I decided I don't understand anything of chess programming.

Chessnix is a UCI chess engine.

• Needs Zig 0.15.2 to compile.
• chessnix is a windows 64 bits exe for modern computers.
• chessnix_bot can be played against on lichess.org.

• Working on it, not pretending to understand anything. Waiting for Zig 0.16.
• Stronger: I hope.
• Changed: Alpha Raises Reduction LMR (instead of depth reduction) after score > alpha.
• Changed: Only update pv when pv-node. Removed update pv in qsearch.
• Changed: Time management.
• Changed: TT structure and entry replacement strategy.
• Changed: Prefetch TT for a slight speed boost.
• Changed: Some minor optimizations.
• Removed: Reward / punish moves in qsearch.
• Bug solved: Another terrible one: qsearch comparing a zero score with TT score instead of the eval.
• Bug solved: Parsing negative time.

• Stronger: ~3168 ELO on CCRL.
• Added: Slight center bias in quiet move ordering for shallow depths (experimental).
• Added: Search history pruning, history reduction.
• Added: Store raw static evaluation in TT.
• Added: Slight capture history bonus and malus in quiescence search.
• Removed: Killer moves.
• Removed: Maintaining pins of both sides. I will not code a 'perfect' SEE in the near future.
• Removed: Unused see_score function.
• Changed: Lots of tweaks in the search algorithm (history pruning + reduction, different LMR table).
• Changed: Correction history.
• Changed: History structure and calculations, especially continuation history.
• Changed: Rescaled history values.
• Changed: Moved 'ply' from position to search.
• Changed: Included a slight score correction during search using rule50.
• Changed: Bound logic when storing to TT. I think it is correct now.
• Changed: Move generation creates ExtMoves (64 bits) instead of the raw moves (16 bits), avoiding copying stuff.
• Bug solved: Terrible LMR table accessing out of bounds value resulting in god knows what.
• Bug solved: Node clearing (before enter search and on making nullmoves).
• Bug solved: Corrected 'id name' uci output.

• Stronger: ~3015 ELO on CCRL.
• Added: Chess960 support.
• Added: Staged movepicking.
• Added: Razoring.
• Added: Continuation History.
• Added: Correction History.
• Added: Internal Iterative Reduction.
• Added: CutNode reduction.
• Added: SEE pruning.
• Changed: Transposition Table with 2 entries per bucket.
• Changed: Aspiration Window strategy.
• Changed: Late Move Pruning. Using an improvement rate + depth formula.
• Changed: Late Move Reduction. Using a precomputed table now.
• Changed: Quiescence search. Does not store bestmove anymore.
• Changed: Node counting. Only incremented after doing a move.
• Changed: Finetuned search in general.
• Move generation refactor.
• Position always storing pins for both sides. For future perfect SEE.
• Bug solved: Invalid principal variation output.
• Bug solved: Bishop pair evaluation bug. The mask for white / black squares was wrong.
• Bug solved: A terrible 'get all attacks' bug affecting evaluation, SEE and move ordering.

Does not exist. This was a local bugfix for pv output.

Engine plays reasonable chess. Entered CCRL with 2842 ELO. Many many things need still to be changed, but it is time for a first version.

Engine plays terrible chess, mainly because I am not interested at all in writing an evaluation function. To be continued...

• Major number changes when the evaluation function changes.
• Minor number changes for improvements and bugfixes.

• Version 1.x 3200+ ELO.
• Version 2.x 3400+ ELO (extended evaluation tuning).
• Version 3.x 3600+ ELO (use neural network training).

• setoption name UCI_Chess960 value true -> enables chess960.
• position startpos moves e2e4 e7e5. (moves" is optional) -> sets the startposition, does the moves.
• position fen rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1 moves e2e4. (moves is optional) -> sets the fen, does the moves.
• eval -> give the static evaluation of the position.
• go depth 7 -> search position to depth 7.
• go movetime 5000 -> search position for 5000 milliseconds.
• perft 5 -> performs a perft test with subtotals per move.
• qperft 5 -> performs a perft without subtotals per move.
• bench -> performs perft speedtests on 4 positions.
• d -> draw the current position and some info.
• quit -> stop the program.
• cls -> clear the terminal.

Aron Petkovski. I used most of the tuned evaluation of an old version of Integral (3) to get me started (from chessnix 2.0 onwards I will do my own tuning). Jonthan, Janez and Colin on Discord. The people of TalkChess, Zig Forum, CCRL. The authors of the engines I used for testing: Bbc, Cheese, Chessplusplus, Colossus, Cwtch, Infrared, Integral, Lambergar, Linx, Lishex, Mess, Monty, OpenCritter, PlentyChess, Pounce, Priessnitz, Seawall, Seredina, Shallow, Shallowguess, Simbelmine, Stash, Supernova, Teki, Yakka. Some extra thanks to these people who made their engine support chess960. I also learned a lot from the source code of all these engines. Chessnix contains, besides my own inventions, a wild mix of ideas from the chess programming wiki as well as several other chess engines (Alexandria, Cwtch, Lambergar, Integral, PlentyChess, Pawnocchio, Stockfish and probably others).

I spent many years on Lemmix, the DOS Lemmings clone. So in my feeling the name had to end with "ix" as well. Later I found out a nix is a kind of water spirit. So then I added the queen water spirit logo.

I am a github noob and only use a main branch as a backup of the current state on my PC. With every release an extra zip of the source code at that moment is provided.