Scope: complete Rust engine for MCTS rollouts #51
Description
Context
The Python engine (packages/engine/) is the source of truth with 100% Java parity for Watcher. The Rust engine (packages/engine-rs/) is a partial port designed for fast MCTS combat rollouts.
Current State
- Python: ~7000 lines, 100% complete, 6183 tests, verified parity
- Rust: ~2700 lines, 35/75 Watcher cards (47% common, 0% uncommon/rare), 63 tests (broken due to PyO3 linking)
- Quality: 6/10 (clean code, good benchmarks, incomplete coverage)
Why This Matters
- MCTS rollouts need fast combat simulation (30ms target per turn)
- Python combat sim is too slow for deep trees
- RTX 3070 could run parallel MCTS if Rust engine is complete
- Current training bottleneck is RL convergence, not speed — so this is Phase 2
Scope
- Fix PyO3 test linking (2h)
- Add missing uncommon cards (~25 cards, 8h)
- Add missing rare cards (~10 cards, 4h)
- Parity verification against Python engine (4h)
- Integration with training pipeline (fallback to Python for unknown cards)
Decision
- Defer until 50%+ WR (recommended by audit)
- Start incremental card porting now
- Prioritize for RTX 3070 MCTS use case
🤖 Generated with Claude Code