prediction/scheduler.py

@@ -437,6 +437,9 @@ def run_analysis_cycle():
             elif final_risk >= 30:
                 final_risk_level = 'MEDIUM'
 
+            # pair_result 는 STRONG/PROBABLE 필터링으로 SUSPECT 는 None.
+            # SUSPECT tier 까지 통계로 남기려면 raw pair 결과도 조회.
+            raw_pair = pair_results.get(mmsi) or {}
             merged_features = {
                 **(c.get('features', {}) or {}),
                 **dark_features,
@@ -445,6 +448,12 @@ def run_analysis_cycle():
                 'gear_violation_evidence': gear_violation_evidence,
                 'pair_trawl_detected': bool(pair_result and pair_result.get('pair_detected')),
                 'pair_trawl_pair_mmsi': (pair_result or {}).get('pair_mmsi', ''),
+                'pair_tier': raw_pair.get('tier') or '',
+                'pair_type': raw_pair.get('pair_type') or '',
+                'pair_reject_reason': raw_pair.get('reject_reason') or '',
+                'similarity': raw_pair.get('similarity', 0),
+                'confidence': raw_pair.get('confidence', 0),
+                'registered_fishery_code': registered_fishery_code or '',
             }
 
             results.append(AnalysisResult(

prediction/scripts/diagnostic-snapshot.sh

@@ -203,9 +203,10 @@ ORDER BY risk_score DESC LIMIT 20;
 SQL
 
 echo ""
-echo "--- 4-4. G-06 쌍끌이 공조 탐지 ---"
+echo "--- 4-4. G-06 쌍끌이 공조 탐지 (tier 포함) ---"
 $PSQL_TABLE << 'SQL'
 SELECT mmsi, zone_code, vessel_type, risk_score,
+       features->>'pair_tier' tier,
        (features->'gear_violation_evidence'->'G-06'->>'sync_duration_min') sync_min,
        (features->'gear_violation_evidence'->'G-06'->>'mean_separation_nm') sep_nm,
        (features->'gear_violation_evidence'->'G-06'->>'pair_mmsi') pair_mmsi,
@@ -216,6 +217,50 @@ WHERE analyzed_at > now() - interval '5 minutes'
 ORDER BY risk_score DESC LIMIT 20;
 SQL
 
+echo ""
+echo "--- 4-4.1 pair_trawl tier 분포 (DAR-03 신호 강도별) ---"
+$PSQL_TABLE << 'SQL'
+SELECT coalesce(features->>'pair_tier', '(none)') tier,
+       count(*) cnt,
+       round(avg((features->>'similarity')::numeric)::numeric, 3) avg_sim,
+       round(avg((features->'gear_violation_evidence'->'G-06'->>'sync_duration_min')::numeric)::numeric, 1) avg_sync_min
+FROM kcg.vessel_analysis_results
+WHERE analyzed_at > now() - interval '5 minutes'
+  AND features->>'pair_trawl_detected' = 'true'
+GROUP BY tier ORDER BY cnt DESC;
+SQL
+
+echo ""
+echo "--- 4-4.2 pair detection reject 사유 (최근 5분 로그) ---"
+ssh redis-211 "sudo journalctl -u kcg-ai-prediction --no-pager --since '5 minutes ago' | grep -oE 'pair detection:[^$]+reject=\{[^}]+\}' | tail -5" 2>/dev/null || true
+
+echo ""
+echo "--- 4-4.3 G-02 금어기 조업 탐지 ---"
+$PSQL_TABLE << 'SQL'
+SELECT mmsi, zone_code, vessel_type, risk_score,
+       features->>'g_codes' g_codes,
+       (features->'gear_violation_evidence'->'G-02'->>'observed_at') observed_at,
+       features->>'registered_fishery_code' fishery_code
+FROM kcg.vessel_analysis_results
+WHERE analyzed_at > now() - interval '5 minutes'
+  AND features->>'g_codes' LIKE '%G-02%'
+ORDER BY risk_score DESC LIMIT 15;
+SQL
+
+echo ""
+echo "--- 4-4.4 G-03 미등록/허가외 어구 탐지 ---"
+$PSQL_TABLE << 'SQL'
+SELECT mmsi, zone_code, vessel_type, risk_score,
+       features->>'g_codes' g_codes,
+       (features->'gear_violation_evidence'->'G-03'->>'detected_gear') detected,
+       (features->'gear_violation_evidence'->'G-03'->>'registered_fishery_code') registered,
+       (features->'gear_violation_evidence'->'G-03'->>'allowed_gears') allowed
+FROM kcg.vessel_analysis_results
+WHERE analyzed_at > now() - interval '5 minutes'
+  AND features->>'g_codes' LIKE '%G-03%'
+ORDER BY risk_score DESC LIMIT 15;
+SQL
+
 echo ""
 echo "--- 4-5. G-04 MMSI 조작 + G-05 어구 이동 ---"
 $PSQL_TABLE << 'SQL'
@@ -341,6 +386,30 @@ WHERE permit_year = EXTRACT(YEAR FROM now())::int
 GROUP BY fishery_code ORDER BY total DESC;
 SQL
 
+echo ""
+echo "--- 7.5-2b. match_method 분포 (NAME_EXACT vs NAME_FUZZY) ---"
+$PSQL_TABLE << 'SQL'
+SELECT coalesce(match_method, '(unmatched)') method,
+       count(*) cnt,
+       round(avg(match_confidence)::numeric, 3) avg_conf
+FROM kcg.fleet_vessels
+WHERE permit_year = EXTRACT(YEAR FROM now())::int
+GROUP BY method ORDER BY cnt DESC;
+SQL
+
+echo ""
+echo "--- 7.5-2c. fishery_code × match_method 교차 ---"
+$PSQL_TABLE << 'SQL'
+SELECT fishery_code,
+       count(*) FILTER (WHERE match_method = 'NAME_EXACT') exact,
+       count(*) FILTER (WHERE match_method = 'NAME_FUZZY') fuzzy,
+       count(*) FILTER (WHERE mmsi IS NULL) unmatched,
+       count(*) total
+FROM kcg.fleet_vessels
+WHERE permit_year = EXTRACT(YEAR FROM now())::int
+GROUP BY fishery_code ORDER BY total DESC;
+SQL
+
 echo ""
 echo "--- 7.5-3. vessel_analysis_results.gear_code 분포 (last 5min) ---"
 $PSQL_TABLE << 'SQL'

prediction/scripts/hourly-analysis-snapshot.sh

@@ -163,6 +163,26 @@ FROM kcg.fleet_vessels
 WHERE permit_year = EXTRACT(YEAR FROM now())::int
 GROUP BY fishery_code ORDER BY total DESC;
 
+\echo
+\echo === P3.5 match_method distribution (NAME_EXACT vs NAME_FUZZY) ===
+SELECT coalesce(match_method, '(unmatched)') method,
+       count(*) cnt,
+       round(avg(match_confidence)::numeric, 3) avg_conf
+FROM kcg.fleet_vessels
+WHERE permit_year = EXTRACT(YEAR FROM now())::int
+GROUP BY method ORDER BY cnt DESC;
+
+\echo
+\echo === P3.6 fishery_code × match_method cross ===
+SELECT fishery_code,
+       count(*) FILTER (WHERE match_method = 'NAME_EXACT') exact,
+       count(*) FILTER (WHERE match_method = 'NAME_FUZZY') fuzzy,
+       count(*) FILTER (WHERE mmsi IS NULL) unmatched,
+       count(*) total
+FROM kcg.fleet_vessels
+WHERE permit_year = EXTRACT(YEAR FROM now())::int
+GROUP BY fishery_code ORDER BY total DESC;
+
 \echo
 \echo === P4. vessel_analysis_results.gear_code distribution (last 1h) ===
 SELECT coalesce(gear_code, '(null)') gear_code,
@@ -258,6 +278,32 @@ WHERE analyzed_at > now() - interval '1 hour'
   AND features->>'pair_trawl_detected' = 'true'
 GROUP BY pair_type ORDER BY cnt DESC;
 
+\echo
+\echo === D3.6 pair_trawl tier distribution (signal-strength tier) ===
+SELECT coalesce(features->>'pair_tier', '(none)') tier,
+       count(*) cnt,
+       round(avg((features->>'similarity')::numeric)::numeric, 3) avg_sim,
+       round(avg((features->'gear_violation_evidence'->'G-06'->>'sync_duration_min')::numeric)::numeric, 1) avg_sync_min,
+       round(avg((features->'gear_violation_evidence'->'G-06'->>'mean_separation_nm')::numeric)::numeric, 3) avg_sep_nm
+FROM kcg.vessel_analysis_results
+WHERE analyzed_at > now() - interval '1 hour'
+  AND features->>'pair_trawl_detected' = 'true'
+GROUP BY tier ORDER BY cnt DESC;
+
+\echo
+\echo === D3.7 G-02 closed-season + G-03 unregistered-gear counts ===
+SELECT
+  count(*) FILTER (WHERE features->>'g_codes' LIKE '%G-02%') g02_count,
+  count(*) FILTER (WHERE features->>'g_codes' LIKE '%G-03%') g03_count,
+  count(*) FILTER (WHERE features->>'gear_judgment' = 'CLOSED_SEASON_FISHING') judg_closed,
+  count(*) FILTER (WHERE features->>'gear_judgment' = 'UNREGISTERED_GEAR') judg_unreg
+FROM kcg.vessel_analysis_results
+WHERE analyzed_at > now() - interval '1 hour';
+
+\echo
+\echo === D3.8 pair detection reject breakdown (last 1h journal) ===
+\! ssh redis-211 "sudo journalctl -u kcg-ai-prediction --no-pager --since '1 hour ago' | grep -oE 'pair detection:[^$]+reject=\{[^}]+\}' | awk -F'reject=' '{print $2}' | sort | uniq -c | sort -rn | head -10" 2>/dev/null || true
+
 \echo
 \echo === D4. G-06 pair trawl detections ===
 SELECT mmsi, zone_code, vessel_type, risk_score,