PLEASE SCALER

customer_support_environment.py

@@ -5,11 +5,12 @@ from typing import Any, Optional
 from uuid import uuid4
 
 from openenv.core.env_server.interfaces import Environment
+from openenv.core.env_server.types import EnvironmentMetadata
 
 from data_loader import build_expected, load_dataset, split_difficulty
 from kb import build_knowledge_base
 from models import SupportAction, SupportObservation, SupportState
-from tasks import grade_task, task_for_difficulty
+from tasks import TASKS_WITH_GRADERS, grade_task, task_for_difficulty
 
 
 class CustomerSupportEnvironment(Environment[SupportAction, SupportObservation, SupportState]):
@@ -167,6 +168,23 @@ class CustomerSupportEnvironment(Environment[SupportAction, SupportObservation,
     def state(self) -> SupportState:
         return self._state
 
+    def get_metadata(self) -> EnvironmentMetadata:
+        task_lines = "\n".join(
+            f"- {t['task_id']} | difficulty={t['difficulty']} | grader={t['grader']}"
+            for t in TASKS_WITH_GRADERS
+        )
+        readme = (
+            "Customer support benchmark with deterministic task graders.\n\n"
+            "Tasks with graders:\n"
+            f"{task_lines}\n"
+        )
+        return EnvironmentMetadata(
+            name="customer_support_benchmark",
+            description="Customer support simulation with 3 graded tasks.",
+            version="1.0.0",
+            readme_content=readme,
+        )
+
     def _make_observation(self, reward: float, done: bool, feedback: str) -> SupportObservation:
         if self._expected is None:
             raise RuntimeError("Environment is not initialized")
@@ -182,6 +200,8 @@ class CustomerSupportEnvironment(Environment[SupportAction, SupportObservation,
             reward=reward,
             feedback=feedback,
             metadata={
+                "tasks_with_graders": TASKS_WITH_GRADERS,
+                "grader": task_for_difficulty(self._expected["difficulty"]).grader_name,
                 "expected_category": self._expected["expected_category"],
                 "expected_action": self._expected["expected_action"],
                 "expected_response": self._expected["expected_response"],

tasks.py

@@ -13,7 +13,7 @@ class TaskSpec:
     grader_name: str
 
 
-TASKS: Dict[str, TaskSpec] = {
+TASK_MAP: Dict[str, TaskSpec] = {
     "easy_classify_respond": TaskSpec(
         task_id="easy_classify_respond",
         difficulty="easy",
@@ -41,17 +41,21 @@ TASKS_WITH_GRADERS: List[dict] = [
         "difficulty": task.difficulty,
         "objective": task.objective,
         "grader": task.grader_name,
+        "grader_name": task.grader_name,
     }
-    for task in TASKS.values()
+    for task in TASK_MAP.values()
 ]
 
+# Additional compatibility export used by some validators.
+TASKS: List[dict] = list(TASKS_WITH_GRADERS)
+
 
 def task_for_difficulty(difficulty: str) -> TaskSpec:
     normalized = difficulty.strip().lower()
-    for task in TASKS.values():
+    for task in TASK_MAP.values():
         if task.difficulty == normalized:
             return task
-    return TASKS["easy_classify_respond"]
+    return TASK_MAP["easy_classify_respond"]
 
 
 def grade_easy(expected: dict, history: List[dict]) -> float:
@@ -91,7 +95,7 @@ GRADERS: Dict[str, Callable[[dict, List[dict]], float]] = {
 
 
 def grade_task(task_id: str, expected: dict, history: List[dict]) -> float:
-    task = TASKS[task_id]
+    task = TASK_MAP[task_id]
     grader = GRADERS[task.grader_name]
     return grader(expected, history)