Adopt paper's 6-level greedy priority (8.590% → 12.188%) by dsjoerg · Pull Request #53 · dacracot/Klondike3-Simulator

and others added 2 commits

April 17, 2026 21:39

Replaces the fixed {s2g, b2b, b2g, s2b} action sequencing with the
prioritized greedy heuristic from Bjarnason, Fern, Tadepalli 2009
("Lower Bounding Klondike Solitaire with Monte-Carlo Planning",
ICAPS-09, pg. 3):

  1. tableau -> foundation that reveals a face-down card
  2. any other move to a foundation stack
  3. tableau -> tableau that reveals a face-down card
  4. deck -> tableau
  5. foundation -> tableau (not implemented; stub returns false)
  6. tableau -> tableau that does not reveal a face-down card

After each move, action selection restarts from priority 1, so an
ace exposed by a board-to-board reveal is sent to the foundation
before any deck play, and so on.

Adds two predicates on Board (columnHasHidden, removeCardWouldReveal)
and splits FromBoard's toBoard / toGoal into reveal / no-reveal
variants that the Player loop selects between.

Measured at --three --attempts 100000 on an M-series Mac:
  seed 1111  before 8.637%  after 12.188%  (+3.55pp, 2:50 -> 1:47)
  seed 2222  before 8.465%  after 12.202%  (+3.74pp, 3:09 -> 2:11)

The cited paper reports 12.992% for its greedy baseline on 1M games.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…antics

- toBoardReveal/toBoardNoReveal: use removeCardWouldReveal(king) for king
  moves (playKingFromBoard moves only the king, not the face-up run);
  columnHasHidden(src) remains correct for playCard run moves.
- toGoalReveal: return after first successful play so the Player loop
  re-evaluates from priority 1, matching the paper's algorithm strictly.
- README: 12.188% -> 12.154% (seed 1111, 100k, three-card; within
  binomial noise of prior measurement).