left side pflug, though I don’t know what the distribution was of each side. When I personally performed reps, I used both of them, sometimes favoring the right side. 

Although these stats show that pflug has the most favorable matchups, if we take total wins per exchange (wins including doubles), pflug comes in second behind Sprechfenster. Still a good result, but a better result for the left foot forward extended position makes me personally feel a little bit better. 

Conclusions

Some quick stats

Here are some quick conclusions based on a few different metrics:

Average (mean) percents
DoublesWinsWins per ExchangeHits landed per exchangeNumber of favorable matchups
Doubles Not includedDoubles Included(doubles + wins w/double)
Ox21.18%46.78%36.84%58%1 (25%)
Pflug19%52.00%42.03%61%3 (75%)
Alber23%50.50%38.06%63%2.5 (62.5%)
VT/Shoulder22.24%49.73%38.68%60.93%1.5 (37.5%)
Sprechfenster18.76%50.98%42.33%59.33%2 (50%)

Most wins per exchange: Sprechfenster

Most wins ignoring doubles: Pflug

Most favorable matchups: Pflug

Most likely to land a hit per exchange (double or clean): Alber

Least doubles: Sprechfenster

How this changes my fencing

There is a lot to take in here, and ultimately it was a project that we did for fun and to train fencing. Following this religiously by always switching to a guard that gives a slight advantage based on this chart would probably not be the best way to proceed. There is more to fencing than what guard you start in, and even if we take these numbers to be one hundred percent accurate, most matchups only give a slight advantage and by no means guarantee a win. 

That being said, there are a few things that I have been paying attention to now that we have done this.

  • If I am in a situation where I don’t want to double (more so than usual), I will be more likely to use sprechfenster.
  • If I am in a situation where a double is advantageous to me, I will be more likely to use alber.
  • I’ll use a lower guard like sprechfenster or alber whenever I am playing against someone in ox, rather than trying to engage in shoulder.
  • If my opponent is using shoulder or vom tag and I don’t want them to, I’ll switch to ox. 

Even though pflug is the best by some metrics, I still don’t like to use it, so I still don’t. Shoulder was and remains my default position, sprechfenster is my second favorite, followed by low guards. Others I use situationally.

I don’t think guards are a largely important part of fencing. I think you should be familiar with what they afford and how they interact, but at the end of the day fencing is not a video game, and having an advantageous guard matchup at all times will not determine the winner. 

Issues and Future Work

There are a lot of potential issues with these tests:

  • All of the matchups were not done with the same people, so the results are skewed based on who happened to show up to class on the day that we did them. 
  • Our club may have biases or tendencies that are different from the greater longsword community.
  • Judging is never perfect, and may disproportionately impact some guards or styles. 
  • Sample sizes varied from one matchup to another, again based on how many people showed up to class that day.
  • I personally participated in the study, so you get biases that come with both running and participating in a study.

What this experiment really demands is replication. We are only one club, and while we have a variety of fencers, we surely have biases and tendencies towards certain things based on the way we train. If you are in a coaching or instructing position at your club, please replicate this. The activities themselves are fun and helpful, and we will get more data points regarding each matchup. 

For our club, the next iteration of this experiment that we plan to do is the same thing, but using a right of way convention instead of treating all doubles equally. We think this may alter the results due to changing the way people fence, as well as revealing information about some of the more double-heavy matchups. This will be done hopefully some time in 2024. 

Full chart:

OxPflugAlberVT/ShoulderSprechfenster
OxDoublesClean HitsDoublesw/Doublew/o DoubleDoublesw/Doublew/o DoubleDoublesw/Doublew/o DoubleDoublesw/Doublew/o Double
26944074608755953880
21.70%78.30%19%35.20%43.50%25%36.25%48.30%22.90%39.50%51.30%17.30%36.40%44.00%
PflugDoublesw/Doublew/o DoubleDoublesClean HitsDoublesw/Doublew/o DoubleDoublesw/Doublew/o DoubleDoublesw/Doublew/o Double
4096321486589378434112
19%45.70%56.50%17.80%82.20%26%35.60%48.10%18.50%42%51.50%13.60%44.80%51.90%
AlberDoublesw/Doublew/o DoubleDoublesw/Doublew/o DoubleDoublesClean HitsDoublesw/Doublew/o DoubleDoublesw/Doublew/o Double
60936596239772894291
25%38.75%51.70%26%38.40%51.90%19.20%80.80%28.80%35.60%50%18.30%39.50%48.40%
VT/ShoulderDoublesw/Doublew/o DoubleDoublesw/Doublew/o DoubleDoublesw/Doublew/o DoubleDoublesClean HitsDoublesw/Doublew/o Double
5590377972894014045101
22.90%37.50%48.60%18.50%39.50%48.50%28.80%35.60%50%22.20%77.80%18.80%42.10%51.80%
SprechfensterDoublesw/Doublew/o DoubleDoublesw/Doublew/o DoubleDoublesw/Doublew/o DoubleDoublesw/Doublew/o DoubleDoublesClean Hits
38102341044297459462178
17.30%46.40%56.00%13.60%41.60%48.10%18.30%42.20%51.60%18.80%39.10%48.20%25.80%74.20%

Posted

in

by