Bot-Policies auf recherchierte Level-Timelines: Runden-Ziele, Rolldown-Runden, Slow-Roll, Not-Roll bei wenig HP

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

backend/tests/test_game.py

@@ -240,3 +240,16 @@ def test_loot_drops_components(art, cfg):
     comp_art = {**art, "component_pool": ["C_A", "C_B"]}
     game = Game(comp_art, cfg, seed=2)
     assert all(i in ("C_A", "C_B") for i in game.items)  # Carousel-Drop
+
+
+def test_bot_leveling_follows_targets(art, cfg):
+    game = Game(art, cfg, seed=12)
+    levels_at = {}
+    while not game.over and game.round["label"] != "4-3":
+        label = game.round["label"]
+        if label in ("2-6", "3-3", "4-2"):
+            levels_at[label] = [b.level for b in game.bots if b.alive]
+        game.step()
+    assert all(lvl >= 5 for lvl in levels_at["2-6"])   # L5 ab 2-5
+    assert all(lvl >= 6 for lvl in levels_at["3-3"])   # L6 ab 3-1/3-2
+    assert any(lvl >= 7 for lvl in levels_at["4-2"])   # fast8/streaker auf 7

backend/tft/sim/policy.py

@@ -1,4 +1,9 @@
-"""Scripted planning policies — drive the bots and the autoplay benchmarks."""
+"""Scripted planning policies — drive the bots and the autoplay benchmarks.
+
+Level-Ziele und Rolldown-Zeitpunkte nach Standard-Spielweisen (Set 17):
+Baseline L4@2-1, L5@2-5, L6@3-2, L7@4-1, L8@4-5. Fast 8 spiked bei 4-2,
+Streaker levelt früh und rollt klein, Reroll cappt bei 6 und slow-rollt >50.
+"""
 
 import random
 
@@ -7,18 +12,40 @@ from tft.sim.player import InvalidAction, PlayerState
 from tft.sim.pool import Pool
 
 ARCHETYPES = {
-    # fast8: früh econ, ab Stage 3 leveln, Rolldown ab Level 8.
+    "fast8": {
-    "fast8": {"level_cap": 10, "xp_stage": 3, "econ_floor": 54,
+        "level_targets": {"2-1": 4, "2-5": 5, "3-2": 6, "4-1": 7, "4-2": 8, "5-2": 9},
-              "roll_stage": 4, "roll_level": 8, "roll_floor": 30},
+        "roll_rounds": {"4-2": 30, "5-2": 20},
-    # reroll: Level-Cap 7, rollt Überschuss über 50 für Upgrades.
+        "roll_low_hp": 40,
-    "reroll": {"level_cap": 7, "xp_stage": 3, "econ_floor": 50,
+        "slow_roll_level": None,
-               "roll_stage": 3, "roll_level": 7, "roll_floor": 50},
+    },
-    # streaker: gibt früh aus, levelt aggressiv, wenig Reserve.
+    "streaker": {
-    "streaker": {"level_cap": 10, "xp_stage": 2, "econ_floor": 20,
+        "level_targets": {"2-1": 4, "2-5": 5, "3-1": 6, "3-5": 7, "4-5": 8, "5-5": 9},
-                 "roll_stage": 4, "roll_level": 8, "roll_floor": 20},
+        "roll_rounds": {"4-1": 30, "4-5": 20},
+        "roll_low_hp": 50,
+        "slow_roll_level": None,
+    },
+    "reroll": {
+        "level_targets": {"2-1": 4, "2-5": 5, "3-2": 6, "5-1": 7, "6-1": 8},
+        "roll_rounds": {},
+        "roll_low_hp": 40,
+        "slow_roll_level": 6,  # ab Level 6: Gold über 50 verrollen (Slow-Roll)
+    },
 }
 
 
+def _round_key(label: str) -> tuple[int, int]:
+    stage, rnd = label.split("-")
+    return int(stage), int(rnd)
+
+
+def _target_level(params: dict, label: str) -> int:
+    key = _round_key(label)
+    return max(
+        (lvl for rnd, lvl in params["level_targets"].items() if _round_key(rnd) <= key),
+        default=0,
+    )
+
+
 def act(p: PlayerState, pool: Pool, artifact: dict, cfg: dict, rng: random.Random,
         rnd: dict, params: dict) -> None:
     """Eine Planungsrunde für einen Spieler."""
@@ -27,18 +54,28 @@ def act(p: PlayerState, pool: Pool, artifact: dict, cfg: dict, rng: random.Rando
 
     _buy_from_shop(p, pool, artifact)
 
-    stage = rnd["stage"]
+    label = rnd["label"]
-    while (stage >= params["xp_stage"] and p.level < params["level_cap"]
+    target = _target_level(params, label)
-           and p.gold >= params["econ_floor"] + cfg["xp"]["buy_cost"]):
+    while p.level < target and p.gold >= cfg["xp"]["buy_cost"]:
         try:
             player.buy_xp(p, cfg)
         except InvalidAction:
             break
 
-    at_target = p.level >= params["roll_level"] or p.level >= params["level_cap"]
+    floor = None
-    if stage >= params["roll_stage"] and at_target:
+    slow_level = params.get("slow_roll_level")
+    if slow_level is not None and p.level >= slow_level:
+        floor = 50
+    if p.level >= max(params["level_targets"].values()):
+        floor = 50  # Endstufe erreicht: Überschuss in Upgrades rollen
+    if label in params["roll_rounds"]:
+        floor = params["roll_rounds"][label]
+    if rnd["stage"] >= 4 and p.hp < params["roll_low_hp"]:
+        floor = 0  # Not-Rolldown bei niedrigem Leben
+
+    if floor is not None:
         _sell_junk(p, pool, artifact)
-        while p.gold > params["roll_floor"]:
+        while p.gold > floor:
             _buy_upgrades(p, pool, artifact)
             try:
                 player.reroll(p, pool, cfg, rng)