Final infrastructure and documentation fix for Phase 2 validation

README.md

@@ -82,9 +82,12 @@ The codebase follows a clean separation-of-concerns architecture across four dis
 
 ```
 .
+├── graders/                     # Unified Graders Package (Root)
+│   ├── easy.py                  # Easy task scoring logic
+│   ├── medium.py                # Medium task scoring logic
+│   └── hard.py                  # Hard task scoring logic
 ├── core/                        # Simulation Logic Layer
 │   ├── drone.py                 # Movement physics & battery drain
-│   ├── graders.py               # Unified scoring logic (0.01 - 0.99)
 │   ├── grid_generator.py        # Map generation logic
 │   ├── obstacles.py             # Collision & terrain detection
 │   ├── state_manager.py         # Episodic state management
@@ -471,7 +474,7 @@ MODEL_NAME=Qwen/Qwen2.5-72B-Instruct python inference.py
 The runner emits structured benchmark-compatible log lines:
 
 ```
-[START] task=drone_env.core.graders:grade_easy env=drone_env_v1 model=Qwen/Qwen2.5-7B-Instruct
+[START] task=drone_env.graders:grade_easy env=drone_env_v1 model=Qwen/Qwen2.5-7B-Instruct
 [STEP] step=1 action=RIGHT reward=0.10 done=false error=null
 [STEP] step=2 action=RIGHT reward=0.10 done=false error=null
 [STEP] step=3 action=RIGHT reward=0.10 done=false error=null
@@ -568,15 +571,15 @@ app: drone_env.server.app:app
 port: 8000
 tasks:
   - id: easy_delivery
-    grader: drone_env.core.graders:grade_easy
+    grader: drone_env.graders:grade_easy
   - id: medium_delivery
-    grader: drone_env.core.graders:grade_medium
+    grader: drone_env.graders:grade_medium
   - id: hard_delivery
-    grader: drone_env.core.graders:grade_hard
+    grader: drone_env.graders:grade_hard
 graders:
-  - id: drone_env.core.graders:grade_easy
-  - id: drone_env.core.graders:grade_medium
-  - id: drone_env.core.graders:grade_hard
+  - id: drone_env.graders:grade_easy
+  - id: drone_env.graders:grade_medium
+  - id: drone_env.graders:grade_hard
 ```
 
 ### Validate Before Submission
@@ -611,8 +614,20 @@ docker build .
 openenv validate
 ```
 
+#### Local Grader Check
+To verify that all 3 tasks have valid, resolvable graders before pushing:
+```bash
+python check_graders.py
+```
+
 A passing run produces:
 ```
+========================================
+  Summary: 3 valid graders found.
+  🚀 LOCAL CHECK PASSED.
+========================================
+```
+
 ========================================
   All 3/3 checks passed!
   Your submission is ready to submit.
@@ -712,7 +727,7 @@ $$r_{\text{shaping}} = (d_{\text{before}} - d_{\text{after}}) \times 0.05$$
 
 ## Grading & Evaluation
 
-Scores are computed by `core/graders.py` using a unified formula:
+Scores are computed by the `graders/` package using a unified formula:
 
 $$\text{score} = 0.8 \times \underbrace{\frac{\text{deliveries done}}{\text{deliveries total}}}_{\text{delivery ratio}} + 0.2 \times \underbrace{\left( 0.5 \cdot \text{battery} + 0.5 \cdot \left(1 - \frac{\text{steps}}{\text{max steps}}\right) \right)}_{\text{efficiency}}$$
 
@@ -774,19 +789,21 @@ The **SkyRelic** ecosystem is divided into four primary layers, interconnected v
 
 ```
 drone_env/
-├── core/
-│   ├── drone.py               # compute_next_pos(), drain_battery()
-│   ├── graders.py             # grade_easy/medium/hard(), GRADERS dict
-│   ├── grid_generator.py      # generate_city_map(), EMOJI, LEGEND
-│   ├── obstacles.py           # check_move() → outcome, cell_type
-│   ├── state_manager.py       # new_episode_state() → DroneState
-│   └── tasks.py               # TASK_CONFIG dict (all difficulty params)
-├── rl/
-│   ├── model.py               # MapEncoder, PathQNet, ACTIONS, CELL2IDX
-│   ├── policy.py              # EpsilonGreedyPolicy
-│   └── trainer.py             # record_episode(), PathLearner, get_action_from_policy()
-├── server/
-│   ├── app.py                 # FastAPI app, all routes, TerminalLogManager
+├── graders/                     # Unified grader package
+│   ├── __init__.py              # GRADERS lookup dictionary
+│   ├── easy.py                  # Easy task logic
+│   ├── medium.py                # Medium task logic
+│   └── hard.py                  # Hard task logic
+├── core/                        # Simulation physics & tasks
+├── rl/                          # DQN & Neural Training
+├── server/                      # FastAPI & Dashboard
+├── data/                        # Persistence (Memory & Logs)
+├── tests/                       # Unit & API tests
+├── check_graders.py             # Local grader validation script
+├── openenv.yaml                 # Mission Manifest
+├── pyproject.toml               # Package configuration
+└── validate-submission.sh       # Submission validator
+```
 │   ├── grid_world_environment.py  # DroneDeliveryEnvironment (OpenEnv base)
 │   ├── Dockerfile             # Multi-stage production image
 │   └── static/                # Browser dashboard (HTML/JS/CSS)
@@ -887,8 +904,8 @@ Build system uses [Meta's BSD-licensed](https://opensource.org/licenses/BSD-3-Cl
 The **SkyRelic** environment has been updated to fully comply with the **Meta PyTorch Hackathon Phase 2 Deep Validation** requirements.
 
 ### 🛡️ Validation Fixes
-- **Strict Score Clamping**: All mission scores and rewards are now strictly clamped to the **(0.01, 0.99)** range in `core/graders.py` and `server/grid_world_environment.py`. This prevents the "out of range" (exactly 0.0 or 1.0) failures reported by the automated validator.
-- **Full Identity Sync (Grader Discovery)**: Task and grader identifiers have been synchronized across the manifest (`openenv.yaml`), backend API, and simulation core using full Python module paths (e.g., `drone_env.core.graders:grade_easy`). This ensures the Meta validator can successfully discover and import the grading functions.
+- **Strict Score Clamping**: All mission scores and rewards are now strictly clamped to the **(0.01, 0.99)** range in the `graders/` package and `server/grid_world_environment.py`. This prevents the "out of range" (exactly 0.0 or 1.0) failures reported by the automated validator.
+- **Full Identity Sync (Grader Discovery)**: Task and grader identifiers have been synchronized across the manifest (`openenv.yaml`), backend API, and simulation core using full Python module paths (e.g., `drone_env.graders:grade_easy`). This ensures the Meta validator can successfully discover and import the grading functions.
 - **Differentiated Reward Scalars**: To provide clearer learning signals, reward scalars for step, wait, and collision penalties have been updated to difficulty-specific tiers:
     - **Easy Mission**: 0.10 (10%)
     - **Medium Mission**: 0.15 (15%)

check_graders.py

@@ -0,0 +1,77 @@
+import yaml
+import importlib
+import sys
+import os
+from pathlib import Path
+
+def check_graders():
+    # Ensure current directory is in path
+    sys.path.insert(0, os.getcwd())
+    
+    yaml_path = Path("openenv.yaml")
+    if not yaml_path.exists():
+        print("❌ Error: openenv.yaml not found!")
+        return
+
+    try:
+        with open(yaml_path, 'r') as f:
+            spec = yaml.safe_load(f)
+    except Exception as e:
+        print(f"❌ Error parsing openenv.yaml: {e}")
+        return
+
+    tasks = spec.get("tasks", [])
+    print(f"🔍 Found {len(tasks)} tasks in openenv.yaml\n")
+
+    valid_count = 0
+
+    for task in tasks:
+        task_id = task.get("id")
+        grader_str = task.get("grader")
+        
+        print(f"--- Task: {task_id} ---")
+        if not grader_str:
+            print("  ❌ No grader field defined.")
+            continue
+        
+        print(f"  Attempting to resolve: {grader_str}")
+        try:
+            if ":" not in grader_str:
+                print(f"  ❌ Invalid format. Expected 'module:function'")
+                continue
+            
+            module_path, func_name = grader_str.split(":")
+            
+            module = None
+            # Try full path first (as it will be in the platform)
+            try:
+                module = importlib.import_module(module_path)
+            except ImportError:
+                # Local fallback: try without 'drone_env.' prefix if we're in the project root
+                if module_path.startswith("drone_env."):
+                    alt_path = module_path.replace("drone_env.", "", 1)
+                    try:
+                        module = importlib.import_module(alt_path)
+                    except ImportError as eb:
+                        print(f"  ❌ Module search failed (tried {module_path} and {alt_path}): {eb}")
+            
+            if module:
+                func = getattr(module, func_name, None)
+                if func and callable(func):
+                    print(f"  ✅ SUCCESS: Found {func_name} in {module.__name__}")
+                    valid_count += 1
+                else:
+                    print(f"  ❌ {func_name} not found or not callable in {module.__name__}")
+            else:
+                pass # Already printed error
+        except Exception as e:
+            print(f"  ❌ Error: {e}")
+
+    print(f"\nSummary: {valid_count} valid graders found.")
+    if valid_count >= 3:
+        print("🚀 LOCAL CHECK PASSED.")
+    else:
+        print("🚨 LOCAL CHECK FAILED.")
+
+if __name__ == "__main__":
+    check_graders()

core/__init__.py

@@ -1,4 +1,13 @@
-# drone_env.core package
-from .graders import grade_easy, grade_medium, grade_hard
+def grade_easy(state):
+    from drone_env.graders.easy import grade_easy as fn
+    return fn(state)
+
+def grade_medium(state):
+    from drone_env.graders.medium import grade_medium as fn
+    return fn(state)
+
+def grade_hard(state):
+    from drone_env.graders.hard import grade_hard as fn
+    return fn(state)
 
 __all__ = ["grade_easy", "grade_medium", "grade_hard"]

core/tasks.py

@@ -4,7 +4,7 @@ Mission configurations for Drone Delivery missions.
 """
 
 TASK_CONFIG = {
-    "drone_env.core.graders:grade_easy": {
+    "drone_env.graders.easy:grade_easy": {
         "width": 10,
         "height": 10,
         "n_buildings": 4,
@@ -24,7 +24,7 @@ TASK_CONFIG = {
         "r_wall": 0.10,
         "r_blocked": 0.10,
     },
-    "drone_env.core.graders:grade_medium": {
+    "drone_env.graders.medium:grade_medium": {
         "width": 14,
         "height": 14,
         "n_buildings": 8,
@@ -44,7 +44,7 @@ TASK_CONFIG = {
         "r_wall": 0.15,
         "r_blocked": 0.15,
     },
-    "drone_env.core.graders:grade_hard": {
+    "drone_env.graders.hard:grade_hard": {
         "width": 18,
         "height": 18,
         "n_buildings": 12,

graders/__init__.py

@@ -0,0 +1,11 @@
+from .easy import grade_easy
+from .medium import grade_medium
+from .hard import grade_hard
+
+GRADERS = {
+    "drone_env.graders.easy:grade_easy": grade_easy,
+    "drone_env.graders.medium:grade_medium": grade_medium,
+    "drone_env.graders.hard:grade_hard": grade_hard,
+}
+
+__all__ = ["grade_easy", "grade_medium", "grade_hard", "GRADERS"]

graders/easy.py

@@ -0,0 +1,40 @@
+"""
+Evaluation logic for easy_delivery task.
+This file is self-contained.
+"""
+
+def _get_attr(state, key, default=0):
+    """Safely get an attribute from a state object or dict."""
+    if isinstance(state, dict):
+        return state.get(key, default)
+    return getattr(state, key, default)
+
+
+def compute_grade(state, max_steps: float) -> float:
+    """
+    Unified grade calculation:
+    - 80% weighted by deliveries completed.
+    - 20% weighted by efficiency (remaining battery and steps).
+    """
+    deliveries_total = _get_attr(state, "deliveries_total", 0)
+    deliveries_done  = _get_attr(state, "deliveries_done",  0)
+    battery          = _get_attr(state, "battery",          1.0)
+    step_count       = _get_attr(state, "step_count",       0)
+
+    if deliveries_total == 0:
+        return 0.5
+
+    delivery_ratio = deliveries_done / deliveries_total
+    efficiency = float(battery) * 0.5 + (1.0 - (float(step_count) / float(max_steps))) * 0.5
+    efficiency = max(0.0, min(1.0, efficiency))
+
+    score = (delivery_ratio * 0.8) + (efficiency * 0.2)
+    if deliveries_done < deliveries_total:
+        score = min(score, 0.49)
+
+    return max(0.01, min(0.99, float(score)))
+
+
+def grade_easy(state) -> float:
+    """Grader for easy_delivery task (10x10 grid, 1 delivery, 60 max steps)."""
+    return compute_grade(state, 60.0)

graders/hard.py

@@ -0,0 +1,40 @@
+"""
+Evaluation logic for hard_delivery task.
+This file is self-contained.
+"""
+
+def _get_attr(state, key, default=0):
+    """Safely get an attribute from a state object or dict."""
+    if isinstance(state, dict):
+        return state.get(key, default)
+    return getattr(state, key, default)
+
+
+def compute_grade(state, max_steps: float) -> float:
+    """
+    Unified grade calculation:
+    - 80% weighted by deliveries completed.
+    - 20% weighted by efficiency (remaining battery and steps).
+    """
+    deliveries_total = _get_attr(state, "deliveries_total", 0)
+    deliveries_done  = _get_attr(state, "deliveries_done",  0)
+    battery          = _get_attr(state, "battery",          1.0)
+    step_count       = _get_attr(state, "step_count",       0)
+
+    if deliveries_total == 0:
+        return 0.5
+
+    delivery_ratio = deliveries_done / deliveries_total
+    efficiency = float(battery) * 0.5 + (1.0 - (float(step_count) / float(max_steps))) * 0.5
+    efficiency = max(0.0, min(1.0, efficiency))
+
+    score = (delivery_ratio * 0.8) + (efficiency * 0.2)
+    if deliveries_done < deliveries_total:
+        score = min(score, 0.49)
+
+    return max(0.01, min(0.99, float(score)))
+
+
+def grade_hard(state) -> float:
+    """Grader for hard_delivery task (18x18 grid, 5 deliveries, 160 max steps)."""
+    return compute_grade(state, 160.0)

core/graders.py → graders/medium.py

@@ -1,14 +1,8 @@
 """
-drone_delivery_env/core/graders.py
-Evaluation logic for scoring Drone Delivery missions.
-
-NOTE: This module is intentionally self-contained with NO external package imports.
-The OpenEnv validator imports grader functions directly via the module path
-(e.g. drone_env.core.graders:grade_easy). Any import failure here means the
-validator counts 0 working graders — causing "Not enough tasks with graders" error.
+Evaluation logic for medium_delivery task.
+This file is self-contained.
 """
 
-
 def _get_attr(state, key, default=0):
     """Safely get an attribute from a state object or dict."""
     if isinstance(state, dict):
@@ -21,7 +15,6 @@ def compute_grade(state, max_steps: float) -> float:
     Unified grade calculation:
     - 80% weighted by deliveries completed.
     - 20% weighted by efficiency (remaining battery and steps).
-    Accepts any object or dict with the required fields.
     """
     deliveries_total = _get_attr(state, "deliveries_total", 0)
     deliveries_done  = _get_attr(state, "deliveries_done",  0)
@@ -29,41 +22,19 @@ def compute_grade(state, max_steps: float) -> float:
     step_count       = _get_attr(state, "step_count",       0)
 
     if deliveries_total == 0:
-        return 0.5  # neutral score when no deliveries configured
+        return 0.5
 
     delivery_ratio = deliveries_done / deliveries_total
-
-    # Efficiency factor (0.0 to 1.0)
     efficiency = float(battery) * 0.5 + (1.0 - (float(step_count) / float(max_steps))) * 0.5
     efficiency = max(0.0, min(1.0, efficiency))
 
     score = (delivery_ratio * 0.8) + (efficiency * 0.2)
-
-    # Cap score if not all deliveries are done
     if deliveries_done < deliveries_total:
         score = min(score, 0.49)
 
-    # Hackathon Requirement: Score must be strictly between 0.0 and 1.0
     return max(0.01, min(0.99, float(score)))
 
 
-def grade_easy(state) -> float:
-    """Grader for easy_delivery task (10x10 grid, 1 delivery, 60 max steps)."""
-    return compute_grade(state, 60.0)
-
-
 def grade_medium(state) -> float:
     """Grader for medium_delivery task (14x14 grid, 3 deliveries, 100 max steps)."""
     return compute_grade(state, 100.0)
-
-
-def grade_hard(state) -> float:
-    """Grader for hard_delivery task (18x18 grid, 5 deliveries, 160 max steps)."""
-    return compute_grade(state, 160.0)
-
-
-GRADERS = {
-    "drone_env.core.graders:grade_easy":   grade_easy,
-    "drone_env.core.graders:grade_medium": grade_medium,
-    "drone_env.core.graders:grade_hard":   grade_hard,
-}

inference.py

@@ -44,7 +44,7 @@ API_KEY = os.getenv("HF_TOKEN") or os.getenv("API_KEY") or os.getenv("OPENAI_API
 API_BASE_URL = os.getenv("API_BASE_URL") or "https://router.huggingface.co/v1"
 MODEL_NAME = os.getenv("MODEL_NAME") or "Qwen/Qwen2.5-7B-Instruct"
 
-TASK_NAME = os.getenv("DRONE_TASK", "drone_env.core.graders:grade_easy")
+TASK_NAME = os.getenv("TASK_NAME", "drone_env.graders.easy:grade_easy")
 BENCHMARK = os.getenv("DRONE_BENCHMARK", "drone_env_v1")
 
 MAX_STEPS = 60

models.py

@@ -14,7 +14,7 @@ class DroneAction(BaseModel):
     )
     task_name: Optional[str] = Field(
         default=None,
-        description="Task to load on reset: drone_env.core.graders:grade_easy | medium | hard",
+        description="Task to load on reset: drone_env.graders.easy:grade_easy | medium | hard",
     )
 
 
@@ -43,7 +43,7 @@ class DroneObservation(BaseModel):
 
 class DroneState(BaseModel):
     episode_id: str = ""
-    task_name: str = "drone_env.core.graders:grade_easy"
+    task_name: str = "drone_env.graders.easy:grade_easy"
     step_count: int = 0
     done: bool = False
     reward_total: float = 0.0

openenv.yaml

@@ -6,8 +6,8 @@ app: server.app:app
 port: 8000
 tasks:
   - id: easy_delivery
-    grader: drone_env.core.graders:grade_easy
+    grader: drone_env.graders.easy:grade_easy
   - id: medium_delivery
-    grader: drone_env.core.graders:grade_medium
+    grader: drone_env.graders.medium:grade_medium
   - id: hard_delivery
-    grader: drone_env.core.graders:grade_hard
+    grader: drone_env.graders.hard:grade_hard

pyproject.toml

@@ -35,5 +35,5 @@ server = "drone_env.server.app:main"
 
 [tool.setuptools]
 include-package-data = true
-packages = ["drone_env", "drone_env.server", "drone_env.core", "drone_env.rl"]
-package-dir = { "drone_env" = ".", "drone_env.server" = "server", "drone_env.core" = "core", "drone_env.rl" = "rl" }
+packages = ["drone_env", "drone_env.server", "drone_env.graders", "drone_env.core", "drone_env.rl"]
+package-dir = { "drone_env" = ".", "drone_env.server" = "server", "drone_env.graders" = "graders", "drone_env.core" = "core", "drone_env.rl" = "rl" }

server/app.py

@@ -25,7 +25,7 @@ if str(BASE_DIR) not in sys.path:
 from drone_env.models import DroneAction, DroneObservation, DroneState
 from drone_env.server.grid_world_environment import DroneDeliveryEnvironment
 from drone_env.core.tasks import TASK_CONFIG
-from drone_env.core.graders import GRADERS
+from drone_env.graders import GRADERS
 from drone_env.rl.trainer import PathLearner, get_action_from_policy
 
 app = FastAPI(
@@ -88,9 +88,9 @@ async def metadata():
         "description": "SkyRelic Drone Delivery reinforcement-learning environment with easy/medium/hard delivery tasks on a grid world.",
         "version": "0.2.1",
         "tasks": [
-            {"id": "easy_delivery",  "grader": "drone_env.core.graders:grade_easy"},
-            {"id": "medium_delivery", "grader": "drone_env.core.graders:grade_medium"},
-            {"id": "hard_delivery",  "grader": "drone_env.core.graders:grade_hard"},
+            {"id": "easy_delivery",  "grader": "drone_env.graders.easy:grade_easy"},
+            {"id": "medium_delivery", "grader": "drone_env.graders.medium:grade_medium"},
+            {"id": "hard_delivery",  "grader": "drone_env.graders.hard:grade_hard"},
         ],
     }
 
@@ -144,9 +144,9 @@ async def get_state():
 
 # Task ID → grader key mapping (matches openenv.yaml task ids)
 TASK_ID_TO_GRADER = {
-    "easy_delivery":  "drone_env.core.graders:grade_easy",
-    "medium_delivery": "drone_env.core.graders:grade_medium",
-    "hard_delivery":  "drone_env.core.graders:grade_hard",
+    "easy_delivery":  "drone_env.graders.easy:grade_easy",
+    "medium_delivery": "drone_env.graders.medium:grade_medium",
+    "hard_delivery":  "drone_env.graders.hard:grade_hard",
 }
 
 @app.get("/grade/{task_name}")
@@ -207,7 +207,7 @@ async def get_terminal_logs():
 @app.get("/rewards")
 async def get_rewards():
     """Return the reward configuration for the current task."""
-    task_name = _env.state.task_name or "drone_env.core.graders:grade_easy"
+    task_name = _env.state.task_name or "drone_env.graders.easy:grade_easy"
     config = TASK_CONFIG.get(task_name, {})
     # Filter only reward keys
     rewards = {k: v for k, v in config.items() if k.startswith("r_")}
@@ -248,7 +248,7 @@ async def get_memory_logs():
 
 @app.post("/predict")
 async def predict(obs: DroneObservation):
-    task_name = _env.state.task_name or "drone_env.core.graders:grade_easy"
+    task_name = _env.state.task_name or "drone_env.graders.easy:grade_easy"
     action_str = get_action_from_policy(obs, task_name)
     return {"direction": action_str}
 

server/grid_world_environment.py

@@ -24,7 +24,7 @@ from drone_env.core.tasks import TASK_CONFIG
 from drone_env.core.drone import compute_next_pos, drain_battery
 from drone_env.core.obstacles import check_move
 from drone_env.core.state_manager import new_episode_state
-from drone_env.core.graders import GRADERS
+from drone_env.graders import GRADERS
 from drone_env.rl.trainer import record_episode
 
 
@@ -33,7 +33,7 @@ class DroneDeliveryEnvironment(Environment):
     def __init__(self):
         super().__init__()
         self._state = DroneState()
-        self._cfg = TASK_CONFIG["drone_env.core.graders:grade_easy"]
+        self._cfg = TASK_CONFIG["drone_env.graders.easy:grade_easy"]
         self._grid: List[List[str]] = []
         self._deliveries: List[Tuple[int, int]] = []
         self._delivered: List[bool] = []
@@ -51,11 +51,11 @@ class DroneDeliveryEnvironment(Environment):
 
         # Map short task IDs (from openenv.yaml) to full grader keys
         _TASK_ID_MAP = {
-            "easy_delivery":   "drone_env.core.graders:grade_easy",
-            "medium_delivery": "drone_env.core.graders:grade_medium",
-            "hard_delivery":   "drone_env.core.graders:grade_hard",
+            "easy_delivery":   "drone_env.graders.easy:grade_easy",
+            "medium_delivery": "drone_env.graders.medium:grade_medium",
+            "hard_delivery":   "drone_env.graders.hard:grade_hard",
         }
-        task = "drone_env.core.graders:grade_easy"
+        task = "drone_env.graders.easy:grade_easy"
         if action and action.task_name:
             name = action.task_name
             if name in _TASK_ID_MAP:

server/static/index.html

@@ -1230,17 +1230,17 @@
           <div class="panel-inner">
             <div class="panel-title">Mission Select</div>
             <div class="task-group" id="taskGroup">
-              <button class="task-btn active" data-task="drone_env.core.graders:grade_easy">
+              <button class="task-btn active" data-task="drone_env.graders.easy:grade_easy">
                 <span class="dot"></span>
                 Easy Delivery
                 <span class="task-tag">10×10</span>
               </button>
-              <button class="task-btn" data-task="drone_env.core.graders:grade_medium">
+              <button class="task-btn" data-task="drone_env.graders.medium:grade_medium">
                 <span class="dot"></span>
                 Medium Delivery
                 <span class="task-tag">14×14</span>
               </button>
-              <button class="task-btn" data-task="drone_env.core.graders:grade_hard">
+              <button class="task-btn" data-task="drone_env.graders.hard:grade_hard">
                 <span class="dot"></span>
                 Hard Delivery
                 <span class="task-tag">18×18</span>

server/static/script.js

@@ -17,7 +17,7 @@ const EMOJI = {
 // ═══════════════════════════════════════════════════════
 //  STATE
 // ═══════════════════════════════════════════════════════
-let currentTask = 'drone_env.core.graders:grade_easy'; 
+let currentTask = 'drone_env.graders.easy:grade_easy'; 
 let autoTimer   = null;
 let logTimer    = null;
 let obs         = null;
@@ -601,12 +601,12 @@ function startNextTask() {
     closeCompletionModal();
     
     const sequence = {
-        'drone_env.core.graders:grade_easy': 'drone_env.core.graders:grade_medium',
-        'drone_env.core.graders:grade_medium': 'drone_env.core.graders:grade_hard',
-        'drone_env.core.graders:grade_hard': 'drone_env.core.graders:grade_easy'
+        'drone_env.graders.easy:grade_easy': 'drone_env.graders.medium:grade_medium',
+        'drone_env.graders.medium:grade_medium': 'drone_env.graders.hard:grade_hard',
+        'drone_env.graders.hard:grade_hard': 'drone_env.graders.easy:grade_easy'
     };
     
-    const nextTask = sequence[currentTask] || 'drone_env.core.graders:grade_easy';
+    const nextTask = sequence[currentTask] || 'drone_env.graders.easy:grade_easy';
     currentTask = nextTask;
     
     // Update active state on buttons
@@ -627,7 +627,7 @@ async function updateMissionLegend() {
         if (!container || !data.tasks) return;
         
         // Ensure tasks are sorted Easy, Medium, Hard
-        const order = ['drone_env.core.graders:grade_easy', 'drone_env.core.graders:grade_medium', 'drone_env.core.graders:grade_hard'];
+        const order = ['drone_env.graders.easy:grade_easy', 'drone_env.graders.medium:grade_medium', 'drone_env.graders.hard:grade_hard'];
         const tasks = data.tasks.sort((a, b) => order.indexOf(a.name) - order.indexOf(b.name));
         
         container.innerHTML = `

tests/test_env.py

@@ -12,14 +12,14 @@ from drone_env.core.grid_generator import generate_city_map
 from drone_env.core.tasks import TASK_CONFIG
 from drone_env.core.obstacles import check_move
 from drone_env.core.drone import compute_next_pos, drain_battery
-from drone_env.core.graders import grade_easy, grade_medium, grade_hard
+from drone_env.graders import grade_easy, grade_medium, grade_hard
 from drone_env.models import DroneState
 
 
 # ── Grid generator ────────────────────────────────────────────────────────────
 
 def test_grid_shape():
-    cfg = TASK_CONFIG["easy_delivery"]
+    cfg = TASK_CONFIG["drone_env.graders.easy:grade_easy"]
     rng = torch.Generator(); rng.manual_seed(42)
     grid, deliveries, start = generate_city_map(cfg, rng)
     assert len(grid) == cfg["height"]
@@ -30,7 +30,7 @@ def test_grid_shape():
 
 
 def test_start_on_road():
-    cfg = TASK_CONFIG["easy_delivery"]
+    cfg = TASK_CONFIG["drone_env.graders.easy:grade_easy"]
     rng = torch.Generator(); rng.manual_seed(7)
     grid, _, start = generate_city_map(cfg, rng)
     sx, sy = start
@@ -38,7 +38,7 @@ def test_start_on_road():
 
 
 def test_delivery_cells_present():
-    cfg = TASK_CONFIG["medium_delivery"]
+    cfg = TASK_CONFIG["drone_env.graders.medium:grade_medium"]
     rng = torch.Generator(); rng.manual_seed(99)
     grid, deliveries, _ = generate_city_map(cfg, rng)
     for dx, dy in deliveries:
@@ -129,9 +129,9 @@ def test_grade_medium_hard_bounds():
 # ── Task configs ──────────────────────────────────────────────────────────────
 
 def test_all_tasks_present():
-    assert "easy_delivery" in TASK_CONFIG
-    assert "medium_delivery" in TASK_CONFIG
-    assert "hard_delivery" in TASK_CONFIG
+    assert "drone_env.graders.easy:grade_easy" in TASK_CONFIG
+    assert "drone_env.graders.medium:grade_medium" in TASK_CONFIG
+    assert "drone_env.graders.hard:grade_hard" in TASK_CONFIG
 
 def test_task_has_required_keys():
     required = ["width","height","n_buildings","n_trees","n_obstacles",
@@ -142,9 +142,9 @@ def test_task_has_required_keys():
             assert key in cfg, f"{task} missing key: {key}"
 
 def test_task_difficulty_ordering():
-    easy   = TASK_CONFIG["easy_delivery"]
-    medium = TASK_CONFIG["medium_delivery"]
-    hard   = TASK_CONFIG["hard_delivery"]
+    easy   = TASK_CONFIG["drone_env.graders.easy:grade_easy"]
+    medium = TASK_CONFIG["drone_env.graders.medium:grade_medium"]
+    hard   = TASK_CONFIG["drone_env.graders.hard:grade_hard"]
     assert easy["width"] < medium["width"] < hard["width"]
     assert easy["n_deliveries"] < medium["n_deliveries"] < hard["n_deliveries"]
     assert easy["max_steps"] < medium["max_steps"] < hard["max_steps"]