v2.5

The gain is from squeezing out more in selfgen, but now I've exhausted all easy improvements.
Expect a proper release on v3.0. (The binary below still requires AVX512)

v2.2

With an embedded NNUE trained using original data from multiple iterations of selfplay, starting from zero. Around 300-400 elo stronger than v1.8.
(binary below requires AVX512)

v1.8

Switches from gcc to clang (which better optimizes movegen).
Also contains various search tweaks.
Will add nnue soon probably

v1.6

Changes the late move reductions implementation to a much more effective version. Makes tweaks to reverse futility pruning and qsearch. Adds dynamic time management (soft and hard time limits).

v1.4

Adds late move reductions and reverse futility pruning. Also tweaks the null move pruning formula.

v1.2

Adds history heuristic and null move pruning. This intermediate release is offered while I wait for conclusive testing on newer versions.
(Note: Releases probably only work for windows, to compile the equivalent source code should be 264613d)

v1.0

Adds TT to engine, ~150 elo gain against v0.9, includes UCI wtime btime support (so now should work with most time controls)
May possibly crash on last move of 50-move rule

v0.9

Includes semi-functional UCI (enough to possibly use with a GUI), iterative deepening with alpha-beta and quiescence search. Uses Pesto Evaluation Function with slight changes.
To do:
Add TT
Add other search enhancements
Tune eval function