100% favorite according to gnu and I still lost

[blitzxz]

Maybe somebody could clarify this to me? Okey, I have thought that error totals in MWC can be used to calculate who was favorite in the match and by how much. However, I have got several times these weird results that I was over 100 % favorite and I still lost the match. For example, in this one 5-point match my opponent's error total was -113,348% and mine was -21,718%. So I gained 91,23% in MWC. If I had 50 % chances in the beginning, this will put me over 100 % favorite to win, in other words I was sure winner. And I still lost 5-4. Where is the problem?

(The match is attached.)

[socksey]

I seem to often lose a game with gnu when I have outplayed the bot.     Luck becomes the factor in this case and I was out of luck!   

socksey



No matter how great and destructive your problems may seem now, remember, you've probably only seen the tip of them. - Lithograph from Despair.com

[sorrytigger]

I hesitate to answer when your statements are not very precise. I had a short look at your match and everything looked ok. According to gnu you played very well

Your MWC should vary in a range from 0% to 100% starting at 50%. It should be 50% whenever the score is n-away n-away, meaning that both players have the same number of points to go, meaning at the beginning of the new game, of course. Your and your opponents MWCs should add to 100%.

Looking at the error totals doesn't show your winning chances, but gives a hint, how much chance you (or your opponent) gave away by making mistakes. If you throw away a lot of MWC you need lucky rolls to bring it back.

If you make perfect moves (according to your analysing tool) you lose the least MWC, i.e. 0%. This doesn't mean that your MWC increases by skill. It may increase or decrease, that depends on the resulting position. But you made the best of it by making the best possible move. Making errors means performing worse than possible. All these little errors sum up to horrible amounts, depending on the matchlength.

When your score was 4:4 your MWC was 50%. But it quickly degraded to below 40%. When you had 3 men on the bar it was around 8%, when you lost it was 0%.

[ah_clem]

Maybe somebody could clarify this to me? Okey, I have thought that error totals in MWC can be used to calculate who was favorite in the match and by how much. However, I have got several times these weird results that I was over 100 % favorite and I still lost the match. For example, in this one 5-point match my opponent's error total was -113,348% and mine was -21,718%. So I gained 91,23% in MWC. If I had 50 % chances in the beginning, this will put me over 100 % favorite to win, in other words I was sure winner. And I still lost 5-4. Where is the problem?

You can't just add up a bunch of individual probabilities to arrive at an overall probability.  For each move, the error is the difference in MWC between the optimal move and the actual move.  You can't just sum these to give an overall probability of MWC for the entire match.  It's like adding fractions by adding the numerators and denominators - it makes no mathematical sense.

Why should the sum of the error totals add up to any particular number?  If you and your opponent play perfectly, they'll add up to zero, if both of you play awfully, they may both be well over 100% (or 200% or more if you're really awful).

The difference between the two sums is an indication of who outplayed whom, but just because the difference is greater than N doesn't imply who wins - there's plenty of randomness (luck) due to the dice and this may outweigh a difference in error total. In particular, even if the difference in error totals is greater than 100% it's still possible for the dice to break for the weaker player.

In other words, your opponent gave away an entire match's worth of errors, but the dice saved him.  Happens all the time.

[blitzxz]

You can't just add up a bunch of individual probabilities to arrive at an overall probability.  For each move, the error is the difference in MWC between the optimal move and the actual move.  You can't just sum these to give an overall probability of MWC for the entire match.  It's like adding fractions by adding the numerators and denominators - it makes no mathematical sense.

Why should the sum of the error totals add up to any particular number?  If you and your opponent play perfectly, they'll add up to zero, if both of you play awfully, they may both be well over 100% (or 200% or more if you're really awful).

The difference between the two sums is an indication of who outplayed whom, but just because the difference is greater than N doesn't imply who wins - there's plenty of randomness (luck) due to the dice and this may outweigh a difference in error total. In particular, even if the difference in error totals is greater than 100% it's still possible for the dice to break for the weaker player.

In other words, your opponent gave away an entire match's worth of errors, but the dice saved him.  Happens all the time.

This is what I thought would explain it. Error totals (in MWC) or luck adjusted result can't be used to calculate anything about overall winning chances. Still this is done regularly in different sites and analyzes. It's even mentioned in gnu tutorials. For example Zorba did this normal favorite calculation after forum match 1.