docs(p2-67-followup): ✅ flip MCTS tactical TreeState impl to done (Wave-3 audit)

Audit confirmed the full surface already in-tree (TacticalTreeState +
apply_tactical_action all 14 variants + score_for_player [0,1] + generic
most_visited_root_action_cloned + bench wiring). No rebuild.

Empirical gates on apricot (CARGO_TARGET_DIR shared):
- cargo check --workspace clean; mc-ai 278/278 lib pass
- claude_real_mcts_vs_heuristic_ais_transcript: 283s/293s (<300s), Claude
  #1 (68 cities vs 19 / eliminated), action diversity move×453/queue_unit×48/
  queue_building×3/found×1, two runs byte-identical (determinism).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

.project/objectives/objectives.json

@@ -1,12 +1,12 @@
 {
-  "generated_at": "2026-06-04T16:45:14Z",
+  "generated_at": "2026-06-04T18:36:52Z",
   "totals": {
-    "done": 241,
+    "done": 242,
     "in_progress": 1,
     "missing": 1,
     "oos": 29,
     "partial": 18,
-    "stub": 9,
+    "stub": 8,
     "superseded": 4,
     "total": 303
   },
@@ -2990,10 +2990,10 @@
       "id": "p2-67-followup-mcts-tactical-state-impl",
       "title": "TreeState impl for TacticalState — wire real MCTS into the AI decision path",
       "priority": "p2",
-      "status": "stub",
+      "status": "done",
       "scope": "game1",
       "owner": "simulator-infra",
-      "updated_at": "2026-05-13",
+      "updated_at": "2026-06-04",
       "blocked_by": [],
       "summary": "After the night's bug-fix pass (Bugs 1-5 closed, simulation fully playable, last-survivor victory firing), the question \"can Claude beat the hardest AI?\" hit a deeper architectural finding:"
     },
@@ -3520,7 +3520,7 @@
   "remaining_by_lead": [
     {
       "owner": "simulator-infra",
-      "remaining": 8
+      "remaining": 7
     },
     {
       "owner": "asset-sprite",

.project/objectives/p2-67-followup-mcts-tactical-state-impl.md

@@ -2,12 +2,12 @@
 id: p2-67-followup-mcts-tactical-state-impl
 title: "TreeState impl for TacticalState — wire real MCTS into the AI decision path"
 priority: p2
-status: stub
+status: done
 scope: game1
 category: simulation
 owner: simulator-infra
 created: 2026-05-13
-updated_at: 2026-05-13
+updated_at: 2026-06-04
 blocked_by: []
 follow_ups: [p2-67]
 ---
@@ -145,14 +145,22 @@ The existing one is specialized on `Tree<GameRolloutState>`. Add a generic versi
 
 ## Acceptance
 
-- ☐ `TacticalTreeState` exists with `TreeState` impl.
-- ☐ `apply_tactical_action` covers all 14 `Action` variants.
-- ☐ `score_for_player` returns [0, 1] reward.
-- ☐ Generic `most_visited_action_at_root<S>` added to `Tree<S>`.
-- ☐ `claude_real_mcts_vs_ai_transcript` test runs in < 5 minutes wall clock.
-- ☐ Claude with 1000-rollout budget MCTS shows meaningful action diversity vs the heuristic baseline (action-frequency table differs).
-- ☐ Result reported: did Claude win? Top 3? Bottom?
-- ☐ Determinism: same seed → byte-identical transcript across two runs.
+- ✓ `TacticalTreeState` exists with `TreeState` impl — `mc-ai/src/tactical/tree_state.rs` (`impl TreeState for TacticalTreeState`, `new_root`, exported in `tactical/mod.rs:52`).
+- ✓ `apply_tactical_action` covers all 14 `Action` variants — `mc-ai/src/tactical/apply.rs:39` (MoveUnit, AttackTarget, Fortify, Heal, FoundCity, SetProduction, EnqueueBuild, AssignCitizen, Scout, IssuePatrol, DeploySiege, PackSiege, Bombard, PromotionPicked).
+- ✓ `score_for_player` returns [0, 1] reward — `mc-ai/src/tactical/scoring.rs:46` (`(0.6*city_share + 0.4*unit_share).clamp(0.0, 1.0)`); 4 unit tests cover balanced/dominant/losing/out-of-range.
+- ✓ Generic most-visited-root-action added to `Tree<S>` — `mc-ai/src/mcts_tree.rs:348 most_visited_root_action_cloned()` (clone-based generic; the older `Copy`-bound `most_visited_action_at_root` is `ActionKind`-specialised, this one is the generic equivalent).
+- ✓ `claude_real_mcts_vs_heuristic_ais_transcript` runs in < 5 min — release run 283.48s (run 1) / 292.97s (run 2), both < 300s ceiling. `mc-player-api/tests/full_game_transcript.rs:2064`.
+- ✓ Action diversity vs heuristic — MCTS action-frequency over 500 turns: `move`×453, `queue_unit`×48, `queue_building`×3, `found`×1 (`.local/demo-runs/2026-05-13-claude-real-mcts/recap.md`).
+- ✓ Result reported — Claude (slot 0, MCTS) finished **#1**: 68 cities / 559 units / 152 gold vs slot-1 heuristic 19 cities and slot-2 heuristic eliminated (0 cities). No `GameOver` (500-turn ceiling reached, no domination victory), but Claude dominant on every axis.
+- ✓ Determinism — two full 500-turn runs produced `TRANSCRIPT_BYTE_IDENTICAL` (`transcript.jsonl`) and identical recap.
+
+## Verification — Wave 3 audit (2026-06-04)
+
+Audit confirmed the full surface already in-tree (commit `378e5799e` "threshold + tree-state mgmt" + later). No rebuild needed. Gates run on apricot (RUN host), `CARGO_TARGET_DIR=~/.cache/mc-cargo-target-shared`:
+
+- `cargo check --workspace` — clean (doc warnings only).
+- `cargo test -p mc-ai --lib` — 278 passed, 0 failed (incl. 4 `tree_state::tests`).
+- `cargo test -p mc-player-api ... --ignored claude_real_mcts_vs_heuristic_ais_transcript` — 1 passed, ×2, byte-identical transcripts.
 
 ## Why this size