feat: update inference, models, and server logic

inference.py

@@ -229,7 +229,15 @@ def main():
     scores.append(run_task(env, "likert",        "likert-scoring"))
     scores.append(run_task(env, "consistency",   "consistency-ranking"))
 
-    print(f"\nOverall avg: {sum(scores) / len(scores):.2f}", flush=True)
+    if scores:
+        print(f"\nOverall avg: {sum(scores) / len(scores):.2f}", flush=True)
+
+    if len(scores) >= 3:
+        print("\n=== CURRICULUM LEARNING DEMO ===")
+        print(f"Task 1 Pairwise (Easy):         {scores[0]:.2f}")
+        print(f"Task 2 Likert (Medium):         {scores[1]:.2f}")  
+        print(f"Task 3 Consistency (Hard):      {scores[2]:.2f}")
+        print(f"Difficulty progression: {scores[0]:.2f} → {scores[1]:.2f} → {scores[2]:.2f}")
 
 
 if __name__ == "__main__":

models.py

@@ -32,9 +32,8 @@ class PairwiseAction(Action):
         description="Optional reasoning for the choice (not used for grading).",
     )
     confidence: float = Field(
-        default=0.8,
-        ge=0.0, le=1.0,
-        description="Annotator confidence in this choice (0.0-1.0)"
+        default=0.8, ge=0.0, le=1.0,
+        description="Annotator confidence (0.0-1.0)"
     )
 
 

server/app.py

@@ -56,6 +56,38 @@ if not ENABLE_WEB_INTERFACE:
         max_concurrent_envs=MAX_CONCURRENT_ENVS,
     )
 
+from collections import defaultdict
+from threading import Lock
+from pydantic import BaseModel, Field
+
+leaderboard = defaultdict(list)
+leaderboard_lock = Lock()
+
+class LeaderboardEntry(BaseModel):
+    model: str = Field(..., min_length=1, max_length=255)
+    score: float = Field(..., ge=0.0, le=1.0)
+
+@app.get("/leaderboard")
+def get_leaderboard():
+    with leaderboard_lock:
+        return {
+            model: {
+                "avg_score": sum(scores)/len(scores) if scores else 0,
+                "runs": len(scores),
+                "scores": scores[-50:]  # Limit returned scores to last 50
+            }
+            for model, scores in leaderboard.items()
+        }
+
+@app.post("/leaderboard/submit")
+def submit_score(entry: LeaderboardEntry):
+    with leaderboard_lock:
+        leaderboard[entry.model].append(entry.score)
+        # Limit stored scores to prevent memory issues
+        if len(leaderboard[entry.model]) > 1000:
+            leaderboard[entry.model] = leaderboard[entry.model][-1000:]
+    return {"status": "recorded"}
+
 
 # ── Browser housekeeping routes ────────────────────────────────
 # Browsers auto-request these; returning proper responses prevents