Develop a more sophisticated quiescence filter for position selection

Develop an improved, computationally feasible quiescence filter for selecting Lichess game snapshots in standard chess that more reliably excludes non-quiescent positions (e.g., immediate recaptures) without introducing dependence on internal valuation systems, so that logistic-regression estimates of piece values reflect stable material imbalances rather than transient capture sequences.

Background

The study estimates piece values by logistic regression using snapshots from millions of Lichess games, and applies a simple no-capture-next-move filter as a crude proxy for quiescence to avoid counting transient imbalances that will be immediately corrected by recapture.

The authors find piece values increase under the single no-capture filter but note tension between isolating genuinely quiescent positions and measuring the contribution of captures to winning chances. They conclude that a more sophisticated, scalable quiescence filter is warranted to improve causal interpretability of material value estimates.