Fix OpenEnv metadata, docker data, tasks endpoint, and demo

README.md

@@ -1,3 +1,17 @@
+---
+title: 911 Dispatch Supervisor
+emoji: 🚨
+colorFrom: red
+colorTo: orange
+sdk: docker
+pinned: false
+tags:
+	- openenv
+	- reinforcement-learning
+	- llm-agent
+	- emergency-dispatch
+---
+
 # 911 City-Wide Emergency Dispatch Supervisor
 
 **LLM-powered 911 dispatch supervision — city scale**
@@ -106,6 +120,33 @@ python demo.py
 python inference.py
 ```
 
+## Reward Function
+
+The reward signal is a weighted combination of five components:
+
+| Component | Weight | Description |
+|-----------|--------|-------------|
+| `response_time` | 30% | How quickly units reach incidents relative to severity benchmarks |
+| `triage` | 25% | Whether the dispatched unit type matches incident requirements |
+| `survival` | 25% | Whether Priority-1 incidents are resolved before survival clock expires |
+| `coverage` | 12% | Geographic distribution of available units across city districts |
+| `protocol` | 8% | Whether the dispatch action was legally valid |
+
+**Safety gate:** If any Priority-1 incident was seen and `survival=0.0`, the total episode score is capped at `0.2` regardless of other components.
+
+## Baseline Scores
+
+Scores from the random baseline agent (`USE_RANDOM=true`):
+
+| Task | Difficulty | Baseline Score |
+|------|-----------|---------------|
+| `single_incident` | Easy | ~0.55 |
+| `multi_incident` | Medium | ~0.48 |
+| `mass_casualty` | Hard | ~0.32 |
+| `shift_surge` | Hard | ~0.38 |
+
+*Run `USE_RANDOM=true python inference.py` to reproduce.*
+
 ## Project Structure
 
 ```

demo.py

@@ -1,9 +1,9 @@
 #!/usr/bin/env python3
 """Demo script showing the 911 dispatch supervisor environment in action.
 
-This non-interactive demo runs a deterministic episode using OpenEnvEnvironment
-directly (no LLM/API server required). It performs a scripted triage sequence
-on the multi-incident task.
+This non-interactive demo runs an episode using OpenEnvEnvironment directly
+(no LLM/API server required). It uses `legal_actions()` so it is seed/task
+independent.
 """
 
 import asyncio
@@ -47,14 +47,12 @@ async def run_demo_episode(
         rewards = []
         errors = []
 
-        scripted_actions = [
-            Action(action_type=DispatchAction.DISPATCH, unit_id="MED-1", incident_id="INC-002"),
-            Action(action_type=DispatchAction.DISPATCH, unit_id="ENG-1", incident_id="INC-001"),
-            Action(action_type=DispatchAction.DISPATCH, unit_id="LAD-1", incident_id="INC-001"),
-            Action(action_type=DispatchAction.DISPATCH, unit_id="PAT-1", incident_id="INC-003"),
-        ]
-
-        for action in scripted_actions:
+        # Step through the environment using only legal actions.
+        while step_count < max_steps:
+            legal = env.legal_actions()
+            if not legal:
+                break
+            action = legal[0]
             step_count += 1
             try:
                 obs, reward, done = await env.step(action)
@@ -73,19 +71,6 @@ async def run_demo_episode(
                 print(f"[STEP {step_count}] ERROR: {e}")
                 break
 
-        # Continue stepping with any legal actions until done/max_steps.
-        while step_count < max_steps:
-            legal = env.legal_actions()
-            if not legal:
-                break
-            action = legal[0]
-            step_count += 1
-            obs, reward, done = await env.step(action)
-            rewards.append(reward)
-            total_reward += reward
-            if done:
-                break
-
         # Final state
         final_state = env.state()
 
@@ -102,6 +87,15 @@ async def run_demo_episode(
         print(f"Final Score:   {final_score:.4f}")
         print(f"Active incidents: {sum(1 for i in final_state.incidents.values() if i.status.value != 'RESOLVED')}")
 
+        print("\n" + "─" * 60)
+        print(f"{'Incident':<12} {'Type':<22} {'Severity':<12} {'Status':<12}")
+        print("─" * 60)
+        for inc in sorted(final_state.incidents.values(), key=lambda i: i.incident_id):
+            print(
+                f"{inc.incident_id:<12} {inc.incident_type.value:<22} {inc.severity.value:<12} {inc.status.value:<12}"
+            )
+        print("─" * 60)
+
         if errors:
             print(f"\nErrors encountered: {len(errors)}")
             for err in errors:

inference.py

@@ -304,7 +304,7 @@ async def main() -> int:
             hf_token = os.environ.get("HF_TOKEN", "")
             agent = LLMAgent(api_key=hf_token, base_url=api_base_url, model=model_name)
 
-        task_ids = ["single_incident", "multi_incident", "mass_casualty"]
+        task_ids = ["single_incident", "multi_incident", "mass_casualty", "shift_surge"]
 
         for task_id in task_ids:
             await run_episode(task_id, model_name, agent)

openenv.yaml

@@ -1,17 +1,20 @@
 name: citywide-dispatch-supervisor
 version: "0.1.0"
-description: 911 dispatch supervisor RL environment powered by LLM orchestration.
+description: >
+	City-wide 911 emergency dispatch supervisor RL environment.
+	An LLM agent learns to manage simultaneous incidents by dispatching
+	police, fire, and EMS units across a city grid under realistic constraints.
 entrypoint: src.openenv_environment:OpenEnvEnvironment
 tasks:
 	- id: single_incident
-		name: Single Incident
-		description: One incident with a small unit pool; learn basic dispatch.
+		name: Single Incident Response
+		description: One incident with a small unit pool; learn basic dispatch, correct unit type, and response time.
 	- id: multi_incident
-		name: Multi Incident
-		description: Multiple concurrent incidents; triage and prioritization under constraints.
+		name: Simultaneous Multi-Incident
+		description: Multiple concurrent incidents requiring triage, prioritization, and correct unit matching.
 	- id: mass_casualty
-		name: Mass Casualty
-		description: Priority-1 surge; maximize survival and response time.
+		name: Mass Casualty Event
+		description: Wave-based Priority-1 surge with resource conflict; maximize survival outcomes.
 	- id: shift_surge
 		name: Shift Surge
-		description: Incident waves plus unit status changes; maintain coverage over time.
+		description: Incident waves combined with units going out of service; maintain coverage over time.

src/server/Dockerfile

@@ -6,6 +6,7 @@ COPY src/server/requirements.txt .
 RUN pip install -r requirements.txt
 
 COPY src/ /app/src/
+COPY data/ /app/data/
 
 EXPOSE 8000
 

src/server/app.py

@@ -48,6 +48,22 @@ async def health() -> dict[str, str]:
     return {"status": "ok"}
 
 
+@app.get("/tasks")
+async def list_tasks() -> list[dict[str, str]]:
+    """List all available tasks."""
+    from src.tasks.registry import TaskRegistry
+
+    return [
+        {
+            "task_id": t.task_id,
+            "name": t.name,
+            "description": t.description,
+            "difficulty": t.difficulty,
+        }
+        for t in TaskRegistry.list_tasks()
+    ]
+
+
 @app.post("/reset")
 async def reset(request: ResetRequest) -> dict[str, Any]:
     global _env

src/visualizer/__init__.py

@@ -1 +1 @@
-"""Visualizer package for 2D airport state viewer."""
+"""Visualizer package for 2D city dispatch state viewer."""

tests/smoke_docker.py

@@ -63,7 +63,7 @@ def test_reset() -> None:
     print("Testing /reset endpoint...")
     response = requests.post(
         f"{HOST}/reset",
-        json={"task_id": "arrival", "seed": 42},
+        json={"task_id": "single_incident", "seed": 42},
         timeout=10,
     )
     assert response.status_code == 200, f"Expected 200, got {response.status_code}"

tests/test_inference.py

@@ -9,7 +9,7 @@ import sys
 
 
 class TestInferenceFormatCompliance:
-    TASK_IDS = ["single_incident", "multi_incident", "mass_casualty"]
+    TASK_IDS = ["single_incident", "multi_incident", "mass_casualty", "shift_surge"]
 
     def _run_inference_capture(self, env: dict[str, str]) -> tuple[int, str, str]:
         cmd = [sys.executable, "inference.py"]