jmo1121109 wrote: ↑

Wed May 29, 2019 11:11 pm

Well, I'm definitely seeing some people not understand how the formula works too. RR is based on 2 things.

1. A base % which is a percentage of how many turns you've missed vs how many you've played with nothing else factored in. So it is not accurate that someone who plays a lot of phases and NMR's a couple times is impacted as badly as someone who plays a few. If you have played 2 turns and missed 1 then your base RR is 50%. This part of the RR is designed to be non-brutal and allow you to recover easily by playing more without missed turns.

2. Additional penalties, of which there are 2 types. 1 type is a short term penalty for 1 month's time of 6%. After 28 days this 6% goes away. The second type is a medium length penalty of 1 year for 5%. These are designed to demolish the RR of people who consistently get kicked into CD in their games. Because at the end of the day if you go CD in your game, you're messing that game up for other people and reducing their experience on the site.

Now that said, live games are more prone to them so I'm happy to make a more accommodating penalty format for live games. What I might do is make it so that the penalty for a live game NMR is only 5% and only lingers for 28 days. Then if you get too unreliable multiple times in a month, you're gonna be done with live games for that month, and you can try to be less rude to the other players again next month. But if it's only a couple times then you won't have a problem. If you really want to play in a tournament, they all give about a month's notice for joining so you'd be able to focus on being reliable that month to avoid having an awful RR for the TD.

New England Fire Squad wrote: ↑

Wed May 29, 2019 9:27 pm

Gunboats typically have a phase length of 5 minutes - with zero excused turns this is only a 10 minute delay, and that happening 2-3 times isn't the end of the world. Press games are usually 7 minute phases, and there was a suggestion that was being taken seriously to make 0 excused phases impossible for live - this would mean that any nmr would delay the game by 21 minutes at minimum in a 7 minute phase game; now that happening 2-3 times would suck. The time issue is pretty real for live press games. Gunboat at least I can dick around on the internet while I wait.

Squigs44 wrote: ↑

Wed May 29, 2019 9:41 pm

There are two types of people in this debate
Type 1: Games take too long because other people are unreliable
Type 2: I get penalized too much when I am unreliable

[...]

The only changes that should be made to RR is if you think the rating is inaccurate. The point about the straight 11% being too harsh on long term players is a good point there

Carl Tuckerson wrote: ↑

Wed May 29, 2019 8:55 pm

On the subject of time, the issue is really that people don't Ready orders in my experience. Gunboats do take a while sometimes, but I am very confident the extensions due to NMRs aren't the main culprit. If you're setting the game to allow one excused missed turn per player, then an individual player can only extend the game by at most 10 minutes before being kicked from the game. That doesn't seem like the reason games take an hour or more longer than one would expect (which is roughly my guess as to why they do, from my own experience).

To that end I think the few minutes saved per game from setting players to Ready instead of Save wouldn't do much. And if people are hitting Ready, then it's probably actively good to set them to Save instead of Ready, so that the game won't progress without them having an opportunity to change orders in light of an NMR. Part of the reason for the new system was to mitigate the imbalances created by players leaving abruptly by making sure everyone knew that the player had left or NMRed.

Gunboats do take some time but I think the main culprit is the players and not the new system. Unfortunately I don't see a realistic solution to that on the site's side. I think it is just on us to encourage people to Ready as a policy.

~~~~~

On the subject of the penalty, I have a question--what exactly do we intend different reliability ratings to communicate to people who are creating games? Not to be obtusely deconstructive, but on its own, a 70% RR vs a 30% RR vs a 90% RR doesn't really "mean" much--it tells you that the 90% guy misses orders less often than the 70% guy and the 30% guy, but that's just a comparison between players. What does it "mean" to have a "70%" RR? Is that "reasonably reliable but prone to missing a phase every now and then"? Is that "misses 30% of his orders"?

Chaqa offhandedly addressed this in his OP, asking if missing three phases recently makes him "only 75% reliable." Does it? Is that even what RR is intending to represent? I don't actually know how to answer this question.

I can't even tell you why I do this, but I read RR like someone's grades in school. Someone with a 90+% RR is golden. 75-89% is reliable enough, prone to occasional absences but good enough for most games, and probably the errors are in good faith when they occur. 65-74% is borderline to me and anything beneath 65% is dicey. I have often wondered why people start RR50% or RR60% games because I don't see how that gates unreliable players off at all. RR70% is what I normally set when I bother with the setting.

But there's nothing inherent about RR that makes my set of interpretations correct. It is ultimately an arbitrary bar, and I can't even promise you I'm completely consistent with that set of groupings.

I think that question has to be addressed before we can talk about tuning the per-miss penalty though... how are we to know what the penalty should be without knowing what we are intending the penalty to represent?

~~~

Regarding live vs nonlive games, I think this is another issue that's best handled by the users. In the acknowledgment that live games are more prone to people getting distracted and missing a turn, just set more excused turns, or set a lower RR threshold for playing. It seems to me that the tools already exist and maybe we aren't optimizing them and that's the crux of the issue. I am not sure that live games are so substantively different from nonlive games that having separate penalties or even separate RR systems for each type of game is necessary, and each has an additional problem. With separate penalties, the information that RR communicates becomes muddled; a guy with 90% RR who only plays live games has a different record of missing turns than a guy with 90% RR who only plays nonlive games. And with separate systems, we add a lot of extra work to the devs.

~~~

That is my 2c on the issues so far. I am especially interested in resolving the second one ("what does RR communicate and what do the different values mean?").