diff --git a/src/ptscrapper/config.py b/src/ptscrapper/config.py
index e0bf8a0..5b6f36f 100644
--- a/src/ptscrapper/config.py
+++ b/src/ptscrapper/config.py
@@ -18,7 +18,7 @@ AUTH_PASSWORD = os.getenv("PTS_PASSWORD", "g5crehAfUCh4Wust")
 # HTTPX timeout (in seconds)
 # You can override any one by setting PTS_TIMEOUT_CONNECT, PTS_TIMEOUT_READ, etc.
 REQUEST_TIMEOUT = {
-    "connect": float(os.getenv("PTS_TIMEOUT_CONNECT", "5.0")),
+    "connect": float(os.getenv("PTS_TIMEOUT_CONNECT", "3.0")),
     "read": float(os.getenv("PTS_TIMEOUT_READ", "10.0")),
     "write": float(os.getenv("PTS_TIMEOUT_WRITE", "10.0")),
     "pool": float(os.getenv("PTS_TIMEOUT_POOL", "5.0")),
@@ -35,3 +35,4 @@ LOCAL_DATA_DIR = os.getenv("LOCAL_DATA_DIR")
 HOST_ID = os.getenv("HOST_ID", socket.gethostname())
 RCLONE_REMOTE = os.getenv("RCLONE_REMOTE", "nas")
 RCLONE_REMOTE_PATH = os.getenv("RCLONE_REMOTE_PATH", "pt-scraper-data")
+MAX_COURSE_FETCH_ATTEMPTS = os.getenv("MAX_COURSE_FETCH_ATTEMPTS", 3)
diff --git a/src/ptscrapper/scheduler.py b/src/ptscrapper/scheduler.py
index 7f84bd3..fd2c53c 100644
--- a/src/ptscrapper/scheduler.py
+++ b/src/ptscrapper/scheduler.py
@@ -7,11 +7,12 @@ import sqlite3
 from datetime import date, datetime, timedelta, timezone
 from itertools import islice
 
+import httpx
 import pyarrow as pa
 from apscheduler.schedulers.asyncio import AsyncIOScheduler
 from apscheduler.triggers.cron import CronTrigger
 
-from .config import LOCAL_DATA_DIR
+from .config import LOCAL_DATA_DIR, MAX_COURSE_FETCH_ATTEMPTS
 from .fetchers import (
     fetch_course_posts,
     fetch_positions,
@@ -44,7 +45,7 @@ os.makedirs(LOCAL_DATA_DIR, exist_ok=True)
 COURSE_WORKER_COUNT = 5
 
 # holds (course_id, ts) tuples to fetch
-course_queue: asyncio.Queue[tuple[int, datetime]] = asyncio.Queue()
+course_queue: asyncio.Queue[tuple[int, datetime, int]] = asyncio.Queue()
 
 # tracks which course_ids are already queued or being processed
 in_flight_courses: set[int] = set()
@@ -128,7 +129,7 @@ async def positions_job():
     for cid in {p["c"] for p in positions}:
         if should_fetch_course(cid) and cid not in in_flight_courses:
             in_flight_courses.add(cid)
-            await course_queue.put((cid, ts))
+            await course_queue.put((cid, ts, 0))
 
 
 async def fetch_and_store_course(course_id: int, ts: datetime, client=None):
@@ -139,24 +140,28 @@ async def fetch_and_store_course(course_id: int, ts: datetime, client=None):
         own_client = True
     try:
         recs = await fetch_course_posts(client, [course_id])
+        # now do your I/O; catch+log only errors here
         for rec in recs:
-            # flatten route stops
-            rows = []
-            for stop in rec.get("r", []):
-                rows.append(
-                    {
-                        "fetch_date": ts,
-                        "course_id": rec["c"],
-                        "stop_id": stop["s"],
-                        "dep_time": stop["t"],
-                    }
-                )
+            rows = [
+                {
+                    "fetch_date": ts,
+                    "course_id": rec["c"],
+                    "stop_id": stop["s"],
+                    "dep_time": stop["t"],
+                }
+                for stop in rec.get("r", [])
+            ]
             table = pa.Table.from_pylist(rows)
             write_course_posts(table, rec["c"], ts)
             record_course_fetched(rec["c"], ts)
             write_course_geometry(rec["p"], rec["c"], ts)
             logger.info("wrote course %s", course_id)
-    except Exception:
+    except Exception as e:
+        # if it's not a ConnectTimeout (which the worker handles), log
+        from httpx import ConnectTimeout
+
+        if isinstance(e, ConnectTimeout):
+            raise
         logger.exception("job=course_posts_job")
     finally:
         if own_client:
@@ -226,9 +231,31 @@ async def daily_compact_and_push_course_geometry():
 async def course_worker(name: str):
     """Continuously process course IDs from the queue."""
     while True:
-        course_id, ts = await course_queue.get()
+        course_id, ts, attempts = await course_queue.get()
         try:
             await fetch_and_store_course(course_id, ts)
+            logger.info("worker=%s succeeded for course=%s", name, course_id)
+        except httpx.ConnectTimeout:
+            if attempts < MAX_COURSE_FETCH_ATTEMPTS:
+                backoff = 2**attempts
+                logger.warning(
+                    "worker=%s timeout on course=%s, retry %s/%s after %ss",
+                    name,
+                    course_id,
+                    attempts + 1,
+                    MAX_COURSE_FETCH_ATTEMPTS,
+                    backoff,
+                )
+                await asyncio.sleep(backoff)
+                # requeue with incremented attempts
+                await course_queue.put((course_id, ts, attempts + 1))
+            else:
+                logger.error(
+                    "worker=%s giving up on course=%s after %s attempts",
+                    name,
+                    course_id,
+                    MAX_COURSE_FETCH_ATTEMPTS,
+                )
         except Exception:
             logger.exception(
                 "worker=%s fetch_and_store_course failed for course=%s", name, course_id