Is this Fair Play? (It's not cheating, but... is it ethical?)

It's simple math that if someone keeps winning all their games their rating will continue to go up. It has nothing to do with the rating system, even if you earn 1 point each win and win 100/100, you'll go up 100 points. You act like the Glicko system can prevent this, and it absolutely can't. All it can do is create a ceiling where at some point a player much higher rated than their opponent can't earn points anymore. For example, on Lichess.org, If I play someone low enough, I get zero points if I win. However as long as I can get 1 point, I could harvest points against that player over and over again, and the rating system can't prevent that. You really need to understand the point being made here, which is abusing the win to loss ratio artificially to an extreme extent. I agree it can't go on forever, as there is a ceiling effect, but that doesn't mean it can't be abused to artificially manipulate your rating several hundred points possibly. If you disagree, please provide the mathematical or other evidence to support your position. I'm familiar with rating systems and their mathematical functions.

-Jordan

The formula is independent of the setting. It doesn't matter if it's OTB or online. The more common idea is that you can inflate your rating by having only higher rated opponents. Both ideas are false, and for the same reason. This has all been discussed exhaustively in other threads, one of them very recent.

I'll do that. 

From a recent thread, where the OP insisted that having higher rated opponents would inflate your rating, it was put to the test, using the actual Glicko formula.

"A friend of mine has recently run up his chess.com rating by aborting his games against higher-rated (or equal) players and fishing for games against lower-rated opponents."

 "the idea of trying to artificially get a higher rating"

It isn't an ethical question since it doesn't affect anyone's rating. You can choose the minimum and maximum rating of your opponents. If you prefer lower rated opponents or higher rated opponents, go for it. Your opponents will have their own choices so you don't have to worry that they won't want to play you because of your rating. No one is harmed... but you might find yourself bored with too many easy wins or, against higher rated players, frustrated by too many losses.

I'm not sure what the criteria to generate this data is being used as it just gives opponents average rating to change in "my rating" to highest rating change, and the blog link isn't working for me there to try and extrapolate more information for this table to better understand it. However, I suspect this doesn't seem to account for what we're talking about in this thread, which is artificially manipulating the win to loss ratio to an extreme extent. Show me how the rating system prevents a win to loss ratio abuse - I play someone 250 points or lower than me and as long as I win and gain a point, I raise my rating artificially high over enough games played and wins had. We're not talking a 60-75% win ratio, but something extreme- 90-100%. In my opinion and understanding, rating systems can only deter (lower the reward of winning/risk or effort given) and cap (eventually getting no reward for winning/risk or effort given) a win to loss ratio abuse as I've already described, but it can't prevent it to a smaller degree, much less altogether. Only actual performances of the players themselves can, and then we can go into the statistics of that probability, which would then be a discussion of theoretical and practical, and I'll say upfront I am discussing this from a theoretical point of view, not a practical/probable one, as we're operating under the premise you made that the Glicko rating system makes it so that playing lower rated players only won't increase your rating any more or less than playing other rating pools, seemingly rejecting the idea of abuse of the win to loss ratio as we've been discussing. Even if the odds are 1:1000 this occurs, their exists the possibility of this abuse, therefore it refutes the idea that the rating system doesn't allow for rating manipulation as we've described.

-Jordan

If you get 1 point per win, Glicko has calculated that you should win 15 games out of 16... and that's likely what will happen... and your 15 points gained will be lost by that one game you happen to lose. If you actually win 100 straight, your rating will be 100 points higher, sure, but it will be proof either that your game has improved or that your previous rating was too low for some reason. You can game the system by cheating, but not by choosing the rating of your opponents.

mine would be a bit... inaccurate by that logic

This just shows you're not understanding the discussion at hand, the probabilities of win to loss ratio by rating difference is NOT what we are discussing here, we are discussing win to loss ratio abuse and how Glicko can't account for it. If I beat someone known to be far less skilled at chess than I am, I am not proving that I am getting better over time by beating them again and again, I am simply proving I am better than them and demonstrating it over and over, giving me 1 or more points each time. If I play 100 games against that player and beat them all 100 times and my rating goes up from say 1500 to 1650, I did not improve 150 points in chess skill, I manipulated my rating by beating someone I know I can beat 100 times and gaining points for it each time I won. That is an abuse of the win to loss ratio we'd anticipate players having, and it'd create an artificial rating for the player abusing it. 

-Jordan

Good point. however, you are assuming they would win every time.

the person the OP was talking about won 22 games in a row as the highest streak.

assuming that they get 1 elo per win and 15 elo a loss then they gained only 5 elo at their highest streak.

the thing is, they wont win every time and there are people that LOVE to haggle for a draw to be annoying.

DefaultedwasTaken, "win every time" is not much of an assumption here:

(names of players omitted purposely):

Depending on his rating, the accuracy might be as suspicious as the 47 out of 50 wins.

Or perhaps this is expected; that if one plays at a 1650 level strength, these high accuracy games and win:loss ratio are expected against lower rated opponents? I am open to that possibility too, since I have no data in front of me on how out-of-the-ordinary this is. 18 and 22 game winning streaks seem rare to be sure--especially back to back--but if you sat me down with a 22 random players in a row rated 900 or lower, I am pretty sure (as a 1200ish player) I'd run up numbers like that in longer time controls too.

ELO calculates the odds of what is expected. Doing better than expected will get you an increase, worse, a decrease. If you win more than expected, you deserve the increase. Your friend's 47-3 and the accuracy scores are suspicious to me... but I did win 16 straight not long ago. I don't know.

28 moves with 98.4% accuracy, 21 with 99.7%  ??? Master-level or computer-aided play.

maybe he's a 2000 level player who sandbagged?

I think their argument is based on a misunderstanding regarding the premise.

Let's say you enter a 10 player round-robin tournament, where you play nine opponents.  Under the Elo system, if you score 6/9 points, it does not matter who you actually won, lost or drew, as long as all players are within 400 points of you.  Your rating will be the same whether you beat the top three players and drew the rest, or beat the three worst players and drew the rest.  The system is premised on the fact that it all evens out.

That I think is a pretty good explanation for why "opponent farming" doesn't work out - AS LONG AS you friend's rating is equivalent to his skill.

No, your friend is cheating.  If all his opponents were, say, 1000 points lower than him in terms of actual playing strength, then yeah, 22 wins in a row would not be a surprise at all.  But the accuracy numbers would still be lower than that.

A seven move game where he checkmates the guy, yeah, his accuracy will likely be close to 100%.  But when it's 25 moves, 30 moves...?  Pretty unlikely.