Dukasaur wrote:Mets, all you've really done is restate with much effort that there's a huge element of luck on CC. That was never in dispute. I clearly stipulated that I know there's a huge element of luck on CC. No sane person has ever denied that there's a huge element of luck on CC.
I understood your point originally. I took the time to clearly explain how the luck works because the way it works in detail is precisely why your original argument is incorrect. The luck is not
noise in the sense that you think it is, and the arguments above illustrate why. I will explain more precisely why at the end of the post in case it was not clear from the above.
That being said, skill still matters. Luck will tend to balance out in the long run, so more skillful players will still win more. Even if the skill component is only 10% or whatever, still against the random noise of the various luck inputs that skill element will still show.
However, even if I agree that skill plays some discrete role that is just associated with a lot of noise, the argument is still an embarrassing one to defend. It is essentially a concession that something which only explains 10% of a person's odds of victory is enough to determine who remains in a tournament and who does not. There are so many other tiebreakers that can be easily defended without going to such trouble, like the combined winning records of all your opponents. Why would you therefore actively defend the one that we should pretty much all be able to agree is the worst at measuring skill? Can you even imagine something which does a worse job while actually still being relevant to the game? I'm having a hard time with that. "My formula gets it right 10% of the time guys" is just not a very convincing argument.
For Christ's sakes, Mets, you're a physicist. You of all people should know that it's possible to measure a variable despite background noise of larger variables. Your entire career depends on that possibility.
If the background noise is larger than the signal, which it will be for any given tournament, then the only thing you can do to beat down the noise is measure this over many tournaments and get some
average estimate of this signal. It can and will be highly variable for any given tournament, and at any rate the average will only make sense if you hold basically every variable (map type, number of players, fog, trench, spoils) constant. Now, your point could at least be defended on your own ground if the noise were small enough that there was no chance of the signal going negative in any given tournament. But it's quite likely that this is untrue. It is quite likely that in any given game, luck will play a much more dominant role than skill. Therefore your tiebreaker disqualifies people who may very well have done a better job than the people who won in fewer rounds, given the actual conditions they faced in your game. For all we know the people who won in fewer rounds did so because of deadbeats. And yet the people who had actual competition are kicked out of the autotournament by your formula. Why? Because on average, in some hypothetical games they didn't actually play in this tournament, you insist that they wouldn't have done as well. You have actually determined that they are less skilled players because they made the poor choice of being randomly assigned to a game where their opponents actually showed up.
This is all kinds of fucked. Your "signal" requires us to ignore most or all of the knowledge of the actual performance of the player in the actual tournament, which is not the hallmark of a tiebreaker; it's a skidmark. Tiebreakers only make sense if they are clearly and uniquely tied to the performance of the players in the tournament, but yours requires outside data to validate, which is immediately unfair to the people who actually competed.
For any given set of luck inputs -- card set x, dice rolls y, turn order ABC -- a player who deploys and attacks according to the theoretically optimum pattern will tend to win more often and more decisively that a player who makes non-optimum moves.
Yes, and this is the point that
I did not dispute, because it is only tangentially relevant to the conversation. The problem is how you determine who played closer to the theoretical optimum, not the fact that this concept exists. Your argument is that because an optimum exists, and we can meaningfully imagine play that is closer to or farther from that optimum, that therefore skill matters in Conquer Club. But the crux of the matter is
not whether skill matters in Risk. The crux is
whether time to victory is a proxy for skill. Only if we can infer skill (even a small amount of skill) from time to victory, can we actually use it as a tiebreaker. You have essentially taken it as a starting assumption that the optimal play pattern will always win you the game in the fewest number of moves compared to sub-optimal play patterns. But
this is not true, and I cannot emphasize plainly enough how starkly incorrect it is. The only place it might have been true was in your fictional board game of the previous post, that was closer to
Diplomacy than
Risk. It is simply false in the game we actually play. Because in the game we actually play, if the optimum tells you how many turns you should expect to win, your tiebreaker favors recklessly
disobeying that optimum on the hope that you'll get lucky and win faster, even if it was suboptimal play. And even if the person who tries that is penalized by ultimately losing the game, the opponent that wins that game will have won because their opponent played sub-optimally, not because they themselves played optimally. The optimum is exactly that: a unique, optimal strategy in a given number of turns. Anything that favors a
distortion of that optimal play is by construction
not measuring skill but something else entirely. That is the fundamental flaw in your approach: we cannot infer more from a player's victory than the fact that they won. If they played perfectly, then they could only have won in the number of turns it took them to win. Encouraging them to deviate from that path and move faster because of a silly tiebreaker is not what I would call a measurement of good skill. You're trying to extract information from the victory that
has to be meaningless, and therefore you're going to fail. Your argument
only makes sense in a world where the optimal solution is the one that always takes the fewest number of turns, but that is not guaranteed to be true in any real Conquer Club situation.
Your problem here, the reason you're making poor arguments, is that you're forgetting that optimization under uncertainty is substantially different than optimization without uncertainty. The nature of the optimum is fundamentally different with uncertainties attached. With uncertainties attached, there is a spectrum of possible winning strategies, and if we like we can arrange this spectrum in terms of aggressiveness or time to victory. The perfect logician always picks the one with the highest likelihood of winning, which will in general be somewhere between the extremes. Your tiebreaker distorts this by providing an incentive to select a winning strategy that is slightly faster/more aggressive but slightly less likely to win in absolute terms. We should not be encouraging suboptimal play.
I also provided concrete examples of game types where time to victory provides a clear negative proxy for skill to help illustrate this point. Some of them apply directly to our AutoTournaments. For example, consider Round Limits. In the Conquer Cup VIII games I recently played, there were 30 round limits on the games. If we get to round 25 of the game, our perfect logician might realize that optimal play is to turtle up and just win on troop count, rather than risk a faster but less certain victory through combat. In an escalating spoils game, it might be possible to win faster by taking a risk on eliminating someone early, but conventional wisdom is that you should wait until the spoils are just large enough to offset the cost in troops you will pay by trying to take someone out. Certain maps (like Das Schloss) might favor taking many turns of simple stacking before busting out.
I cannot emphasize strongly enough, in case it is still not clear, that these problems are not
noise that perturbs your
signal. They are fundamental problems with the nature of the tiebreaker itself.
The tiebreaker does not measure what you think it measures.
For all of the above reasons, I therefore submit that this tiebreaker actively does harm to our tournament structure, and encourage Conquer Club to use a more conventional system used by any number of professional tournaments worldwide.
Posted: Tue Apr 21, 2015 8:48 am
(I lost my first try then won all my games till the title of the CC4). It's just dumb to create a structure that will create unfair situations for some.
