The "Regan Method" Suggests Sandu is Playing Hones

[Sean Evans]

The "Regan Method" Suggests Sandu is Playing Honestly, But Some Suspect "Intelligent Cheating"

Friday, 29.05.2015 11:45

Associate Professor of the University of Buffalo, and a member of the joint FIDE / ACP anti-cheating commission, Kenneth Regan, gave his assessment of Mihaela Sandu. You will recall that fifteen participants of the European Championship have openly accused the Roumanian player of cheating.

Regan, whose specialization is computer science and computing systems, is the author of a special program, with which games can be reviewed for possible cheating. The method of validation involves many factors, details of which are unknown to the general public.

On the community page on Facebook "No more cheating in chess", the scientist commented after the sixth round of the championship. When asked what percentage of Sandu's moves coincide with those recommended by the computer, Regan said: "It is absolutely normal."

However, not all GMs who have looked at the Roumanian's games, agree. For example, in the comments on the same page, Peter Heine Nielsen confirmed the assessment of Reagan: "It is obvious that she is innocent," but on our website Michal Krasenkow expressed doubts: "I know many wonderful Roumanian chess players of the same generation: Peptan, Kozma and others, who have always been dangerous rivals to our Polish girls. But I have never heard of Sandu. Even in adulthood, this player had a rating of 2100+. Of course, players develop at different paces, and anything can happen, but the principle of "trust but verify", especially as noted by David [Navara], in our times, is quite logical. The games of Sandu are without the gross errors typical, alas, of women's chess, but contain brilliant ideas, unfeasible for ordinary mortals - see the very convincing (to me) post by Dmitry Kryavkin on Facebook (according Kryakvin, in the game Sandu - Stefanova White "played brilliantly" - CN). It is very similar to "intelligent cheating" (when the assistant does not automatically select the computer's first line, and looks for every opportuny to look "human") - in any case, it must be checked."

It is this version - "intelligent cheating" - which those who suspect foul play by Sandu are basing their case on, and contrast it with the "Regan method."

[michiguel]

The "Regan Method" Suggests Sandu is Playing Honestly, But Some Suspect "Intelligent Cheating"

Friday, 29.05.2015 11:45

Associate Professor of the University of Buffalo, and a member of the joint FIDE / ACP anti-cheating commission, Kenneth Regan, gave his assessment of Mihaela Sandu. You will recall that fifteen participants of the European Championship have openly accused the Roumanian player of cheating.

Regan, whose specialization is computer science and computing systems, is the author of a special program, with which games can be reviewed for possible cheating. The method of validation involves many factors, details of which are unknown to the general public.

On the community page on Facebook "No more cheating in chess", the scientist commented after the sixth round of the championship. When asked what percentage of Sandu's moves coincide with those recommended by the computer, Regan said: "It is absolutely normal."

However, not all GMs who have looked at the Roumanian's games, agree. For example, in the comments on the same page, Peter Heine Nielsen confirmed the assessment of Reagan: "It is obvious that she is innocent," but on our website Michal Krasenkow expressed doubts: "I know many wonderful Roumanian chess players of the same generation: Peptan, Kozma and others, who have always been dangerous rivals to our Polish girls. But I have never heard of Sandu. Even in adulthood, this player had a rating of 2100+. Of course, players develop at different paces, and anything can happen, but the principle of "trust but verify", especially as noted by David [Navara], in our times, is quite logical. The games of Sandu are without the gross errors typical, alas, of women's chess, but contain brilliant ideas, unfeasible for ordinary mortals - see the very convincing (to me) post by Dmitry Kryavkin on Facebook (according Kryakvin, in the game Sandu - Stefanova White "played brilliantly" - CN). It is very similar to "intelligent cheating" (when the assistant does not automatically select the computer's first line, and looks for every opportuny to look "human") - in any case, it must be checked."

It is this version - "intelligent cheating" - which those who suspect foul play by Sandu are basing their case on, and contrast it with the "Regan method."

I just checked. Sandu is a 2300 WGM who kept that rating since 2008. The only suspicion is that she beat two WGM ~150 elo stronger? Ridiculous.

If the bar for accusation is gong to be so low, you can be accusing 100 players a month.

Miguel

[Ferdy]

A quick run of move error evaluator.
Her first 7 games including that of her opponents. She has blunders too but 3 of her opponents had blunders on 3 or more pawns.

Code: Select all

Move Error Evaluator v9.2

Moves Error calculation based from Stockfish 6 64 POPCNT and players from
Event: ewcc-2015, after round 7.2

Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz
Engine id name Stockfish 6 64 POPCNT
Hash 128 MB, Threads 1, MoveTime 0.10 sec/pos
Engine error margin: 10 centipawn

 Rank                       Players  AveDepth  TotalPos  AveError  DECnt  E300CP  E200CP  E100CP  E50CP  E20CP  EECnt   EERate    W    L    D  Games  Score  SRate

    1            Batsiashvili, Nino        12        54     13.83      4       0       0       2      3      4      4    7.41%    1    0    0      1    1.0  100.0%
    2                Sandu, Mihaela        12       275     25.51     35       0       2       5     17     24     15    5.45%    6    1    0      7    6.0   85.7%
    3       Goryachkina, Aleksandra        11        27     36.89      7       1       0       0      0      5      4   14.81%    0    1    0      1    0.0    0.0%
    4           Javakhishvili, Lela        11        23     44.48      1       0       1       1      0      2      1    4.35%    0    1    0      1    0.0    0.0%
    5          Stefanova, Antoaneta        12        16     53.75      1       1       0       0      1      0      2   12.50%    0    1    0      1    0.0    0.0%
    6                   Girya, Olga        12        49     57.39      9       0       0       2      5      4      3    6.12%    0    1    0      1    0.0    0.0%
    7              Mammadova, Aysel        12        25     62.16      6       1       0       2      0      4      0    0.00%    0    1    0      1    0.0    0.0%
    8                 Melia, Salome        12        45     63.76     12       0       0       0      4      9      3    6.67%    0    1    0      1    0.0    0.0%


Legend:
Rank     - ranking based on low average error
AveDepth - average depth reached by the engine in analyzing those positions
AveError - average difference between the engine and player score, in centipawn (1p = 100)
DECnt    - count of positions where position score is below zero, as a result of move error.
         - This is called Dangerous Error Counts
E300CP   - count of positions where player has blundered by 3 or more pawns
E200CP   - count of positions where player has blundered by 2 to 2.99 pawns
E100CP   - count of positions where player has blundered by 1 to 1.99 pawns
E50CP    - count of positions where player has blundered by 0.5 to 0.99 pawns
E20CP    - count of positions where player has blundered by 0.2 to 0.49 pawns
EECnt    - count of positions where player move score is better than engine move score
EERate   - 100*EECnt/TotalPos or the percentage of the Engine Error Counts

Notes:
1. Analysis of positions start at move 12.
2. If best score of engine is already above 3 pawns and the score of player move
   is also above 3 pawns then error for this position is set to zero. The more cases
   like this only proves that that player is able to maintain its advantage and this
   will minimize its average error.
   On the other hand if the best score of engine is already below minus 3 pawns, no further
   analysis will be done, and this position will not be included in error calculation.
3. When a player move is a blunder that brings down the score below minus 3 pawns,
   error for this position is set to 6 pawns in the calculation. The reason being once
   the opponent found the refutation, the succeeding positions will no longer be evaluated.
   It is just fair to penalize this heavily to also reflect on the average error which
   should be high for players that have more game loses or fatal blunders.
4. move_error = engine_score - player_score.
   engine_score = score returned by the engine after searching for a given amount of time.
   player_score = score returned by the engine after searching the player move for a given amount of time.
   AveError = sum of [move_error] / totalPos.
5. Error is increased if the score of position after player move has dropped to below zero.
   6x error if (score <= -200 cp), 5x error if (score <= -150 cp), 4x error if (score <= -100 cp),
   3x error if (score <= -50 cp), and 2x error if (score < 0 cp).
   The idea is only those errors which will result to inferior position (below zero score) should be penalized more.
   If the increased error penalty is more than 3 pawns, error is set to 3 pawns.
6. Regarding EECnt or Engine error counts, this is a counter check to engine where the human move
   is sometimes better than the engine move, this is because the analysis time given to the engine is
   shorter and also perhaps the position is complicated that bigger depths are required to determine
   correct assessment of the position.
7. DECnt or Dangerous Error Counts is an info to see which player is probably always in trouble.
   Strong players usually have low count of this.
8. Engine error margin is the minimum difference between the player score and the engine score in centipawn
   where the error of the engine is not counted. Engine error happens when the move of the player is better
   than the move suggested by the engine in a given position and analysis time. If this value is high say
   1/2 of a pawn or 50 cp, then there could be less error from engine that will be recorded.

Comparison with other tournaments and championships. Here also we can compare the average error of the event. This is sorted by individual average error.

Code: Select all

Players                     Event                             TotalPos  AveDepth  AveErrorCP  EventAveErrorCP  Score     SRate 
--------------------------  --------------------------------  --------  --------  ----------  ---------------  --------  ------
HYDRA                       Adams-Hydra,London2005            194       11        6.21        12.68             5.5/6    91.67%
Carlsen, Magnus             GashimovMemorial,Shamkir2015      280       12        9.54        15.18             7.0/9    77.78%
So, Wesley                  GashimovMemorial,Shamkir2015      280       11        11.59       15.18             5.0/9    55.56%
Anand, Viswanathan          GashimovMemorial,Shamkir2015      274       12        11.66       15.18             6.0/9    66.67%
Mamedov, Rauf               GashimovMemorial,Shamkir2015      255       11        12.10       15.18             3.5/9    38.89%
Batsiashvili, Nino          ewcc-2015                         54        12        13.83       44.72             1.0/1    100.00
Caruana, Fabiano            GashimovMemorial,Shamkir2015      321       12        15.13       15.18             5.0/9    55.56%
Giri, Anish                 GashimovMemorial,Shamkir2015      296       11        15.64       15.18             3.5/9    38.89%
Mamedyarov, Shakhriyar      GashimovMemorial,Shamkir2015      300       11        15.66       15.18             4.0/9    44.44%
Vachier-Lagrave, Maxime     GashimovMemorial,Shamkir2015      304       12        16.60       15.18             3.5/9    38.89%
Adams, Michael              Adams-Hydra,London2005            191       11        19.16       12.68             0.5/6    8.33% 
Adams, Michael              GashimovMemorial,Shamkir2015      250       12        20.94       15.18             3.5/9    38.89%
Kramnik, Vladimir           GashimovMemorial,Shamkir2015      336       11        22.95       15.18             4.0/9    44.44%
Steinitz, William           WorldChampionship3rd,NewYork1890  495       12        25.26       28.44            10.5/19   55.26%
Sandu, Mihaela              ewcc-2015                         275       12        25.51       44.72             6.0/7    85.71%
Steinitz, William           WorldChampionship1st,USA1886      608       11        25.88       30.34            12.5/20   62.50%
Gunsberg, Isidor            WorldChampionship3rd,NewYork1890  501       11        31.61       28.44             8.5/19   44.74%
Steinitz, William           WorldChampionship2nd,Havana1889   535       11        33.03       36.55            10.5/17   61.76%
Zukertort, Johannes Herman  WorldChampionship1st,USA1886      572       11        34.79       30.34             7.5/20   37.50%
Goryachkina, Aleksandra     ewcc-2015                         27        11        36.89       44.72             0.0/1    0.00% 
Chigorin, Mikhail           WorldChampionship2nd,Havana1889   519       11        40.08       36.55             6.5/17   38.24%
Javakhishvili, Lela         ewcc-2015                         23        11        44.48       44.72             0.0/1    0.00% 
Stefanova, Antoaneta        ewcc-2015                         16        12        53.75       44.72             0.0/1    0.00% 
Girya, Olga                 ewcc-2015                         49        12        57.39       44.72             0.0/1    0.00% 
Mammadova, Aysel            ewcc-2015                         25        12        62.16       44.72             0.0/1    0.00% 
Melia, Salome               ewcc-2015                         45        12        63.76       44.72             0.0/1    0.00% 

[CRoberson]

I looked over the games manually to identify specific computer like moves. I saw a few that were borderline. Then I reviewed her loses. The interesting thing was that her playing style seemed to change in the last two games currently displayed on Chessbase news: particularly her value of pawn structures seems to have changed.

It is easy enough to change the personality of many Chess programs, thus using a statistical matching to known programs technique may be insufficient.

[AdminX]

I looked over the games manually to identify specific computer like moves. I saw a few that were borderline. Then I reviewed her loses. The interesting thing was that her playing style seemed to change in the last two games currently displayed on Chessbase news: particularly her value of pawn structures seems to have changed.

It is easy enough to change the personality of many Chess programs, thus using a statistical matching to known programs technique may be insufficient.

"We have quickly scanned the games of Mihaela Sandu and reached the following preliminary conclusion: it would seem that the Romanian WGM was not using computer assistance. Naturally there is a theoretical possibility that players may be getting very sophisticated forms of help, which may conceal the deceit from cursory investigation. But that seems unlikely in this case. However, our readers can judge for themselves, using the following PGN file – and perhaps the computer correlation function in ChessBase 13 (highlight the games in the games list, right click and use "Edit – Let's Check Engine Correlation")."

I did not even realize I could do this in Chessbase 13.

[Stan Arts]

I always think, some random titled player take this guy/girl to the back and have a blitz session. Any chessplayer can get a good feel for another players abilities in less than half a game. Very easy to tell if someone is legit <whatever rating> or miles off.

[Sean Evans]

Support for Mihaela Sandu

[syzygy]

I just checked. Sandu is a 2300 WGM who kept that rating since 2008. The only suspicion is that she beat two WGM ~150 elo stronger? Ridiculous.

If the bar for accusation is gong to be so low, you can be accusing 100 players a month.

Exactly, statistically there will always be players that for some period of time perform better than "statistically" predicted.

Some years ago in the Netherlands a nurse working at the intensive care unit was sentenced to life in prison for being a statistical outlier (in terms of number of deaths on her watch). After many years and campaigning by people with a brain the courts finally realised their error and she was released.

What is the chance of winning the lottery? One in a million? That's impossibly low! So let's arrest the lottery winner for cheating.

[Laskos]

I looked over the games manually to identify specific computer like moves. I saw a few that were borderline. Then I reviewed her loses. The interesting thing was that her playing style seemed to change in the last two games currently displayed on Chessbase news: particularly her value of pawn structures seems to have changed.

It is easy enough to change the personality of many Chess programs, thus using a statistical matching to known programs technique may be insufficient.

"We have quickly scanned the games of Mihaela Sandu and reached the following preliminary conclusion: it would seem that the Romanian WGM was not using computer assistance. Naturally there is a theoretical possibility that players may be getting very sophisticated forms of help, which may conceal the deceit from cursory investigation. But that seems unlikely in this case. However, our readers can judge for themselves, using the following PGN file – and perhaps the computer correlation function in ChessBase 13 (highlight the games in the games list, right click and use "Edit – Let's Check Engine Correlation")."

I did not even realize I could do this in Chessbase 13.

I used that feature. I am not sure how it works, engine to what depth, I suspect these are ponder hits.

Tourney results:

Code: Select all

Rk.	SNo	Title	Name	FED	Rtg	Pts.	 TB1 	 TB2 	 TB3 
1	7	IM	Batsiashvili Nino	GEO	2473	8.0	2416	46.0	49.0
2	15	GM	Zhukova Natalia	UKR	2456	7.5	2345	41.5	44.5
3	30	IM	Guseva Marina	RUS	2391	6.5	2415	44.5	48.5
4	14	GM	Danielian Elina	ARM	2458	6.5	2413	44.5	48.0
5	4	WGM	Girya Olga	RUS	2479	6.5	2405	47.0	51.5
6	16	IM	Melia Salome	GEO	2452	6.5	2333	43.5	46.5
7	45	WGM	Sandu Mihaela	ROU	2300	6.0	2472	50.5	53.0     --------------
8	20	IM	Kashlinskaya Alina	RUS	2423	6.0	2434	47.0	50.5
9	25	IM	Bodnaruk Anastasia	RUS	2402	6.0	2385	44.0	47.5
10	11	GM	Socko Monika	POL	2467	6.0	2354	43.5	48.0
11	29	IM	Gaponenko Inna	UKR	2391	6.0	2333	40.5	43.0
12	6	WGM	Goryachkina Aleksandra	RUS	2474	6.0	2332	40.5	44.5
13	8	IM	Paehtz Elisabeth	GER	2473	6.0	2332	38.5	42.0
14	13	IM	Mkrtchian Lilit	ARM	2459	6.0	2314	40.0	43.5
15	19	WGM	Zawadzka Jolanta	POL	2434	6.0	2266	37.0	40.0

"Engine Correlation" (probably ponder hits):

Code: Select all

Zhukova,N: Corr= 61% from 9 games
Socko,M: Corr= 61% from 9 games
Batsiashvili,N: Corr= 60% from 9 games
Stefanova,A: Corr= 58% from 9 games
Sandu,M: Corr= 57% from 9 games    --------------
Melia,S: Corr= 54% from 9 games
Danielian,E: Corr= 54% from 9 games
Hoang,T: Corr= 54% from 9 games
Goryachkina,A: Corr= 50% from 9 games
Paehtz,E: Corr= 49% from 9 games
Savina,A: Corr= 49% from 9 games
Arabidze,M: Corr= 48% from 9 games
Ziaziulkina,N: Corr= 47% from 9 games
Khurtsidze,N: Corr= 47% from 9 games
Mkrtchian,L: Corr= 46% from 9 games

By this analysis it seems nothing special about the questioned woman chessplayer.