A UCI-compliant chess engine written in C# targeting .NET 6. Charlie uses classic alpha–beta search with iterative deepening, quiescence, and simple heuristics for move ordering and time management. The codebase is small, readable, and intended as a practical reference/starting-point for chess-engine work in C#.

• Protocol: UCI (Universal Chess Interface)
• Language/runtime: C# on .NET 6
• Search: Alpha–beta with aspiration windows, principal-variation search, late move reductions, and tactical/extensions
• Evaluation: Material + piece-square tables, territory/space, hanging-piece penalties, and pawn-structure features
• Board: Bitboards with precomputed sliding attacks and Zobrist hashing

• .NET 6 SDK or newer

# From the repo root
dotnet build -c Release

# Run the engine as a console app that speaks UCI over stdin/stdout
# Windows
dotnet run -c Release --project Charlie

In a UCI-compatible GUI (e.g., CuteChess, Arena), configure an engine pointing to the Charlie executable.

uci
isready
position startpos moves e2e4 e7e5 g1f3
go depth 10
stop
quit

• go depth <N>: fixed-depth search
• go wtime <ms> btime <ms> [winc <ms> binc <ms>]: time-managed search using simple budgeting and an increment-aware time manager

• setoption name Clear Hash: clears the transposition table

• bench [depth <N>]: run a bench across a suite of positions (see Charlie.BenchTest.Bench)
• perft <depth>: run a perft from the current position and print per-move counts and totals
• eval fen <FEN...>: evaluate a position once and print a centipawn/mate score

• Bitboards for all piece sets and occupancy
• Precomputed directional attack rays for rooks/bishops and king/knight move masks
• Castling legality includes square occupancy and threat checks
• En passant support for both sides; EP affects hashing (so threefold is position-accurate)

• Iterative deepening from depth 1 upward; root move ordering by simple promise heuristic
• Aspiration windows centered on the previous iteration’s score
• Principal-variation (PV) search with a narrow re-search on fail-high/low
• Reductions/extensions:
  • Late move reductions on quiet non-first moves
  • Extensions for promotions, in-check, and a small PV extension around shallow depths
• Quiescence search on captures, promotions, and en-passant; forward-promotion pushes included in qsearch from 7th rank
• Simple repetition detection via Zobrist history (BoardState.IsThreeMoveRepetition)

• Material values: P=100, N=320, B=338, R=525, Q=920
• Piece-square tables for all pieces; phase-aware king tables (opening vs endgame)
• Territory/space: counts of attacked squares and uncontested territory
• Hanging-piece penalties and simple pawn-structure features (isolated, doubled, passed)
• Light lazy-evaluation cutoff for clearly won/lost positions

• Zobrist hashing of full positions including side-to-move, castling, and en passant
• Table stores the best move and search depth for ordering; no bound types or scores recorded (kept intentionally simple)

• Heuristic budgeting using available time and increment
• More time is allocated when the score is poor or when the PV best move changes; less time when we are confident in the best move

CharlieTest/Program.cs shows how to run self-play tournaments via CuteChess on Windows. A typical CLI invocation looks like:

"C:\\Program Files (x86)\\cutechess\\cutechess-cli.exe" \
  -engine cmd="<path-to-Charlie-exe>" name="Charlie" \
  -engine cmd="<path-to-Charlie-exe>" name="Charlie-Alt" \
  -each tc=5+0.05 \
  -games 2 -rounds 50 -recover -concurrency 4

Adjust the paths and time controls to your environment.