Statistically, is this the perfect play opening?

So, I had an idea. Maybe a stupid one.

I created an opening book from over 9million games and then I used a 350K game database of both players over 2500 elo and the 'learn from a database' Fritz feature and I came up with 100% certainty that 1.d4d5 is the best opening.

After that 2.Nf3 76.% or 2.Bf4 23.8%

2.Nf3 continues to ...Nf6 3.Bg5 while

2.Bf4 continues with ...Nf6 3.e3 76.9% or Nc3 23.1%

I don't know what to think of them. Are they really the best? 2.c4 is a no-no (0%)! Is the Torre Attack the perfect play opening? OR am I misinterpreting Fritz's findings?

@1

See Figure 30 of Acquisition of Chess Knowledge in AlphaZero

1. Ruy Lopez Berlin Variation
2. Giuoco Piano
3. Four Knights
4. Ruy Lopez Morphy Variation
5. Caro-Kann Advance Variation with ...Bf5
6. Nimzovich Defense
7. Petrov
8. Sicilian Defense Najdorf Variation
9. French Defense Classical Variation
10. Caro-Kann Defense Advance Variation c5
11. French Defense Classical Variation, other move order
12. Caro-Kann Defense Classical Variation
13. French Defense Tarrasch Variation
14. Queen's Gambit Declined with Nf3
15. Queen's Gambit Declined with Nc3
16. Queen's Gambit Declined with Nf3, other move order
17. Queen's Gambit Declined Tarrasch Variation
18. Queen's Gambit Declined with Nf3, other move order

Indeed 1. e4 e5 2. Nf3 Nc6 3. Bb5 is what I think are the best moves. I don't know after  that.

But my question remains: why does Fritz come to its odd conclusion (through stats of GM games), is it yet another bug or am I missing something?

1. d4 d5 2. c4 is more than fine , I think you're getting a bit confused on how to use and interpret a DB, Steve Giddins had written a nice book on this topic, how to build your opening repertoire. It's old so could be a bit outdated for today but tbh why bother with DBs to see what's happening on move 2, pick 1-2 model players whose style you like and see what they play.

Also, I don't know Fritz's database capabilities and how good they are, for DB usage chessbase or the new scid are more typically used.

Don't use DBs to find the "best opening", there's no such thing. If you want to get a taste of how play in various openings look like, better to get your hands on annotated games in a variety of openings, e.g. good game collections.

Creating your own tailored repertoire using a combination of books & DBs is for later stages.

Thanks for the reply, but it is not exactly that which is puzzling me. It s a feature in Fritz, after you build an opening book you can ask it to learn from another or the same database. Each time white wins that line's weight increases and if a defeat it is reduced. The same for black. So, the final product-book is not just db stats like an untuned book.

No, I m not trying to build my opening repertoire with that, just wanted to see what is playable.

For example I took the advise of a teacher and started with the Italian game, but I was not getting the lines I wanted since Black wouldnt cooperate, I tried the London, didnt like my bishop locked in front of my pawns, I tried the Scotch didnt like the positions and now I m playing the Vienna. Hardly any of them are main, fancy lines that db stats would propose. I picked them. I ve also definitely settled to the Caro-Kann as black and I still have absolutely no idea what to play against d4, but I doubt anyone really knows .

To see what's playable, instead of the statistical approach on the 2500+ range see what openings super GMs have played over the last decades. This may miss a few sound openings they didn't play due to fashion but almost all their choices will be openings that are sound and have rich positions.

More or less all major openings and all major lines in them are entirely playable.

You've hopped from Italian to London to Scotch to Vienna. Playing some of these on occasion for fun is of course fine but hopping around will ultimately be bad as you won't learn anything in depth.

Pick one and stick to it, through good games and bad games. Every difficulty you face in an opening is not solved by changing openings but by analysing your games, finding flaws in your game and improving in those areas.

The value will come from playing 50, 100, 200x the Italian game, then analysing your games. Then you will know the structures very well, the typical tactical motifs, when to release tension, which pieces to keep for your endgames etc.

If you divide this by 4, which is what you do when you hop between openings, then you only get a smaller fraction of that value.

If you want to get a broader exposure, do it systematically for a low % of your games. E.g. every 10th game ( 10% of games ) try out something new.

Also, if you have a teacher, follow the teacher's advice, assuming he been teaching the Italian for years, he can give a big boost to your learning.

If you like the Caro against d4, QGD is an option as there is some overlap i.e. in the Carlsbad and IQP structures, so knowledge gained on one helps the other. In another thread here, someone pointed out that actually there's a QGD line that transposes to a Panov line directly and there's a book/course on that ( haven't read the book so don't have an opinion ).

Some folks also pair the Caro with the Slav or Semi Slav.

But if you have a teacher, the right answer is to play what your teacher recommends, he knows more about your strengths and areas of improvement.

In chess, individual moves mean nothing without the underlying ideas. This DB analysis (and all similar attempts) fall into the ditch of thinking that the moves are the important thing, and trying to work out the best plan, starting from the moves.

That's putting the cart in front of the horse. You should be choosing the correct plan FIRST, and then finding the moves to tactically support that plan.

Another completely different objection is that databases are heavily biased by the fact that people play tournament games intending to WIN, not just to explore the possible results.

Consider this hypothetical example:

A particular line of defense for Black has been played hundreds of times in Master tournaments, and has always proved to be a solid and reliable defense. It has accumulated excellent performance stats, with many solid draws and even has more wins than losses on record... an excellent score for an opening with the Black pieces.

Then White finds a convincing refutation of the entire line. And plays it in a tournament game, which causes quite a stir among tournament players. Suddenly everybody abandons that whole line of defense, not wanting to end up in a position that has clearly been refuted.

What will this look like in the database?

Answer : Since nobody is willing to enter that line anymore, no new games are played and the statistics cannot update to reflect the fact that the entire line is now known to be WORTHLESS. The database continues to show that line as a GOOD option for Black, even though it's WELL KNOWN to be a forced loss!

Purely hypothetical? 

Hardly.

Consider this well-known line in the Petroff :

All analysis based on database percentages is vulnerable to this flaw.