ACE 1.0.3

Library

• Fixed a major bug in alias move selection by including cards from the current trick
• Optimized the tree structure to reduce memory usage and improve performance
• Improved move ordering by comparing node value when visit counts are equal
• Refactored Game class for cleaner state handling and public API exposure
• Added Library.Version to the public API for version retrieval
• Added version metadata to the library project file (.csproj)

ACE 1.0.2

Library

• Fixed tracking of the best outcome among expanded nodes during simulation step
• Fixed UCB utility bonus to correctly rely on completed results instead of virtual loss
• Clarified Engine.Evaluate(...) summary documentation and evaluation wording

ACE 1.0.1

Library

• Fixed outcome flipping to correctly handle overtricks across player perspectives

Console app

• Replaced set command with filter for clearer hand constraint management
• Added dummy command to set and reveal dummy's hand after the opening lead

ACE 1.0

Library

• Released version 1.0, officially moving the project out of WIP into stable release
• Fixed position evaluation to correctly account for players' overtricks and undertricks
• Improved Limiter to extend search beyond current trick based on remaining depth
• Minor code cleanup

ACE 0.09 (WIP)

Library

• Improved alias detection for equivalent moves to better share node statistics
• Simplified final move selection to always use visit count (removed Metric)
• Added Engine.SetSeed(int seed) to make runs repeatable with fixed seed
• Added Game.SetDummy(string hand) for dummy hand setup and reveal
• Minor code cleanup

ACE 0.08 (WIP)

Library

• Replaced previous search with a Monte Carlo Tree Search (ISMCTS) approach
• Added tree reuse to preserve and continue subtrees when depth is unlimited
• Improved search by expanding multiple nodes per iteration instead of just one
• Engine.Evaluate now returns evaluation statistics (action, value, visits, depth)
• Added optional Config parameter to Engine.Search / Engine.Continue:
  • Exploration - adjusts how strongly less-visited moves are preferred
  • Limiter - limits search expansion to the current trick when enabled
  • Metric - defines how final moves are selected from the search tree
• Code cleanup

Console app

• Added iteration limit via -l / -limit flag for search / continue
• Updated output to match new evaluation format
• Code cleanup

ACE 0.07 (WIP)

Library

• Major update: rebuilt tree search and evaluation logic from the ground up
• Tree representation changed from deterministic children to stochastic edges
• Simplified evaluation by removing the mixed opponent/partner model variants
• Updated ISS to score moves by expected outcome with a simple minimax rule
• Replaced the old rolling info-state hash with a new extended key structure
• Engine.Evaluate now takes only a smoothing prior: Evaluate(double prior)
• Search depth is now limited to the current trick horizon for extra performance
• Random sampling is now deterministic per thread, making runs reproducible
• The dummy seat is now determined and stored at startup via Game.Dummy

Console app

• evaluate command simplified: removed setting of partner/opponent models

ACE 0.06 (WIP)

Console app

• Fixed a major bug that caused inconsistent number parsing

ACE 0.05 (WIP)

Console app

• Clarified the usage example of evaluate with a correct model for example deal

Library

• Default prior in policy calculations is now set to 0 (previously was 1)

ACE 0.04 (WIP)

General

• All card input, output, and display now use standardized suit-rank (PBN) format
• Updated the readme file to document these changes and reflect the correct usage