I signed off last week's post on A Database for ICCF Finals promising myself to concentrate on the middlegames and endgames in this unique collection of accumulated chess knowledge. I haven't been able to develop any useful techniques to do this, because the openings keep getting in the way. Let's look at those first.
The chart in the 'Database' post already offers a number of talking points. The first is the relative success of the initial Knight moves. The move 1.Nf3 shows a 59.9% success rate for White, while 1.Nc3 shows 65.9%, almost 2/3. I can't explain the success of 1.Nf3. It leads into the same variations that one expects from the 1.d4/1.c4/1.Nf3 complex, but is apparently more successful than either of the alternatives. Perhaps it's just a statistical anomaly.
The move 1.Nc3 is easier to explain. It was used by a single competitor, Ove Ekebjaerg of Denmark; see The chess games of Ove C Ekebjaerg on Chessgames.com, where the 1.Nc3 connection has also been discovered. He played in the 9th (ended in 1983), 14th (2000), and 16th (2004) ICCF Correspondence Finals, where his best finish was second place in the 14th final, behind Tonu Oim. His overall score with 1.Nc3 in the three events was +8-1=13. The single loss was to Oim in the 9th final, which Oim also won. According to Chesslab.com, the critical line is 1.Nc3 d5 2.e4 dxe4 (or 2...d4) 3.Nce2 e5, when 4.Ng3 is more successful than 4.Nf3. It's also worth noting that when engines go head-to-head without a book, 1.Nc3 is a top candidate move; see Chess960 Opening Theory for more.
Back to the 'Database' post, I was also intrigued by the responses to 1.e4. The worst response, at least according to the ICCF statistics, is 1...e6, with a 63.1% success rate for White. The best response is 1...g6, with 51.7% for White. I would not have been surprised to see these results reversed. The stats for 1...e6 come largely from the Winawer (1.e4 e6 2.d4 d5 3.Nc3 Bb4 4.e5 c5 5.a3 Bxc3+ 6.bxc3), where 27 games brought 74.0% success to White. The move 3...Nf6 was also in White's favor, with 24 games going 66.6% for White.
The number of games starting 1...g6 is probably too small to draw any conclusions. After 2.d4, the number increases with the transposition 1.d4 g6 2.e4 coming into consideration. After 2...Bg7, the stats start to skew in White's favor.
Finally, what about the 631 games starting 1.d4 Nf6? Which move, for example, shows better results for Black -- the King's Indian or the Gruenfeld? Here are the stats from the ICCF finals for the sequence 1.d4 Nf6 2.c4, followed by 2...g6 3.Nc3 (the differences in frequency from one table to the next are due to transpositions from 1.c4, etc.).
After 2.c4, the move 2...e6 brings somewhat better results than 2...g6. Similarly, after 3.Nc3 the move 3...d5 works a little better than 3...Bg7. This is not good news for players of the King's Indian, like me, but I'm sure we'll continue to play it. Statistics are, after all, only 50% of the story.