fix

backend/api/openenv.py

@@ -10,6 +10,9 @@ GET  /env/info    → env metadata
 
 from __future__ import annotations
 
+import json
+import sys
+
 from fastapi import APIRouter, HTTPException
 from pydantic import BaseModel
 from typing import Optional
@@ -20,6 +23,11 @@ from env.tasks import get_all_tasks
 router = APIRouter()
 
 
+def _log(tag: str, payload: dict) -> None:
+    """Emit a single structured log line to stdout: [TAG] <json>"""
+    print(f"[{tag}] {json.dumps(payload)}", flush=True)
+
+
 # ─── Request Models ───────────────────────────────────────────────
 
 class ResetRequest(BaseModel):
@@ -42,6 +50,12 @@ async def env_reset(req: ResetRequest):
         obs = env.reset_with_question(req.task_id, req.question_id)
     else:
         obs = env.reset(req.task_id)
+    _log("START", {
+        "task_id": obs.task_id,
+        "task_difficulty": obs.task_difficulty,
+        "question": obs.question,
+        "max_attempts": obs.max_attempts,
+    })
     return obs
 
 
@@ -55,6 +69,25 @@ async def env_step(req: StepRequest):
             custom_sql=req.custom_sql,
         )
         obs, reward = await env.step(action)
+        _log("STEP", {
+            "attempt": obs.attempt_number,
+            "action": req.repair_action,
+            "sql": obs.current_sql or "",
+            "error": obs.error_message,
+            "error_class": obs.error_class,
+            "reward": round(reward.value, 4),
+            "success": reward.success,
+            "done": reward.done,
+        })
+        if reward.done:
+            ep = env._episode
+            _log("END", {
+                "success": reward.success,
+                "attempts": obs.attempt_number,
+                "total_reward": round(
+                    sum(s.reward for s in ep.steps) if ep and ep.steps else reward.value, 4
+                ),
+            })
         return {
             "observation": obs.model_dump(),
             "reward": reward.model_dump(),