We updated the above dataset at to now 100M SF16 + SF17.1 games, 100k for each starting position.
I asked ChatGPT for a first ansatz to identify pattern and White does much better whenever …
... At least one rook is already on a corner square (a1 or h1).
... Queen is NOT blocking the king-rook corridor.
... No knight starts in a corner.
... At least one bishop already has a free diagonal.
... King starts on file b, c or d (not a / e / h).
White owns the first tempo. Any start that saves more tempi (or avoids self-blocks) lets White cash that advantage; starts that impose extra self-tempo taxes hand Black an easier life.
Is that a reasonable analysis? -- Many thanks!
I hope this is the right place to announce our github project
Opening Analysis for Chess960
We identify openings across all Chess960 variants.
👉 We analyzed about 20 million Chess960 matches we generated with Stockfish16.
👉 We give a formal definition of an opening and apply it to the dataset.
👉 We rediscover many well-known chess openings.
👉 Using this technique, we identify novel openings for all Chess960 variants.
Please send us a message if
👉 ... you have comments about our Stockfish setup.
👉 ... you see better ways to generate datasets.
👉 ... you have reasons to believe that we should increase the database size.
👉 ... you see reasonable ways to group the indentified openings into categories.
👉 ... if you think we should improve the data presentation on the github project.
Disclaimer:
We do not know much about chess beyond the rules, and we never posted into this forum
The github name is /stumpc5/chess960