fix: always run all 3 task levels regardless of TASK_LEVEL env var

Root cause: Dockerfile had ENV TASK_LEVEL=easy which caused inference.py
to only run the easy task — validator saw 1 task, needs 3.

- Dockerfile: remove ENV TASK_LEVEL=easy entirely
- inference.py: always run [easy, medium, hard] unconditionally;
TASK_LEVEL env var only affects the uvicorn server, not inference

Dockerfile

@@ -23,7 +23,8 @@ ENV PYTHONPATH=/app:/app/src
 ENV ENABLE_WEB_INTERFACE=true
 
 # ── Runtime config ────────────────────────────────────────────
-ENV TASK_LEVEL=easy
+# Do NOT set TASK_LEVEL — inference.py must run all 3 levels for the validator.
+# The uvicorn server reads TASK_LEVEL at request time; leaving it unset is fine.
 ENV PORT=7860
 
 # ── Install only what the base image does NOT already provide ─

inference.py

@@ -217,15 +217,9 @@ async def main() -> None:
 
     client = OpenAI(api_key=api_key, base_url=api_base_url)
 
-    # Determine which task levels to run.
-    # If TASK_LEVEL is set to a specific level, run only that one;
-    # otherwise run ALL THREE so the validator sees 3 graded tasks.
-    requested = os.environ.get("TASK_LEVEL", "").strip().lower()
-    if requested in ("easy", "medium", "hard"):
-        task_levels = [requested]
-    else:
-        task_levels = ["easy", "medium", "hard"]
-
+    # Always run all three task levels — the validator requires at least 3 graded tasks.
+    # TASK_LEVEL env var is intentionally ignored here; it only affects the uvicorn server.
+    task_levels = ["easy", "medium", "hard"]
     print(f"[DEBUG] Running task levels: {task_levels}", flush=True)
 
     if LOCAL_IMAGE_NAME: