What if Chess960 had asymmetric setups?
Why do we always use the same starting position for both players?
It’s standard practice — generate one random 960 setup, and both sides begin from it. This preserves symmetry and feels fair. But it also means we never test how a given configuration performs against all possible others.
So I’m exploring an alternative idea: What if each player were “anchored” to their own preferred starting position — like a general returning to their favorite battle formation — and played against every other possible setup?
This shifts the perspective: instead of asking "How does this position play?", we ask: "Which positions are strong across the widest range of opponents?"
In a few days, I’ll launch chess960² — a year-long computational run where Stockfish 15.1 (the last version before NNUE was enabled by default) plays all 960×960 = 921,600 possible pairings of starting positions.
White uses position A, Black uses position B — for every combination. Time control: 3 seconds per move (fixed). Goal: Build a public dataset mapping the relative strength of all 960 positions when tested systematically against each other. But before launch, I’m curious: Is symmetry essential to fairness in Chess960 — or does it limit our understanding of positional strength? And could a format where players specialize in one setup lead to deeper strategic identity?
Would love to hear historical context, theoretical objections, or prior work on asymmetric starting conditions.