[Haer]

One of the most frequent complaints I see on the forums/IRC is that a particular round has 'bad' stats'. To some extent such judgements will be based on personal taste, but it has lead me to wonder - is there some set of objective criteria by which we can say a set of stats are better/worse than another set? Personally, I think if we could find some kind of criteria like this it would help us to be a lot more certain about which rounds really had good stats and which didn't, and in the end lead to better stats for future rounds.

To get things started, I'd like to propose a few things which could be looked at. I will begin by saying that I am in general sceptical of the ability of people to judge a set of stats just by looking at them; it may be possible, but I suspect it's just too complex a problem to accurately judge in all but the most obvious of cases. I would be interested in any ideas people have about such a method.

Moving onto specific criteria, I think there are two things a good set of stats should have; balance and variation.

People often throw around the word balance without really defining what it means. In my mind, there are actually two separate things which need to be balanced - races and attack/defence.

Race balance means that each race has a roughly equal chance of doing well. People often look at the distribution of races in the top100, and I think this is a fairly good measure, although you should probably look at the top10/50/200 as well, and perhaps normalise by the total number of each race.

For attack/defence balance, this means roughly how easy it is to land/stop an attack. This can be a little harder to measure, but something that could be looked at is the number of launched attacks which land and the number of launched attacks which cap. If further data was available, you could look at for each withdrawn attack, what ratio of defending value to attacking value was necessary on average.

For variance, there are again a few different things you could look at. I think you could break this down into per race ship strat and overall strategy.

For each individual race, you'd ideally want a least 2 feasible ship strategies and maybe more. If data was available about the total distribution of ships in the universe, you could try to do some kind of clustering and then count the numbers.

For an overall strategy, you could look at doing something similar within statistics like # of attacks launched, # of def fleets launched, # of FCs/MCs/Dists, Gov. choice, etc. Again, you could try a similar kind of analysis - perform clustering and count the number of clusters.

While figuring this sort of stuff out isn't trivial, I dont think it would take a massive amount of work either; much of this can be calculated using standard software libraries. The main obstacle as I see it is lack of data; We already get some random stats about the universe, it'd be really nice if we could see this sort of information included for future rounds.