Prevent CPU players from swapping 8+ value cards (8, 9, 10, J, Q) into
face-down positions, which is statistically bad since expected hidden
card value is ~4.5.
Fixes applied:
- Add value threshold (7) to unpredictability random swap path
- Restrict comeback bonus to cards with value < 8
- Reduce speculative wolfpack Jack bonus from 6x to 2x aggression
- Add safety filter to remove hidden positions for 8+ cards
- Fix endgame logic to discard 8+ instead of forcing swap into hidden
- Skip hidden positions in denial candidate list for 8+ cards
- Add swapped_high_into_unknown tracking to SimulationStats
Reduces "swapped 8+ into unknown" dumb moves from ~85 per 200 games
to ~6 per 500 games (0.054% rate, down from ~2%).
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
AI now considers the next player's visible cards before discarding:
- Checks if discarding would give opponent a pair opportunity
- Calculates denial value based on card value and game phase
- May keep a worse card to deny opponent when cost is acceptable
- Denial threshold varies by AI personality (aggression)
Also updates simulation to recognize denial as a valid reason for
swapping good cards, preventing false "swapped good for bad" flags.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Enable testing AI behavior under different rule sets via CLI:
- --preset flag for named configurations (baseline, eagle_eye, etc.)
- --rules flag for custom comma-separated rules
- --compare flag for side-by-side preset comparison with metrics
- Improved dumb move detection for negative_pairs_keep_value rule
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Features:
- Multiplayer WebSocket game server (FastAPI)
- 8 AI personalities with distinct play styles
- 15+ house rule variants
- SQLite game logging for AI analysis
- Comprehensive test suite (80+ tests)
AI improvements:
- Fixed Maya bug (taking bad cards, discarding good ones)
- Personality traits influence style without overriding competence
- Zero blunders detected in 1000+ game simulations
Testing infrastructure:
- Game rules verification (test_game.py)
- AI decision analysis (game_analyzer.py)
- Score distribution analysis (score_analysis.py)
- House rules testing (test_house_rules.py)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
