GM Kaufman v.Novag Constellation 3.6 QR Odds (Moves 1-40)

[lkaufman]

Well the earliest Hallsworth list i have is his first
it was called the "News Sheet" back then
on News Sheet #1(Sept.1985) Connie 3.6 (released in 1984)is shown at 1881 elo
of course chess computers were always rated higher when they first appeared on any list and tend to go down as more games are played

scanning ahead just a couple of years ... to News sheet #13(Spring 87)she is shown as 1830 Elo and one more jump of two years to News sheet #24 (Sept.89) she clocks in at 1741
this last rating was based upon 803 games
also on #24 Hallsworth show a rating for her against humans of 1861 bases upon only 17 games

the SSDF adjusted downward all of their ratings in the year 2000 by 100 Elo to adjust for ratings inflation
i think i recall they were following an adjustment made by Fide for human players
not sure when/if Hallsworth adjusted down his ratings as well but i think he might have

it should be noted however that the German dedicated chess computer site has ratings for the dedicated computers which are independent of the SSDF and Hallsworth
their current day "Tournament" rating for Connie is 1786 elo(based upon only 20 games)
http://www.schachcomputer.info/html/bod ... liste.html


Regards
Steve

Some comments: I believe that SSDF reduced ratings by 100 points on multiple occasions, certainly not just once. Probably same for Hallsworth, though I'm less sure about his list. These adjustments were not to adjust for inflation as such, because the ratings of any given set of engines were not inflating. Rather they were done because as new, stronger engines came out, they got ratings that were too high relative to human ratings, and it was considered more important to get the ratings of the new engines "right" than to worry about the old ones. There was some debate about the cause of this phenomenon, but for me it was quite clear: engine vs. engine ratings exaggerate rating differences in comparison to engine vs. human ratings, by something like a 4 to 3 ratio. Finally FIDE never adjusted human ratings downward in general, they only lowered the ratings of female players not named Polgar down by 100 points at one time (on the grounds that women players other than Polgars mostly played each other).
So the relevant numbers seem to be Hallsworth 1830 vs. engines and 1861 vs. humans, and SSDF 1786 vs. humans. In those days we used 100 Elo as adjustment for England and 200 for Sweden to convert to USCF, so those numbers would imply 1930 or 1961 or 1986 USCF rating, which average to 1959, close to my 1950 estimate. The Superconnie human rating of 2018 was based on 48 games at 40/2, so allowing for the improvement over Connie 3.6 again the 1950 figure looks right.

[Steve B]

Well the earliest Hallsworth list i have is his first
it was called the "News Sheet" back then
on News Sheet #1(Sept.1985) Connie 3.6 (released in 1984)is shown at 1881 elo
of course chess computers were always rated higher when they first appeared on any list and tend to go down as more games are played

scanning ahead just a couple of years ... to News sheet #13(Spring 87)she is shown as 1830 Elo and one more jump of two years to News sheet #24 (Sept.89) she clocks in at 1741
this last rating was based upon 803 games
also on #24 Hallsworth show a rating for her against humans of 1861 bases upon only 17 games

the SSDF adjusted downward all of their ratings in the year 2000 by 100 Elo to adjust for ratings inflation
i think i recall they were following an adjustment made by Fide for human players
not sure when/if Hallsworth adjusted down his ratings as well but i think he might have

it should be noted however that the German dedicated chess computer site has ratings for the dedicated computers which are independent of the SSDF and Hallsworth
their current day "Tournament" rating for Connie is 1786 elo(based upon only 20 games)
http://www.schachcomputer.info/html/bod ... liste.html


Regards
Steve

Some comments: I believe that SSDF reduced ratings by 100 points on multiple occasions, certainly not just once. Probably same for Hallsworth, though I'm less sure about his list. These adjustments were not to adjust for inflation as such, because the ratings of any given set of engines were not inflating. Rather they were done because as new, stronger engines came out, they got ratings that were too high relative to human ratings, and it was considered more important to get the ratings of the new engines "right" than to worry about the old ones. There was some debate about the cause of this phenomenon, but for me it was quite clear: engine vs. engine ratings exaggerate rating differences in comparison to engine vs. human ratings, by something like a 4 to 3 ratio. Finally FIDE never adjusted human ratings downward in general, they only lowered the ratings of female players not named Polgar down by 100 points at one time (on the grounds that women players other than Polgars mostly played each other).
So the relevant numbers seem to be Hallsworth 1830 vs. engines and 1861 vs. humans, and SSDF 1786 vs. humans. In those days we used 100 Elo as adjustment for England and 200 for Sweden to convert to USCF, so those numbers would imply 1930 or 1961 or 1986 USCF rating, which average to 1959, close to my 1950 estimate. The Superconnie human rating of 2018 was based on 48 games at 40/2, so allowing for the improvement over Connie 3.6 again the 1950 figure looks right.

i guess this is possible
still..i have so many computers rated so much higher then the Connie 3.6 ..that extrapolating from a base of 1950 for the Connie(or 300 Pts more for a conversion of the Hallsworth list now ..to USCF ..)
then many of my high-end dedicated's would be USCF Senior Masters
a quick count would give me something like 35 dedicated's
..just doesn't sound right to me..
also.. i still think we need to add something to your rating given your expertise in dedicated computers and in PC Engines in general
however i wont argue the point
i still think the game is meaningful for my experiment even given a rating difference of only 500 (if we adopt your rating estimate for Connie)
perhaps if Connie can somehow manage to win this game then the experiment still holds some validity given that the extra R was able to defeat the 500+ difference in rating and not just draw with it

even failing that ..the game is enjoyable

Steve

[lkaufman]

[quote="lkaufman"]
"Finally FIDE never adjusted human ratings downward in general, they only lowered the ratings of female players not named Polgar down by 100 points at one time (on the grounds that women players other than Polgars mostly played each other). "

Correction: FIDE RAISED the ratings of female players not named Polgar by 100 points.

[lkaufman]

i guess this is possible
still..i have so many computers rated so much higher then the Connie 3.6 ..that extrapolating from a base of 1950 for the Connie(or 300 Pts more for a conversion of the Hallsworth list now ..to USCF ..)
then many of my high-end dedicated's would be USCF Senior Masters
a quick count would give me something like 35 dedicated's
..just doesn't sound right to me..


Steve

But this is a mistake. If some engine is 500 above Connie on the Hallsworth or SSDF list, you should not expect it to get a USCF rating of 1950 plus 500 = 2450, but rather 1950 plus 3/4 of 500 = 2325. Does this make the high-end ratings seem reasonable to you?

[Steve B]

i guess this is possible
still..i have so many computers rated so much higher then the Connie 3.6 ..that extrapolating from a base of 1950 for the Connie(or 300 Pts more for a conversion of the Hallsworth list now ..to USCF ..)
then many of my high-end dedicated's would be USCF Senior Masters
a quick count would give me something like 35 dedicated's
..just doesn't sound right to me..


Steve

But this is a mistake. If some engine is 500 above Connie on the Hallsworth or SSDF list, you should not expect it to get a USCF rating of 1950 plus 500 = 2450, but rather 1950 plus 3/4 of 500 = 2325. Does this make the high-end ratings seem reasonable to you?

well i didnt do that exactly..
i simply added 300 pts (the difference between Connie's Hallsworth rating of 1646 and your estimate of 1950 USCF) to all dedicated rated at or near 2100 on his list
that gives me about 35 computers rated 2400 + USCF
not counting Clones(same program and hardware specs but in different housings)


for example...

the Kasparov Sparc is shown as 2208 ELO
adding the 300 pts i get 2508 USCF

according to your formula what would be the constant i can add to the computers on his list given a 300 pt spread between Hallsworth's elo rating and your USCF rating for Connie?

Steve

[lkaufman]

for example...

the Kasparov Sparc is shown as 2208 ELO
adding the 300 pts i get 2508 USCF

according to your formula what would be the constant i can add to the computers on his list given a 300 pt spread between Hallsworth's elo rating and your USCF rating for Connie?

Steve

The mistake is adding a constant. If you consider 1650 on his list equal to 1950 USCF rating (vs. humans), you should add 300 but then subtract 25% of the excess over 1650. So 2200 would be 2200 + 300 - 25% of 550 which equals 2362.5. If some engine is rated at 2850 on his list, the 300 added would be fully offset by the 25% reduction to predict a USCF rating of 2850, roughly what I would expect from someone around World number 10.

[Steve B]

for example...

the Kasparov Sparc is shown as 2208 ELO
adding the 300 pts i get 2508 USCF

according to your formula what would be the constant i can add to the computers on his list given a 300 pt spread between Hallsworth's elo rating and your USCF rating for Connie?

Steve

The mistake is adding a constant. If you consider 1650 on his list equal to 1950 USCF rating (vs. humans), you should add 300 but then subtract 25% of the excess over 1650. So 2200 would be 2200 + 300 - 25% of 550 which equals 2362.5. If some engine is rated at 2850 on his list, the 300 added would be fully offset by the 25% reduction to predict a USCF rating of 2850, roughly what I would expect from someone around World number 10.

well we lose roughly 1/2 of the 35 computers on his list if we use 2200 instead of 2100
so about 17-18 computers that would be rated at or close to a Senior Master USCF
Still seems a bit high too high to me
but thats just a Gut feeling
my guess is i dont own but a handful of computers that wouldn't be eaten for lunch by a solid USCF Senior master in a standard(no odds) 40/2 match
although i imagine the Masters book knowledge would probably make a big difference in this case

anyway...
its your move regards


Steve

[lkaufman]

Connie finds a novel way of providing a flight square for her Q with...

22..Bf5

[d] r2r2k1/p1p1nppp/1pn5/4Pb1q/3pR3/Q2B1N1P/P4PPK/2B5 w - - 0 23

Cramped Regards
Steve

I play 23.g4, so we should end up with me four pawns down as well as down by rook and knight for two bishops, which is about another pawn. Bad as it is, I think it's better than just being a rook down with no play. I considered 23.Bg5 Bxe4 24.Bxe4 to force ...f6, but I don't think I'd have enough material left to conduct a successful attack.

[Steve B]

I play 23.g4, so we should end up with me four pawns down as well as down by rook and knight for two bishops, which is about another pawn. Bad as it is, I think it's better than just being a rook down with no play. I considered 23.Bg5 Bxe4 24.Bxe4 to force ...f6, but I don't think I'd have enough material left to conduct a successful attack.

Connie begins the tactical melee with 23..Bxg4

[d] r2r2k1/p1p1nppp/1pn5/4P2q/3pR1b1/Q2B1N1P/P4P1K/2B5 w - - 0 24


She has also authorized me to offer you this conditional move..

IF 24.Rxg4 THEN 24..Nxe5

Multi-Tasking Regards
Steve

[Terry McCracken]

I play 23.g4, so we should end up with me four pawns down as well as down by rook and knight for two bishops, which is about another pawn. Bad as it is, I think it's better than just being a rook down with no play. I considered 23.Bg5 Bxe4 24.Bxe4 to force ...f6, but I don't think I'd have enough material left to conduct a successful attack.

Connie begins the tactical melee with 23..Bxg4

[d] r2r2k1/p1p1nppp/1pn5/4P2q/3pR1b1/Q2B1N1P/P4P1K/2B5 w - - 0 24


She has also authorized me to offer you this conditional move..

IF 24.Rxg4 THEN 24..Nxe5

Multi-Tasking Regards
Steve

We're on track with this line, I posted a day back...23. g4..Bxg4 24. Rxg4..Nxe5 25. Nxe5..Qxe5+ etc.