diff --git a/prediction/algorithms/pair_trawl.py b/prediction/algorithms/pair_trawl.py
index 683fb72..d2e21d7 100644
--- a/prediction/algorithms/pair_trawl.py
+++ b/prediction/algorithms/pair_trawl.py
@@ -390,7 +390,7 @@ def _trajectory_similarity(
 # bbox 1차 탐색 반경 (도 단위)
 BBOX_DEG = 0.01                   # 약 1.1km — 주변 후보만 컷
 SIMILARITY_PAIR = 0.70            # 유력 페어
-SIMILARITY_OBSERVE = 0.50         # 관찰 페어
+SIMILARITY_OBSERVE = 0.45         # 관찰 페어 (0.50 → 0.45 완화: pool 확대 후 recall 확보)
 
 
 def find_pair_candidates(
diff --git a/prediction/scheduler.py b/prediction/scheduler.py
index 91e39f1..abe0e49 100644
--- a/prediction/scheduler.py
+++ b/prediction/scheduler.py
@@ -212,15 +212,28 @@ def run_analysis_cycle():
             pt_sub_registered: set[str] = set()  # TODO: fishery_code=PT-S 구분
             base_mmsis: set[str] = {c['mmsi'] for c in classifications}
             base_mmsis |= pt_registered
-
-            # pool은 전체 24h 누적 tracks (중국 어선 중심 8k+ 선박)
+            # 조업 속력대(1.5~5.0kn)에서 움직이는 모든 중국 선박을 base로 확장.
+            # classifications 500척만으로는 bbox 기점이 부족해 실제 공조 페어를 놓침.
             pool_tracks = getattr(vessel_store, '_tracks', {}) or vessel_dfs
+            for mmsi, df in pool_tracks.items():
+                if not mmsi.startswith('412'):
+                    continue
+                if df is None or df.empty or 'sog' not in df.columns:
+                    continue
+                try:
+                    mean_sog = float(df['sog'].tail(12).mean())
+                    if 1.5 <= mean_sog <= 5.0:
+                        base_mmsis.add(mmsi)
+                except Exception:
+                    continue
+
             pair_candidates = find_pair_candidates(
                 base_mmsis=base_mmsis,
                 vessel_dfs=pool_tracks,
                 get_vessel_info=vessel_store.get_vessel_info,
                 pt_registered=pt_registered,
                 pt_sub_registered=pt_sub_registered,
+                min_common_samples=4,
             )
             pt_det = 0; coop_det = 0
             for cand in pair_candidates: