Keep reset scores inside the open interval

Author: PathikSingh

server/environment.py

@@ -85,7 +85,7 @@ def reset(
         self._raw_total_reward = 0.0
         self._world = self._build_world(self._task)
 
-        return self._make_observation(reward=0.0, done=False)
+        return self._make_observation(reward=clamp_strict_score(0.0), done=False)
 
     def step(
         self,

tests/test_environment.py

@@ -41,7 +41,7 @@ def test_reset_returns_valid_observation():
     obs = env.reset(task_id="rollback_alpha")
     assert obs.task_id == "rollback_alpha"
     assert obs.available_tools
-    assert obs.reward == 0.0
+    assert 0.0 < obs.reward < 1.0
     assert obs.done is False