In online chess, our opponent's rating changes during the game. (Usually, for most of us simultaneously play several games, which finish at different points of time.)

When I finish a game against a certain opponent,
which rating of his/hers counts for the calculation of my new rating -
a) the one at the beginning of our game? //I very hope so!
OR
b) the one at the end of our game? //completely idiotic approach

The latter (b) would be completely idiotic, since for example in the case of over-the-board team championships, which embrace a period of several months, thus several FIDE lists (jan 1, apr 1, etc) with changing ratings for the players, it is only the rating valid at the start of the championship that counts. (And this does have logic, while b) has no logic at all, it is only more comfortable for the programmers...)

If b) is true, then I won't resign games even with a piece down!!! I will wait until my opponent reaches, say, 2400. Only then will I give up, so that I will lose less points...

Please assure me that a) is true of the rating system here!!!

As I understand it, the rating used is the rating after deductions for the loss, as chess.com feel this is the more accurate rating.

@jaf299:
The question was not that. :)
Your opponent's rating is what has an impact on your rating change after a game.
But which rating? The one he had when you started the game with him, or the one he happens to have at the time of finishing (reflecting the results of other games of his that were finished while you were playing with him)?

I believe it is the one he has at the time your game with him ends. So if he has finished another couple of games from the time you started playing him, the results of these games are reflected in the rating used to adjust the ratings at the end of your game with him / her. This is just my understanding of how the rating system works, made possible by the fact that all games are monitored by a server that can adjust ratings on the fly. If anyone has a different take on this please let us know.    

 I just looked back at some of my finished games and the rating used does seem to be the rating the opponent had when our game ended, which was different from the rating he started with. 

That would be very very bad... :-(
A paralell example. When you sit down to play an opponent in an OTB game, it is natural that you want to know exactly by how many points your rating will increase/decrease after the game depending ONLY on the result. But with ratings changing "on the fly", that wouldn't be possible...

This is not only my problem, I think.
When I asked about it at another turn-based chess site, the answer was, b) because I am the programmer and I didn't want to bother with it.
I agree, this is a more demanding task for the programmers, but still..

I know what you are getting at, but I can also see the logic in the way ratings are currently calculated, which means that the most up to date grading is used to calculate the result of each game. Of course it does make it almost impossible to calculate how much rating points you will gain or lose from a particular game. I can see the difficulty in the implementation of point a) from a programming point of view, the way it's done at present is much simpler to implement.

If a game goes on for 2-3 months (not at all unbelieveable), what's the most accurate assesment of his playing strength, the one he has currently, or the one he had 2 months ago?

OTB isn't the same thing as correspondance.

Hey, I meant knowing in advance only the number by which the rating is expected to increase/decrease--not the exact new rating. (As is the case with the Élő System used by FIDE. You have multiple own ratings for each tournament (game), based on your rating valid at the time when the tournament (game) commenced. What you can always know is that the resulting change will be the sum of the multiple changes.)
And hey, one can play every game for months, making only one move a week in each game...while all his opponents are playing at a more regular pace. (So, the argument regarding the connection between duration and playing strength might be subjective.)
And hey, I played a number of real correspondence games (via snail mail) for years--a) was applied there, so I guess I know its essence.

Another parable.
Suppose you have 2000 rating points.
Your seek is accepted by a 2600+ player. Wow!! (what a strong opponent, what an opportunity, the game of your life, etc etc)
You start playing the game and it lasts four months. During the game, your opponent loses his points to other opponents, and becomes as low as 1800.
Then it's time to finish the game, since only the two kings survived.
So, your game is a draw (what an achievement vs. a 2600!), for he played his best especially against you, nevertheless you're going to lose rating points for being equal to an 1800-player--instead of a 2600-player...  Now are you satisfied?

If your opponent really dropped fom 2600 to 1800, then his 2600 rating wasn't an accurate representation of his strength anyway. Probably the 1800 is more accurate. Also, they use Glicko instead of Elo system. Glicko takes into account how stable the rating is moving up or down for the purposes of calculation. So if some moves from 2600 to 1800 and back again you would get less points than from some whose rating stays closer to 2600.