Rank went down, but I won?

LOL right - too many ugly moves!

Actually, the graph ends at the 21st (where I lost a game). Maybe it isn’t current with the win today. Usually the rank and graph update right away. I’ve never noticed a delay.

Nah. You can select a period of time to zoom in… Mar 22 is clearly there, but since you won one and lost one, the net effect seems to be -10 pts.

It would be cool to have a “rating change” column in the game history. Then we could track change in time much better.

2 Likes

It is likely because the rating of one of your previous 15 opponents went down since your last game. I believe the rating is calculated off the last 15 games and is recalculated after each game using every players’ updated rating … or something like that.

In any case this happens, certainly no mistake.

2 Likes

That’s a great idea, in theory. As Stephen pointed out, this number would not actually be constant: it varies with your opopnents’ recent results. For this reason, I think it would be too expensive to provide: it’d have to be recalculated too often.

One way to help answer this question is now on the pie chart (cropped from your screenshot). If you mouse over the rating point in question on the graph, the pie chart shows how many results are in that dot, and of what type. So at least it’s clear how many results are in a point.

Not quite. Suppose you go from “2000 > 2010”, that’s what the game record will say. Suppose the calculation is updated in between games, the next game will say “2012 > 2020”. It’s entirely consistent.

At what time are you referring to this change?

The rank change you get from a game changes over time, after the game finishes.

I’m not sure how to express it more exactly than that.

T=Time, R=Rating, R1=Rating timestamp game x begins, R2=Rating timestamp game x ends, R3 Rating timestamp game y begins, R4= Rating timestamp game y ends

T0: R0=2000
T1: R1=2000
T2: R2=2010
T3: R3=2012
T4: R4=2020

This is not the case. R1 can be different to R0 due to the rank of the opponent of the first game changing after T1.

Or that’s how I understand it.

In otherwords, I am pretty sure that it is the case that your rank can change after the end of your last game, without you playing any more.

Therefore, the delta ascribed to each game changes over time.

Okay, let’s try this again.

Account starts: rating 1500.
Game starts: rating 1500.
Game ends: rating 1600. [1500>1600] record saved.
[opponent’s rating stays the same]
Game starts: rating 1600.
Game ends: rating 1700. [1600>1700] record saved.
[opponents ratings change]
Game starts: rating 1750.
Game ends: rating 1850. [1750>1850] record saved.

It doesn’t matter, because we simply record the rating at the beginning and end of each game.

I see what you mean now. It wasn’t obvious because it never would have occurred to me that this is a good idea :slight_smile:

To paraphrase, you are saying “lets record the effect on ranking of each game, at the time that the game ended (acknowledging that this is not the actual rank effect of that game, long term, so the sum of the deltas that we record will not add up the overall delta of the person’s rank)”

Well, as I said it wouldn’t have occurred to me, due to the partial nature of it, but I agree it seems eminently feasible, and possible useful.

The downside would be the stream of questions “why did my rank change differently on the graph, on this day, to what this table says happened on that day?”.

So you think people asking questions is a bad thing? People trying to understand the system? I think it’s a good thing. If there were a “stream of questions” (I assume you mean the same questions over and over), which I doubt, OGS could finally get around to sport an FAQ. Other sites solve this by adding a ? symbol you can hover over to get useful explanations on the spot.

The goal of any system is to be as self-explanatory as possible.

Of course “asking questions” - in general - is fine. And I in no way implied it was not fibe

However, designing a feature in a way that is inherently going to generate questions is not such a great idea.

You’re right that it can possibly be solved in this instance by a hover or similar.

Or we could come up with a solution that is internally consistent and not needing explanation :slight_smile:

Come on, you’re just grasping for straws now. “The goal of any system” is not to be self-explanatory, what are you on about? Red herring?

As for internal consistency, I can make up similarly bogus arguments by saying “It’s nonsense to have a pie chart of ‘won/lost games vs weaker/stronger opponents’ since whether they’re weaker or stronger can change based on rating fluctuation. Since it’s not internally consistent it’s better to not have a pie chart at all”.

We are clearly having trouble communicating. I assure you I am not throwing out red herrings.

I was addressing your question:

Generally, asking questions is a good thing.

However, web designs that cause people to have to ask questions are a bad thing.

It is “web application design 101” topic that your web application (system) should be self explanatory.

If you have to adorn all your features with little pop-ups to explain them, you need to ask yourself if you did a poor job of this.

The better designed a web application, the less questions it will generate about “what does this mean?” or “how do I do this?”.

The feature you proposed will generate such question and/or require little popups to explain itself.

Therefore, it seems prudent to keep looking for a cleaner way of achieving the goal.

GaJ

3 Likes

This argument is a furphy. The pie chart remains consistent with the fluctuating rankings. The page remains internally consistent with the pie chart on it, unlike the delta table that you proposed.

Also, I didn’t say “it’s nonsense” to have the delta rank values for the game.

What I did say is that they will end up inconsistent with the actual rank deltas visible in the graph, which will generate questions.

GaJ

1 Like

You mean it will generate questions like

Why would my rank decrease if I just won a game? My aggregate rank was 1587 and dropped to 1577 when an opponent resigned. I’m confused. Why would this be?

It’s quite hard to have a civilized discussion with a person who’s every response is sarcastic.

I think what you were trying to say is:

“We already have a situation where the display of rank information generates questions - it’s genuinely hard to understand the behaviour of the graph from the game history. So why are you objecting to my suggestion?”.

It is clear that the current situation generates questions and we’d like a solution to that.

We’d also like a solution that doesn’t just shift the questions.

As I said, in regards to a deltas column in the games table:

… but the downside is that it would generate questions itself.

I’d actually like to see the graph plot rank by game rather than by date. That would solve the whole problem, and also let people see more clearly how they are progressing each game they play.

Is there some reason why this couldn’t or shouldn’t be done?

…Why not just create the graph and +/- values based on people’s current rankings? If an opponent’s rating changes, and the page is reloaded, just recompute everything and it’ll be consistent with everything else.

When smurph proposed the deltas table I was worried about the recompute load to keep it up to date.

But actually, I just realised that there must be a process every “so often” (is it 15 days) where a big recompute is done, and that’s where the ranks change. So it would be natural to update the delta table at that time - basically for free - and we all could live happily ever after!

@anoek, is this sound, in principle?

I was thinking you’d just compute all of the deltas whenever a user requested the table. Grab the ratings of everyone that fits on the table, which is already done, and retroactively compute the deltas based on the user’s current rating. All of the computation could be done on demand on the client, with the server just returning a column of opponent Glicko ratings in addition to the user names, game results, and user rankings that it already returns.

Alternately, just include the ranking adjustments in the deltas. That way, someone can look at the table, see that they won a game, but their ranking went down, and understand that it happened because the rankings were adjusted. The ranking graph doesn’t appear to update historic ratings based on rank changes for other players, so if the goal is to match the deltas to the ranking graph, including the adjustments might be a good idea.

1 Like