Stockfish Contributions

Stockfish is widely considered the strongest chess engine in existence. Every change to the engine must pass sequential probability ratio tests (SPRT) on Fishtest, a distributed testing framework that plays tens of thousands of games against the current master before a patch is accepted. Below are my merged contributions.

15Elo gainers
68Commits

Elo-gaining patches

  1. Move hindsight reductions above cutoffs (2025-11-22)
  2. Post-NNUEv10 tune (2025-11-22)
  3. Refine constant in correction history update (2025-11-03)
  4. Check evaluation after ttMove before doing a tt cut (2025-05-18)
  5. Tweak futility formula (2025-04-20)
  6. Decrease lmr depth if static eval decreases a lot (2025-02-12)
  7. Retroactive reduction decrease if eval improves (2025-01-16)
  8. Refine probcut (2025-01-12)
  9. Corrplexity for futility pruning (2025-01-02)
  10. Bonus for a prior capture that causes a fail low (2024-11-22)
  11. Reduce on ttcaptures if not capture (2024-09-01)
  12. Grade countermove bonus for low statscores (2024-06-29)
  13. Permute NNUE feature-transformer weights to increase sparsity (2023-06-23)

Elo-gaining patches based on my ideas

  1. Tweak History Bonus (2025-04-27), implemented by FauziAkram
  2. Tune Fail Low Bonus (2024-05-27), implemented by Shawn Xu

Simplifications

Plus 52 simplifications passing non-regression testing, removing code while keeping playing strength. See all commits.