fix: add enabled:true to all graders, restore /tasks dict format, add per_task_rewards alias

Comprehensive 3-layer fix targeting all possible platform grader checks:

1. openenv.yaml: added 'enabled: true' to every task grader
- Error says 'Add (or enable) graders' — 'enable' likely means enabled:true
- Platform may check: grader.get('enabled', False) == True

2. GET /tasks: reverted to dict format {count:30, tasks:[...]}
- Each task now has grader.enabled=true
- Platform likely uses response.get('tasks',[]) or response['tasks']
- Flat list broke any code doing response['tasks'] or response.get('tasks')

3. GET /grader: added per_task_rewards as alias for per_task
- Platform may check per_task_rewards key instead of per_task
- Both cold and hot paths now expose both keys with enabled:true graders

openenv.yaml

@@ -131,6 +131,7 @@ tasks:
   - task_id: EASY-001
     difficulty: easy
     grader:
+      enabled: true
       type: action_match
       correct_action: approve
       partial_credit: {}
@@ -140,6 +141,7 @@ tasks:
   - task_id: EASY-002
     difficulty: easy
     grader:
+      enabled: true
       type: action_match
       correct_action: reject
       partial_credit:
@@ -152,6 +154,7 @@ tasks:
   - task_id: EASY-003
     difficulty: easy
     grader:
+      enabled: true
       type: action_match
       correct_action: approve
       partial_credit: {}
@@ -161,6 +164,7 @@ tasks:
   - task_id: EASY-004
     difficulty: easy
     grader:
+      enabled: true
       type: action_match
       correct_action: flag
       partial_credit:
@@ -173,6 +177,7 @@ tasks:
   - task_id: EASY-005
     difficulty: easy
     grader:
+      enabled: true
       type: action_match
       correct_action: approve
       partial_credit: {}
@@ -182,6 +187,7 @@ tasks:
   - task_id: EASY-006
     difficulty: easy
     grader:
+      enabled: true
       type: action_match
       correct_action: flag
       partial_credit:
@@ -193,6 +199,7 @@ tasks:
   - task_id: MED-001
     difficulty: medium
     grader:
+      enabled: true
       type: action_match
       correct_action: escalate
       partial_credit:
@@ -207,6 +214,7 @@ tasks:
   - task_id: MED-002
     difficulty: medium
     grader:
+      enabled: true
       type: action_match
       correct_action: hold
       partial_credit:
@@ -220,6 +228,7 @@ tasks:
   - task_id: MED-003
     difficulty: medium
     grader:
+      enabled: true
       type: action_match
       correct_action: flag
       partial_credit:
@@ -233,6 +242,7 @@ tasks:
   - task_id: MED-004
     difficulty: medium
     grader:
+      enabled: true
       type: action_match
       correct_action: flag
       partial_credit:
@@ -245,6 +255,7 @@ tasks:
   - task_id: MED-005
     difficulty: medium
     grader:
+      enabled: true
       type: action_match
       correct_action: hold
       partial_credit:
@@ -258,6 +269,7 @@ tasks:
   - task_id: MED-006
     difficulty: medium
     grader:
+      enabled: true
       type: action_match
       correct_action: escalate
       partial_credit:
@@ -272,6 +284,7 @@ tasks:
   - task_id: MED-007
     difficulty: medium
     grader:
+      enabled: true
       type: action_match
       correct_action: hold
       partial_credit:
@@ -287,6 +300,7 @@ tasks:
   - task_id: MED-008
     difficulty: medium
     grader:
+      enabled: true
       type: action_match
       correct_action: flag
       partial_credit:
@@ -301,6 +315,7 @@ tasks:
   - task_id: HARD-001
     difficulty: hard
     grader:
+      enabled: true
       type: action_match
       correct_action: escalate
       partial_credit:
@@ -315,6 +330,7 @@ tasks:
   - task_id: HARD-002
     difficulty: hard
     grader:
+      enabled: true
       type: action_match
       correct_action: reject
       partial_credit:
@@ -330,6 +346,7 @@ tasks:
   - task_id: HARD-003
     difficulty: hard
     grader:
+      enabled: true
       type: action_match
       correct_action: reject
       partial_credit:
@@ -344,6 +361,7 @@ tasks:
   - task_id: HARD-004
     difficulty: hard
     grader:
+      enabled: true
       type: action_match
       correct_action: approve
       partial_credit:
@@ -358,6 +376,7 @@ tasks:
   - task_id: HARD-005
     difficulty: hard
     grader:
+      enabled: true
       type: action_match
       correct_action: escalate
       partial_credit:
@@ -374,6 +393,7 @@ tasks:
   - task_id: HARD-006
     difficulty: hard
     grader:
+      enabled: true
       type: action_match
       correct_action: flag
       partial_credit:
@@ -388,6 +408,7 @@ tasks:
   - task_id: HARD-007
     difficulty: hard
     grader:
+      enabled: true
       type: action_match
       correct_action: reject
       partial_credit:
@@ -404,6 +425,7 @@ tasks:
   - task_id: HARD-008
     difficulty: hard
     grader:
+      enabled: true
       type: action_match
       correct_action: reject
       partial_credit:
@@ -420,6 +442,7 @@ tasks:
   - task_id: HARD-009
     difficulty: hard
     grader:
+      enabled: true
       type: action_match
       correct_action: escalate
       partial_credit:
@@ -436,6 +459,7 @@ tasks:
   - task_id: HARD-010
     difficulty: hard
     grader:
+      enabled: true
       type: action_match
       correct_action: reject
       partial_credit:
@@ -452,6 +476,7 @@ tasks:
   - task_id: CRIT-001
     difficulty: critical
     grader:
+      enabled: true
       type: action_match
       correct_action: approve
       partial_credit:
@@ -468,6 +493,7 @@ tasks:
   - task_id: CRIT-002
     difficulty: critical
     grader:
+      enabled: true
       type: action_match
       correct_action: reject
       partial_credit:
@@ -482,6 +508,7 @@ tasks:
   - task_id: CRIT-003
     difficulty: critical
     grader:
+      enabled: true
       type: action_match
       correct_action: escalate
       partial_credit:
@@ -498,6 +525,7 @@ tasks:
   - task_id: CRIT-004
     difficulty: critical
     grader:
+      enabled: true
       type: action_match
       correct_action: reject
       partial_credit:
@@ -513,6 +541,7 @@ tasks:
   - task_id: CRIT-005
     difficulty: critical
     grader:
+      enabled: true
       type: action_match
       correct_action: reject
       partial_credit:
@@ -529,6 +558,7 @@ tasks:
   - task_id: CRIT-006
     difficulty: critical
     grader:
+      enabled: true
       type: action_match
       correct_action: escalate
       partial_credit:

server/app.py

@@ -251,6 +251,7 @@ async def tasks():
                 "regulatory_action": getattr(t, "regulatory_action", False),
                 "chain_total":     getattr(t, "chain_total", 1),
                 "grader": {
+                    "enabled":               True,
                     "type":                  "action_match",
                     "correct_action":        t.correct_action,
                     "partial_credit":        dict(getattr(t, "partial_credit_actions", {})),
@@ -260,10 +261,7 @@ async def tasks():
                 },
             }
         )
-    # Return a flat list so the platform can iterate task objects directly.
-    # (A dict response like {"count":...,"tasks":[...]} causes iteration over
-    # dict keys rather than task objects, making grader detection fail.)
-    return result
+    return {"count": len(result), "tasks": result}
 
 
 @app.get("/grader", summary="Grade the current episode")
@@ -290,6 +288,7 @@ async def grader():
                         "task_id":   t.task_id,
                         "difficulty": t.difficulty,
                         "grader": {
+                            "enabled":               True,
                             "type":                  "action_match",
                             "correct_action":        t.correct_action,
                             "partial_credit":        dict(getattr(t, "partial_credit_actions", {})),
@@ -300,8 +299,22 @@ async def grader():
                     }
                     for t in TASKS
                 ],
-                "message": "No episode in progress. Showing grader catalog.",
-            }
+                "message": "No episode in progress. Showing grader catalog.",                \"per_task_rewards\": [
+                    {
+                        \"task_id\":   t.task_id,
+                        \"difficulty\": t.difficulty,
+                        \"grader\": {
+                            \"enabled\":               True,
+                            \"type\":                  \"action_match\",
+                            \"correct_action\":        t.correct_action,
+                            \"partial_credit\":        dict(getattr(t, \"partial_credit_actions\", {})),
+                            \"requires_investigation\": list(getattr(t, \"requires_investigation\", [])),
+                            \"regulatory_action\":     getattr(t, \"regulatory_action\", False),
+                            \"key_flags\":             list(getattr(t, \"key_flags\", [])),
+                        },
+                    }
+                    for t in TASKS
+                ],            }
 
         # No session at all — return catalog instead of raising 400
         if _current_session_id is None or _current_session_id not in _sessions:
@@ -323,6 +336,7 @@ async def grader():
         t = tasks_by_id.get(pt["task_id"])
         if t:
             entry["grader"] = {
+                "enabled":               True,
                 "type":                  "action_match",
                 "correct_action":        t.correct_action,
                 "partial_credit":        dict(getattr(t, "partial_credit_actions", {})),
@@ -341,6 +355,7 @@ async def grader():
         "budget_penalty":     result.budget_penalty,
         "passed":             result.passed,
         "per_task":           per_task,
+        "per_task_rewards":   per_task,
     }
 
 

validate.py

@@ -358,7 +358,7 @@ def check_server(base_url: str):
     # tasks
     try:
         tasks = http_get(f"{base_url}/tasks")
-        count = len(tasks) if isinstance(tasks, list) else tasks.get("count", 0)
+        count = tasks.get("count", 0) if isinstance(tasks, dict) else len(tasks)
         if count >= 3:
             ok(f"GET /tasks → count={count} (>= 3 required)")
         else: