Is There Even A Statistical Difference Between KdF Guards?

At my fencing club Bucks Historical Longsword, our normal class format is to go over a passage from the Lew gloss, then play a game or drill based on it and talk about it. Lately we’ve also been collecting data on asymmetric games to see which side has an inherent advantage, if any. We just happened to reach the section on the vier leger (four positions or guards) in the midst of our data collection streak, which provided an interesting opportunity: can we use data to determine which guard out of the four is best? 

At first this seems like a silly question, and it is in a way. Fencing is fencing, and while different guards do afford different opportunities, the options of both sides aren’t what determine who will succeed in fencing, but how and when they are used. 

When we talk about guards, or when we do something that is supposed to counter a guard, we tend to play a game with very simple rules: regular fencing, but one side must exclusively use the guard in question. From this we get ox game, pflug game, alber game, and vom tag game. First up was the ox game, and before we did it I didn’t think we needed to take data on it, it’s not like a normal asymmetric game where both sides have some kind of constraint placed upon their fencing. With the guard games, only one side has a constraint, while the other can do whatever they want. The point of this game is not to be perfectly balanced, but to compel the fencers to explore the tactical landscape of each individual guard, both fencing from it and against it. 

However, after asking, a couple people wanted to take data, so we did it. Taking data is unobtrusive, all you need to do is report the score of each set of 10 reps (as a positive side effect, making sure you keep an accurate score also forces people to make sure they know who succeeded in each rep, which is critical to game-based learning – if you think you’re succeeding when you are not, then your takeaway of the game will be skewed). Doubles were resolved by throwing them out and ignoring the exchange; each set must end in a score adding up to 10 (IE 5-5, 6-4, etc) (in hindsight if I were to run this as a real experiment, I would record doubles as an additional data point, but at that point data collection starts to become a bit obtrusive for a class in which people are trying to improve at fencing and are not experiment test subjects). Everyone must also do both sides of the game with each other in order to normalize for skill difference between fencers. 

The null hypothesis was that the guard side would always be losing, though maybe to varying extents. The reason for this is that the guard side has constraints on their actions (must start from a specific guard) and the other side does not, therefore the side with more options should have the advantage. The non-guard side can play however they feel most comfortable rather than being forced to use a guard that they might not be familiar with. My prediction was that pflug would probably do best, followed by alber, then ox and vom tag (only overhead variant of vom tag would be tested, that is the only one in Lew that is named vom tag), because the latter tend to be used less than the former. The most commonly used guards at our club tend to be shoulder guard (shoulder vom tag for Ringeck and Danzig enjoyers), pflug, and longpoint/sprechfenster. Alber and ox see moderate use, and vom tag is rare. 

The test was underway, and data collection began. We had 9 people in class the first day, forming 3 groups of 3, with each group doing a full rotation twice, resulting in 18 sets and a sample size of 180 repetitions. The result was 90 for the ox side, 90 for the non-ox side. 

After the game, we discussed what happened, what was good or bad for either side, and in this case, why we got the unexpected result. We do have some fencers who regularly use ox and have developed effective games with it, but by and large it isn’t very commonly used. It could be that we found it just as difficult to attack ox as we did fencing from it, so it evened out. People also found that the “limited options” factor of the ox side was not a huge deal because you can simply leave the guard with a preparation when you attack (which is something that you always do no matter what guard you are in, because you need to move in order to attack). At any rate, I was eager to see what would happen when we did the pflug game because I know that we are in general more comfortable in pflug than ox, but surely the guard side can’t be better than equal with the non-guard side. 

With pflug, we had a similar class size with different people, and out of 180 reps it was once again 90 for each side. By now a pattern is starting to emerge. We can see that both sides are tending to be roughly balanced. I’ll post the rest of the numbers below before discussing further.

Given these numbers, I’m comfortable calling them all about even. Alber is a bit of an asterisk, because not everyone did both sides with each other (though most did), which can explain why the guard side did slightly better. Still it is within one 8-2 score in non-alber’s favor. Vom Tag was a little bit more skewed in the direction that we expected in the null hypothesis, but is still quite close. I have a couple of speculations about why this happened:

• Vom Tag is our least familiar guard – during the discussion I asked for a show of hands to see who uses the guard regularly in fencing, and no one raised their hands.
• It is scary to fence from – Unlike other guards where your sword is in front of you, in this guard your sword is out of sight, making you feel exposed, and possibly reluctant to attack (which is important because the threat of attack is arguably the main feature of this position)
• It is harder to move in – raising the sword over your head raises your center of gravity, requiring slightly different movement patterns to move around than people are used to, which can result in feeling “stuck” in place and not moving as much as you may normally. 

As a follow up to this, I would be interested to see if the results even out if the game is played after some other games like Direct Attack and Soviet Foil are played while starting in Vom Tag, which would allow fencers to get used to moving and attacking from the position. 

At any rate, all of the games ended up roughly even. What possible conclusions can we draw from this?

Guards don’t matter(?)

Yes and no. I think there is an element of this data that suggests that you can be successful in any guard, but I think it still makes a difference what guard you stand in. One other thing we can look at is the volatility of the data – some guards resulted in more scores closer to 5-5, and some had a wide variety of scores. This can partially be the result of fencers of different skill level coming to each class, but it also might suggest that a guard can have an equalizing effect. So maybe if you are fencing someone who you think is more skilled than you, or if your opponent has scored a few points and you want to change the game to stop their momentum, you might try using ox. Conversely, you can use a more volatile guard like alber or vom tag to press an advantage. The optimistic takeaway of this conclusion is that there is probably no best guard, because then the game of fencing would degenerate to just using that guard, and variety is cool. 

Presenting one side with more options than the other does not necessarily give that side an advantage

I think this is the most solid takeaway from this data, and it has implications in both fencing and game design. Before taking data on these games, I had intuitively assumed that, all other things being equal, giving one side fewer options than another would make that side weaker, but that seems not to be the case. From a fencing perspective this makes sense, you don’t need a wide variety of moves to win, and developing a very refined narrow game can be extremely effective (IE “area of excellence” from Epee 2.0). From a game design perspective, in the future I might not be as apprehensive of adding constraints to one side and not the other, or fewer on one side than the other as I have been. 

Future Research

Here are some ideas for future growth that I would consider trying some day:

Guard matchups

We can see that it is fairly even if one side is in a guard and the other is free to do whatever they want, but what if both sides must be in a specific guard? I would like to see how it plays out when each guard is put up against all other guards. Do any guards hard counter any other guards, or possibly give a slight edge? Maybe someday I will try to tackle this, but it would require a lot of sets to accomplish.

Shoulder guard and Sprechfenster

These are my most commonly used guards, I would like to see how well the pattern holds if we test them.

Doubles

I am interested to see how many doubles tend to result in these guard games. Are they all equal, or are some guards more likely to result in doubles than others? Along these same lines, I would like to see what happens when doubles are resolved in different ways, such as right-of-way priority, as opposed to throwing out doubles. 

The Follow-Up

I couldn’t rest knowing that Vom Tag disrupted the pattern of all guards being equal, so in our next class (after writing the article above), I gave everyone a choice: we can revisit the Vom Tag game, only this time we well do a round of Direct Attack drill and a round of Soviet Foil drill in which the attackers must start in vom tag instead of shoulder guard prior to re-doing the Vom Tag game, or we can move on and never look back. The consensus was that we revisit vom tag, so that’s what we did. 

The null hypothesis was that the results of the Vom Tag game would be closer to 50/50. The Direct Attack drill was meant to target a fencer’s possible reluctance to attack from the high position; it will optimize their attack and get them used to the ideal distance for it. The Soviet Foil drill was meant to address the movement issue; a wide variety of movement is required to succeed in the SFD. 

There were 9 people at class, with 7 of them also having participated in the initial Vom Tag game last Friday. Out of 170 reps (one group forgot to report one set), the result was 75 (44%) for the vom tag side, 95 (56%) for the non-vom tag side. 

Instead of improving the performance of the vom tag side, the drills made it worse! It’s definitely not what I hoped for or expected, but it’s a more interesting and exciting result, because now we get to think about why it happened. Here are a couple of possibilities we discussed after the results were in, and I came up with after the fact:

Everyone was tired

VT is a tiring guard to stand in because you need to hold your sword over your head (even if you rest your hands on your mask), and now we’re preceding the game by 2 other games where you need to hold the position. It’s possible that fatigue negatively affected the performance of the VT side. 

Fencers were better at defending against it

It’s possible that between the games and doing the full drill on Friday, everyone had improved more at exploiting the weaknesses of VT than at actually using the guard. Maybe a little bit of the “Kote Dilemma” effect going on.

Fencers became more confident at doing strikes that are easy to counter

This one I came up with on my own after class. It’s possible that by doing the DA and SFD, everyone tended to become more confident at throwing direct attacks without fully optimizing them for the higher guard, and thus were easily countered. I think threatening and throwing a direct attack from an overhead position is necessary to make the guard work, but the mechanics and how your body optimizes the cut are different than doing it from the shoulder, and require practice. Also you need to mix it up with other moves in order for it to have a chance of landing, if you do it too much then the opponent will be able to easily counter it. So it’s possible that doing the games made people more likely to throw cuts that are suboptimal and/or too frequent. 

Not enough time to manifest the skill

The area where CLA-based games really shine is with long-term retention. Often in the short term, classical blocked IP-based drills can have equal or better results than CLA, but are just terrible for retention. It’s possible that the timeframe between practicing the skills in the games and trying to apply them was too small, and the result may be different if we had waited. 

The games don’t work

It’s possible that the games just don’t work, and won’t make anyone better at fencing. Thankfully I think we can throw this one out, because long term results of using them as training tools suggest that they do work very well. However, what this does suggest is that they may target skills in a different way than I think, and may produce unexpected results. 

Full Data Set