Implement SOR.Ai Phase 2: decision engines, candidate generator, heuristic scoring
Phase 2.3 — Random & Heuristic engines:
- IAiDecisionEngine interface + AiAction discriminated union (Draw, PlayCard, Move, Attack, Sacrifice, Ascend, EndTurn)
- AiDifficultyProfile / AiDifficultyProfiles: in-code profiles for tiers 1–10 with timing, mistake rates, and scoring weights
- RandomDecisionEngine: picks random legal action each step (Difficulty 1 — Novice)
- HeuristicDecisionEngine: greedy loop with JSON-clone simulation + mistake injection (Difficulties 2–10)
Phase 2.4 — ActionCandidateGenerator:
- Enumerates all legal moves: Spirit/Terramorpha/Ruin/Repentance/Instinct channelings, movement (Walk/Flight/Jump/Teleport), attacks, sacrifices, ascends
- Delegates range/path/affinity logic to SOR.Game.Grid (Pathfinding, AttackSystem, AffinitySystem)
Phase 2.5 — Heuristic scoring:
- ScoringFunctions.Evaluate: material, affinity control, threat, positional, resource weights
- ThreatDetector: counts enemy units that can reach AI ArchSpirit (with move-into-range lookahead)
AiPlayerSession wired to select engine by difficulty profile, submit full action plan
with humanized think/action delays, and catch GameException gracefully per action.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>