From f42bc4aa7d4dd63c2a496ebc15a7f26e1929a99b Mon Sep 17 00:00:00 2001
From: Jarca <genichin@users.noreply.github.com>
Date: Fri, 3 Jul 2026 08:25:01 +0900
Subject: [PATCH] Refactor import and workflow services plus adapter seams

---
 dashboard/app.py                              |  123 +-
 .../design.md                                 |   23 +
 .../implementation.md                         |   46 +
 .../index.md                                  |   21 +
 .../planning.md                               |   28 +
 .../verification.md                           |   37 +
 .../design.md                                 |   13 +
 .../implementation.md                         |   48 +
 .../index.md                                  |   21 +
 .../planning.md                               |   28 +
 .../verification.md                           |   36 +
 .../design.md                                 |    7 +
 .../implementation.md                         |   43 +
 .../index.md                                  |   12 +
 .../planning.md                               |   12 +
 .../verification.md                           |   43 +
 docs/batches/index.md                         |    3 +
 ...write-import-workflow-adapter-narrowing.md |   18 +-
 docs/project-structure.md                     |   60 +-
 docs/runtime-flows.md                         |   23 +-
 ...pter-surface-narrowing-baseline-refresh.md |   63 +
 ...peline-application-infrastructure-split.md |   80 +
 ...advanced-finance-workflow-service-split.md |   70 +
 docs/srs/index.md                             |   11 +-
 jarca_ledger_mcp/adapters/mcp/server.py       |  114 +-
 .../adapters/mcp/tool_contracts.py            |   60 +
 .../adapters/streamlit/__init__.py            |   12 +-
 .../adapters/streamlit/accounts_gateway.py    |    6 +
 .../adapters/streamlit/assets_gateway.py      |   10 +
 .../streamlit/categorization_gateway.py       |   20 +
 .../adapters/streamlit/imports_gateway.py     |   10 +
 .../adapters/streamlit/overview_gateway.py    |   13 +
 .../streamlit/reconciliation_gateway.py       |    3 +
 .../adapters/streamlit/service_gateway.py     |   50 +-
 .../streamlit/transactions_gateway.py         |    3 +
 .../adapters/streamlit/transfers_gateway.py   |   11 +
 jarca_ledger_mcp/application/__init__.py      |    5 +
 .../application/asset_account_services.py     |  502 +-----
 .../application/assets_service.py             |  490 ++++++
 .../application/categorization_services.py    |  224 +--
 .../application/cross_transfer_service.py     |  593 +++++++
 .../application/import_transfer_services.py   | 1530 +----------------
 .../application/imports_service.py            |  922 ++++++++++
 .../application/reconciliation_service.py     |   94 +
 jarca_ledger_mcp/application/shared.py        |   14 +-
 .../transaction_report_services.py            |   93 +-
 .../uncategorized_workbench_service.py        |  246 +++
 .../infrastructure/parsers/__init__.py        |    4 +
 .../infrastructure/parsers/kb_bank_pdf.py     |    1 +
 .../infrastructure/parsers/seoulpay_xlsx.py   |    1 +
 .../infrastructure/parsers/toss_bank_pdf.py   |    1 +
 .../parsers/toss_securities_pdf.py            |    1 +
 jarca_ledger_mcp/service.py                   |   15 +
 53 files changed, 3420 insertions(+), 2497 deletions(-)
 create mode 100644 docs/batches/bat-012_20260703_import-pipeline-split/design.md
 create mode 100644 docs/batches/bat-012_20260703_import-pipeline-split/implementation.md
 create mode 100644 docs/batches/bat-012_20260703_import-pipeline-split/index.md
 create mode 100644 docs/batches/bat-012_20260703_import-pipeline-split/planning.md
 create mode 100644 docs/batches/bat-012_20260703_import-pipeline-split/verification.md
 create mode 100644 docs/batches/bat-013_20260703_advanced-workflow-service-split/design.md
 create mode 100644 docs/batches/bat-013_20260703_advanced-workflow-service-split/implementation.md
 create mode 100644 docs/batches/bat-013_20260703_advanced-workflow-service-split/index.md
 create mode 100644 docs/batches/bat-013_20260703_advanced-workflow-service-split/planning.md
 create mode 100644 docs/batches/bat-013_20260703_advanced-workflow-service-split/verification.md
 create mode 100644 docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/design.md
 create mode 100644 docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/implementation.md
 create mode 100644 docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/index.md
 create mode 100644 docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/planning.md
 create mode 100644 docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/verification.md
 create mode 100644 docs/srs/Interface/req-018_adapter-surface-narrowing-baseline-refresh.md
 create mode 100644 docs/srs/Non-Functional/req-016_import-pipeline-application-infrastructure-split.md
 create mode 100644 docs/srs/Non-Functional/req-017_advanced-finance-workflow-service-split.md
 create mode 100644 jarca_ledger_mcp/adapters/mcp/tool_contracts.py
 create mode 100644 jarca_ledger_mcp/adapters/streamlit/accounts_gateway.py
 create mode 100644 jarca_ledger_mcp/adapters/streamlit/assets_gateway.py
 create mode 100644 jarca_ledger_mcp/adapters/streamlit/categorization_gateway.py
 create mode 100644 jarca_ledger_mcp/adapters/streamlit/imports_gateway.py
 create mode 100644 jarca_ledger_mcp/adapters/streamlit/overview_gateway.py
 create mode 100644 jarca_ledger_mcp/adapters/streamlit/reconciliation_gateway.py
 create mode 100644 jarca_ledger_mcp/adapters/streamlit/transactions_gateway.py
 create mode 100644 jarca_ledger_mcp/adapters/streamlit/transfers_gateway.py
 create mode 100644 jarca_ledger_mcp/application/assets_service.py
 create mode 100644 jarca_ledger_mcp/application/cross_transfer_service.py
 create mode 100644 jarca_ledger_mcp/application/imports_service.py
 create mode 100644 jarca_ledger_mcp/application/reconciliation_service.py
 create mode 100644 jarca_ledger_mcp/application/uncategorized_workbench_service.py
 create mode 100644 jarca_ledger_mcp/infrastructure/parsers/__init__.py
 create mode 100644 jarca_ledger_mcp/infrastructure/parsers/kb_bank_pdf.py
 create mode 100644 jarca_ledger_mcp/infrastructure/parsers/seoulpay_xlsx.py
 create mode 100644 jarca_ledger_mcp/infrastructure/parsers/toss_bank_pdf.py
 create mode 100644 jarca_ledger_mcp/infrastructure/parsers/toss_securities_pdf.py

diff --git a/dashboard/app.py b/dashboard/app.py
index 354b15d..4f53e3d 100644
--- a/dashboard/app.py
+++ b/dashboard/app.py
@@ -15,7 +15,14 @@ import plotly.express as px
 import plotly.io as pio
 import streamlit as st
 
-from jarca_ledger_mcp.adapters.streamlit import service_gateway as service
+from jarca_ledger_mcp.adapters.streamlit import accounts_gateway as accounts_gateway
+from jarca_ledger_mcp.adapters.streamlit import assets_gateway as assets_gateway
+from jarca_ledger_mcp.adapters.streamlit import categorization_gateway as categorization_gateway
+from jarca_ledger_mcp.adapters.streamlit import imports_gateway as imports_gateway
+from jarca_ledger_mcp.adapters.streamlit import overview_gateway as overview_gateway
+from jarca_ledger_mcp.adapters.streamlit import reconciliation_gateway as reconciliation_gateway
+from jarca_ledger_mcp.adapters.streamlit import transactions_gateway as transactions_gateway
+from jarca_ledger_mcp.adapters.streamlit import transfers_gateway as transfers_gateway
 from jarca_ledger_mcp.db import resolve_db_path
 
 st.set_page_config(page_title="Jarca Ledger", page_icon="💰", layout="wide")
@@ -146,7 +153,7 @@ def get_db_path() -> str:
     return str(resolve_db_path())
 
 
-def get_service_callable(name: str, required_params: list[str] | None = None):
+def get_service_callable(module, name: str, required_params: list[str] | None = None):
     """Return a service function, reloading once for Streamlit hot-reload edge cases.
 
     Streamlit can keep an already-imported service module in memory while the
@@ -155,7 +162,7 @@ def get_service_callable(name: str, required_params: list[str] | None = None):
     signature. In that case reload the service module before returning it.
     """
     required_params = required_params or []
-    fn = getattr(service, name, None)
+    fn = getattr(module, name, None)
     if callable(fn):
         try:
             params = inspect.signature(fn).parameters
@@ -164,8 +171,8 @@ def get_service_callable(name: str, required_params: list[str] | None = None):
         except (TypeError, ValueError):
             if not required_params:
                 return fn
-    importlib.reload(service)
-    return getattr(service, name, None)
+    importlib.reload(module)
+    return getattr(module, name, None)
 
 
 def inject_app_css() -> None:
@@ -431,10 +438,10 @@ def parse_optional_int(value: str, field_name: str) -> int | None:
 
 
 def get_category_options(db_path: str) -> tuple[list[str], dict[str, list[str]]]:
-    result = service.list_ledger_categories(active_only=True, include_usage=False, db_path=db_path)
+    result = categorization_gateway.list_ledger_categories(active_only=True, include_usage=False, db_path=db_path)
     if result.get("success") and result.get("count", 0) == 0:
-        service.seed_default_ledger_categories(db_path=db_path)
-        result = service.list_ledger_categories(active_only=True, include_usage=False, db_path=db_path)
+        categorization_gateway.seed_default_ledger_categories(db_path=db_path)
+        result = categorization_gateway.list_ledger_categories(active_only=True, include_usage=False, db_path=db_path)
     grouped = result.get("grouped", {}) if result.get("success") else {}
     categories = [""] + sorted(grouped.keys())
     subcategories = {cat: [""] + sorted([s for s in subs if s]) for cat, subs in grouped.items()}
@@ -524,7 +531,7 @@ def update_transaction_form(tx: dict[str, Any], key_prefix: str, db_path: str |
     submitted = st.button("저장", type="primary", key=f"{key_prefix}_submit")
 
     if submitted:
-        result = service.update_ledger_transaction(
+        result = transactions_gateway.update_ledger_transaction(
             tx["id"], category=category, subcategory=subcategory, person=person,
             is_fixed=is_fixed, memo=memo, db_path=db_path,
         )
@@ -532,7 +539,7 @@ def update_transaction_form(tx: dict[str, Any], key_prefix: str, db_path: str |
             st.error(result.get("error", result))
             return
         if save_rule and tx.get("merchant") and category:
-            rule = service.create_category_rule(
+            rule = categorization_gateway.create_category_rule(
                 match_field="merchant",
                 pattern=tx["merchant"],
                 category=category,
@@ -560,7 +567,7 @@ def transaction_edit_dialog(tx: dict[str, Any], key_prefix: str, db_path: str) -
 def linked_transfer_candidates_dialog(tx: dict[str, Any], db_path: str) -> None:
     st.caption(f"기준 거래: `{tx['id']}`")
     st.write(f"**{tx.get('transaction_datetime') or tx.get('transaction_date')}** · {tx.get('merchant')} · {format_krw(tx.get('amount'))}")
-    result = service.find_self_transfer_candidates(tx["id"], limit=20, db_path=db_path)
+    result = transfers_gateway.find_self_transfer_candidates(tx["id"], limit=20, db_path=db_path)
     if not result.get("success"):
         st.error(result.get("error", result))
         return
@@ -578,7 +585,7 @@ def linked_transfer_candidates_dialog(tx: dict[str, Any], db_path: str) -> None:
         cols[3].write(format_krw(candidate.get("amount")))
         cols[4].write(candidate.get("merchant") or candidate.get("counterparty") or "")
         if cols[5].button("이 거래와 연결", key=f"link_pair_{tx['id']}_{candidate['id']}_{idx}", use_container_width=True, type="primary"):
-            link_res = service.link_self_transfer_pair(tx["id"], candidate["id"], db_path=db_path)
+            link_res = transfers_gateway.link_self_transfer_pair(tx["id"], candidate["id"], db_path=db_path)
             if not link_res.get("success"):
                 st.error(link_res.get("error", link_res))
             else:
@@ -603,7 +610,7 @@ def linked_transfer_candidates_dialog(tx: dict[str, Any], db_path: str) -> None:
 def cancel_original_candidates_dialog(tx: dict[str, Any], db_path: str) -> None:
     st.caption(f"취소/환급 거래: `{tx['id']}`")
     st.write(f"**{tx.get('transaction_datetime') or tx.get('transaction_date')}** · {tx.get('merchant')} · {format_krw(tx.get('amount'))}")
-    result = service.find_cancel_original_candidates(tx["id"], limit=20, db_path=db_path)
+    result = transfers_gateway.find_cancel_original_candidates(tx["id"], limit=20, db_path=db_path)
     if not result.get("success"):
         st.error(result.get("error", result))
         return
@@ -621,7 +628,7 @@ def cancel_original_candidates_dialog(tx: dict[str, Any], db_path: str) -> None:
         cols[3].write(format_krw(candidate.get("amount")))
         cols[4].write(candidate.get("merchant") or candidate.get("counterparty") or "")
         if cols[5].button("이 원거래와 연결", key=f"cancel_pair_{tx['id']}_{candidate['id']}_{idx}", use_container_width=True, type="primary"):
-            link_res = service.link_cancel_to_original(tx["id"], candidate["id"], db_path=db_path)
+            link_res = transfers_gateway.link_cancel_to_original(tx["id"], candidate["id"], db_path=db_path)
             if not link_res.get("success"):
                 st.error(link_res.get("error", link_res))
             else:
@@ -683,7 +690,7 @@ def category_rule_create_dialog(db_path: str) -> None:
             st.error(str(exc))
         else:
             fixed_value = None if is_fixed == "변경안함" else is_fixed == "예"
-            res = service.create_category_rule(match_field, pattern, category, subcategory, tx_type, match_type, int(priority), person, fixed_value, 1.0, memo, amount_min, amount_max, day_of_month, db_path=db_path)
+            res = categorization_gateway.create_category_rule(match_field, pattern, category, subcategory, tx_type, match_type, int(priority), person, fixed_value, 1.0, memo, amount_min, amount_max, day_of_month, db_path=db_path)
             clear_cache()
             if res.get("success"):
                 st.success("규칙을 저장했습니다.")
@@ -735,7 +742,7 @@ def category_rule_edit_dialog(rule: dict[str, Any], db_path: str) -> None:
             st.error(str(exc))
         else:
             edit_fixed_value = None if edit_is_fixed == "변경안함" else edit_is_fixed == "예"
-            update_rule = getattr(service, "update_category_rule", None)
+            update_rule = getattr(categorization_gateway, "update_category_rule", None)
             if update_rule is None:
                 st.error("현재 실행 중인 서비스에 update_category_rule 함수가 없습니다. 앱을 재시작해 최신 코드를 다시 불러오세요.")
             else:
@@ -787,7 +794,7 @@ def save_merchant_rule_from_recommendation(merchant: str, tx_type: str, rec: dic
         return {"success": False, "error": "merchant가 비어 있어 규칙을 만들 수 없습니다."}
     if not rec.get("category"):
         return {"success": False, "error": "추천 분류가 없어 규칙을 만들 수 없습니다."}
-    return service.create_category_rule(
+    return categorization_gateway.create_category_rule(
         match_field="merchant",
         match_type="contains",
         pattern=merchant,
@@ -817,7 +824,7 @@ def render_uncategorized_groups(groups: list[dict[str, Any]], db_path: str, repe
         cols[3].write(recommendation_text(rec))
         apply_disabled = not bool(rec.get("category"))
         if cols[4].button("그룹 적용", key=f"uncat_group_apply_{idx}_{merchant}", disabled=apply_disabled, use_container_width=True):
-            res = service.apply_uncategorized_group_recommendation(
+            res = categorization_gateway.apply_uncategorized_group_recommendation(
                 group.get("merchant", ""), group.get("raw_summary", ""), group.get("type", ""), db_path=db_path
             )
             clear_cache()
@@ -860,7 +867,7 @@ def render_uncategorized_work_items(rows: list[dict[str, Any]], db_path: str, re
             action_cols = st.columns([1.0, 1.0, 1.0, 1.1, 1.1, 2.5], vertical_alignment="center")
             apply_disabled = not bool(rec.get("category"))
             if action_cols[0].button("추천 적용", key=f"uncat_apply_rec_{row['id']}_{idx}", disabled=apply_disabled, use_container_width=True, type="primary"):
-                res = service.apply_uncategorized_recommendation(row["id"], db_path=db_path)
+                res = categorization_gateway.apply_uncategorized_recommendation(row["id"], db_path=db_path)
                 clear_cache()
                 if res.get("success"):
                     st.success(f"{recommendation_text(res.get('recommendation'))}로 분류했습니다.")
@@ -892,7 +899,7 @@ def category_rule_preview_dialog(rule: dict[str, Any], db_path: str) -> None:
     st.caption(f"규칙 ID: `{rule_id}`")
     st.write(f"**{pattern}** → **{result_text}**")
 
-    preview_fn = get_service_callable("preview_category_rule_matches")
+    preview_fn = get_service_callable(categorization_gateway, "preview_category_rule_matches")
     if not callable(preview_fn):
         st.error("현재 실행 중인 서비스에서 단일 규칙 미리보기 함수를 찾지 못했습니다. Streamlit 서버를 완전히 종료한 뒤 다시 시작해 주세요.")
         return
@@ -922,7 +929,7 @@ def category_rule_preview_dialog(rule: dict[str, Any], db_path: str) -> None:
         st.info("이 규칙에 부합하는 거래가 없습니다.")
 
 def page_overview(db_path: str) -> None:
-    health = service.health_check(db_path)
+    health = overview_gateway.health_check(db_path)
     render_page_header(
         "가계부 운영 대시보드",
         "이번 달 실제 현금흐름, 자산 유동성, 운영 경고를 우선순위대로 확인합니다.",
@@ -931,14 +938,14 @@ def page_overview(db_path: str) -> None:
     )
     st.caption(f"DB: `{health.get('db_path')}`")
     month = current_kst_month()
-    cashflow = service.get_monthly_cashflow_report(month, db_path=db_path)
-    asset_report = service.get_asset_liquidity_report(db_path=db_path)
-    budget_report = service.get_budget_vs_actual_report(month, db_path=db_path)
-    uncategorized_res = service.list_uncategorized_transactions(limit=1000, db_path=db_path)
-    transfer_queue_res = service.list_cross_transfer_review_queue(limit=300, db_path=db_path)
-    reconciliation_res = service.get_balance_reconciliation_report(db_path=db_path)
-    backups_res = service.list_ledger_backups(limit=1)
-    imports_res = service.list_ledger_import_batches(limit=1, db_path=db_path)
+    cashflow = overview_gateway.get_monthly_cashflow_report(month, db_path=db_path)
+    asset_report = overview_gateway.get_asset_liquidity_report(db_path=db_path)
+    budget_report = overview_gateway.get_budget_vs_actual_report(month, db_path=db_path)
+    uncategorized_res = overview_gateway.list_uncategorized_transactions(limit=1000, db_path=db_path)
+    transfer_queue_res = overview_gateway.list_cross_transfer_review_queue(limit=300, db_path=db_path)
+    reconciliation_res = reconciliation_gateway.get_balance_reconciliation_report(db_path=db_path)
+    backups_res = overview_gateway.list_ledger_backups(limit=1)
+    imports_res = overview_gateway.list_ledger_import_batches(limit=1, db_path=db_path)
 
     render_section_label("💎", "이번 달 수입/지출 현황")
     st.caption(f"기준: KST 현재 월 {month}, transactions.transaction_date, ledger_categories.cashflow_kind 실제 현금흐름")
@@ -1037,7 +1044,7 @@ def page_assets(db_path: str) -> None:
         memo = st.text_input("운영 메모", value="")
         submitted = st.form_submit_button("수기 자산 추가", type="primary")
     if submitted:
-        res = service.create_manual_asset(name, int(amount), valuation_date.isoformat(), bucket, source_note=source_note, memo=memo, db_path=db_path)
+        res = assets_gateway.create_manual_asset(name, int(amount), valuation_date.isoformat(), bucket, source_note=source_note, memo=memo, db_path=db_path)
         if res.get("success"):
             clear_cache()
             st.success("수기 자산을 추가했습니다.")
@@ -1045,7 +1052,7 @@ def page_assets(db_path: str) -> None:
         else:
             st.error(res.get("error", res))
 
-    assets_res = service.list_manual_assets(active_only=False, db_path=db_path)
+    assets_res = assets_gateway.list_manual_assets(active_only=False, db_path=db_path)
     assets = assets_res.get("manual_assets", []) if assets_res.get("success") else []
     render_section_label("📋", "수기 자산/부채 목록")
     if assets:
@@ -1071,13 +1078,13 @@ def page_assets(db_path: str) -> None:
             do_update = col_update.form_submit_button("선택 항목 수정", type="primary")
             do_deactivate = col_deactivate.form_submit_button("선택 항목 비활성", type="secondary")
         if do_update:
-            res = service.update_manual_asset(selected["id"], name=new_name, valuation_amount=int(new_amount), valuation_date=new_date, liquidity_bucket=new_bucket, source_note=new_source, memo=new_memo, active=make_active, db_path=db_path)
+            res = assets_gateway.update_manual_asset(selected["id"], name=new_name, valuation_amount=int(new_amount), valuation_date=new_date, liquidity_bucket=new_bucket, source_note=new_source, memo=new_memo, active=make_active, db_path=db_path)
             if res.get("success"):
                 clear_cache(); st.success("수정했습니다."); st.rerun()
             else:
                 st.error(res.get("error", res))
         if do_deactivate:
-            res = service.deactivate_manual_asset(selected["id"], db_path=db_path)
+            res = assets_gateway.deactivate_manual_asset(selected["id"], db_path=db_path)
             if res.get("success"):
                 clear_cache(); st.warning("비활성 처리했습니다."); st.rerun()
             else:
@@ -1094,9 +1101,9 @@ def page_assets(db_path: str) -> None:
     correction_reason = c3.text_input("정정 사유(재마감 시 필수)", value="")
     p1, p2 = st.columns(2)
     if p1.button("스냅샷 Preview", type="secondary"):
-        st.session_state["asset_snapshot_preview"] = service.preview_asset_snapshot_close(basis_date, db_path=db_path)
+        st.session_state["asset_snapshot_preview"] = assets_gateway.preview_asset_snapshot_close(basis_date, db_path=db_path)
     if p2.button("스냅샷 Close 실행", type="primary"):
-        st.session_state["asset_snapshot_close"] = service.close_asset_snapshot(basis_date, source_note=close_note, correction_reason=correction_reason, db_path=db_path)
+        st.session_state["asset_snapshot_close"] = assets_gateway.close_asset_snapshot(basis_date, source_note=close_note, correction_reason=correction_reason, db_path=db_path)
         clear_cache()
     preview = st.session_state.get("asset_snapshot_preview")
     close_res = st.session_state.get("asset_snapshot_close")
@@ -1170,10 +1177,10 @@ def page_cross_transfer_review(db_path: str) -> None:
     c1, c2 = st.columns(2)
     with c1:
         if st.button("미리보기(드라이런)"):
-            st.session_state["cross_transfer_preview"] = service.preview_cross_account_transfers(db_path=db_path)
+            st.session_state["cross_transfer_preview"] = transfers_gateway.preview_cross_account_transfers(db_path=db_path)
     with c2:
         if st.button("자동 매칭 실행"):
-            res = service.run_cross_account_transfer_match(db_path=db_path)
+            res = transfers_gateway.run_cross_account_transfer_match(db_path=db_path)
             clear_cache()
             st.session_state["cross_transfer_last_run"] = res
             st.success(f"매칭 {res.get('matched_count', 0)}건 적용, 확인필요 {res.get('review_count', 0)}건")
@@ -1190,7 +1197,7 @@ def page_cross_transfer_review(db_path: str) -> None:
 
     st.divider()
     render_section_label("📌", "확인필요 큐")
-    list_queue_fn = getattr(service, "list_cross_transfer_review_queue", None)
+    list_queue_fn = getattr(transfers_gateway, "list_cross_transfer_review_queue", None)
     if callable(list_queue_fn):
         queue = list_queue_fn(db_path=db_path)
         rows = queue.get("transactions", [])
@@ -1242,7 +1249,7 @@ def page_uncategorized(db_path: str) -> None:
     sort_label = filter_cols[4].selectbox("작업목록 정렬", ["금액순", "날짜순(최신순)"], index=0)
     sort_by = "date_desc" if sort_label == "날짜순(최신순)" else "amount_desc"
 
-    workbench_fn = get_service_callable("get_uncategorized_workbench", required_params=["sort_by"])
+    workbench_fn = get_service_callable(categorization_gateway, "get_uncategorized_workbench", required_params=["sort_by"])
     if not callable(workbench_fn):
         st.error("현재 실행 중인 서비스에서 미분류 작업대 함수를 찾지 못했습니다. Streamlit 서버를 완전히 재시작해 주세요.")
         return
@@ -1449,7 +1456,7 @@ def render_rule_list_items(rules: list[dict[str, Any]], db_path: str) -> None:
             if action_cols[0].button("🔎 적용(미리보기)", key=f"rule_expander_preview_{rule_id}_{idx}", use_container_width=True):
                 category_rule_preview_dialog(rule, db_path)
             if action_cols[1].button("✅ 적용", key=f"rule_expander_apply_{rule_id}_{idx}", use_container_width=True):
-                apply_fn = get_service_callable("apply_category_rule")
+                apply_fn = get_service_callable(categorization_gateway, "apply_category_rule")
                 if not callable(apply_fn):
                     st.error("현재 실행 중인 서비스에서 단일 규칙 적용 함수를 찾지 못했습니다. Streamlit 서버를 완전히 종료한 뒤 다시 시작해 주세요.")
                 else:
@@ -1465,7 +1472,7 @@ def render_rule_list_items(rules: list[dict[str, Any]], db_path: str) -> None:
             with action_cols[3]:
                 st.markdown('<div class="rule-danger-note">삭제보다 비활성화를 권장합니다.</div>', unsafe_allow_html=True)
             if action_cols[4].button("🗑️ 삭제", key=f"rule_expander_delete_{rule_id}_{idx}"):
-                delete_rule = getattr(service, "delete_category_rule", None)
+                delete_rule = getattr(categorization_gateway, "delete_category_rule", None)
                 if delete_rule is None:
                     st.error("현재 실행 중인 서비스에 delete_category_rule 함수가 없습니다. 앱을 재시작해 최신 코드를 다시 불러오세요.")
                 else:
@@ -1491,7 +1498,7 @@ def page_rules(db_path: str) -> None:
     with action_cols[1]:
         if st.button("추가", key="rule_add_dialog_open", use_container_width=True, type="primary"):
             category_rule_create_dialog(db_path)
-    rules = service.list_category_rules(active_only=False, db_path=db_path).get("rules", [])
+    rules = categorization_gateway.list_category_rules(active_only=False, db_path=db_path).get("rules", [])
     render_section_label("📚", "현재 분류 규칙")
     render_rule_list_items(rules, db_path)
 
@@ -1499,12 +1506,12 @@ def page_rules(db_path: str) -> None:
 
     c1, c2, c3 = st.columns(3)
     if c1.button("기본 규칙 seed/update"):
-        res = service.seed_default_category_rules(db_path=db_path)
+        res = categorization_gateway.seed_default_category_rules(db_path=db_path)
         clear_cache(); st.success(res)
     if c2.button("규칙 preview"):
-        st.session_state["rule_preview"] = service.preview_categorization(limit=5000, db_path=db_path)
+        st.session_state["rule_preview"] = categorization_gateway.preview_categorization(limit=5000, db_path=db_path)
     if c3.button("규칙 적용"):
-        res = service.apply_categorization_rules(limit=5000, db_path=db_path)
+        res = categorization_gateway.apply_categorization_rules(limit=5000, db_path=db_path)
         clear_cache(); st.success(f"{res.get('applied_count', 0)}건 적용")
     if "rule_preview" in st.session_state:
         prev = st.session_state["rule_preview"]
@@ -1528,7 +1535,7 @@ def page_categories(db_path: str) -> None:
 
     c1, c2 = st.columns(2)
     if c1.button("기본 카테고리 seed/update"):
-        res = service.seed_default_ledger_categories(db_path=db_path)
+        res = categorization_gateway.seed_default_ledger_categories(db_path=db_path)
         clear_cache()
         st.success(f"{res.get('created_or_updated', 0)}개 카테고리를 생성/갱신했습니다.")
         st.rerun()
@@ -1545,7 +1552,7 @@ def page_categories(db_path: str) -> None:
         submitted = st.form_submit_button("추가/재활성화", type="primary")
     if submitted:
         cashflow_kind = next(k for k, v in CASHFLOW_KIND_LABELS.items() if v == cashflow_kind_label)
-        res = service.create_ledger_category(category, subcategory, int(sort_order), memo, cashflow_kind, db_path=db_path)
+        res = categorization_gateway.create_ledger_category(category, subcategory, int(sort_order), memo, cashflow_kind, db_path=db_path)
         clear_cache()
         if res.get("success"):
             st.success("카테고리를 저장했습니다.")
@@ -1553,7 +1560,7 @@ def page_categories(db_path: str) -> None:
         else:
             st.error(res.get("error", res))
 
-    result = service.list_ledger_categories(active_only=not show_inactive, include_usage=True, db_path=db_path)
+    result = categorization_gateway.list_ledger_categories(active_only=not show_inactive, include_usage=True, db_path=db_path)
     rows = result.get("categories", [])
     render_section_label("🗂️", "카테고리 목록")
     if rows:
@@ -1577,7 +1584,7 @@ def page_categories(db_path: str) -> None:
     hard_delete = st.checkbox("DB에서 완전 삭제", value=False, help="보통은 끄고 비활성화만 하세요. 거래 기록은 별도 보존됩니다.")
     force = st.checkbox("사용 중이어도 강제", value=False, help="거래/예산/규칙에서 사용 중인 카테고리도 삭제/비활성화합니다. 거래 category 값은 변경하지 않습니다.")
     if st.button("삭제/비활성화 실행", type="secondary"):
-        res = service.delete_ledger_category(target["category"], target.get("subcategory", ""), hard_delete=hard_delete, force=force, db_path=db_path)
+        res = categorization_gateway.delete_ledger_category(target["category"], target.get("subcategory", ""), hard_delete=hard_delete, force=force, db_path=db_path)
         clear_cache()
         if res.get("success"):
             st.warning(f"처리했습니다. 사용 거래 {res.get('transaction_count', 0)}건")
@@ -1593,7 +1600,7 @@ def page_accounts(db_path: str) -> None:
         kicker="🏦 Accounts",
         pills=["잔액", "기관", "statement_status", "수동 잔액대조"],
     )
-    accounts_res = service.list_ledger_accounts(active_only=False, db_path=db_path)
+    accounts_res = accounts_gateway.list_ledger_accounts(active_only=False, db_path=db_path)
     accounts_data = accounts_res.get("accounts", []) if accounts_res.get("success") else []
     accounts = pd.DataFrame(accounts_data)
     if not accounts.empty:
@@ -1638,7 +1645,7 @@ def page_accounts(db_path: str) -> None:
                 except ValueError:
                     st.error("기준 잔액은 정수로 입력하세요.")
                     return
-            res = service.update_ledger_account(selected["id"], db_path=db_path, **update_fields)
+            res = accounts_gateway.update_ledger_account(selected["id"], db_path=db_path, **update_fields)
             clear_cache()
             if res.get("success"):
                 st.success("계좌 상태를 저장했습니다.")
@@ -1654,7 +1661,7 @@ def page_accounts(db_path: str) -> None:
     tolerance = c2.number_input("허용 오차(원)", min_value=0, value=1000, step=100)
     source_note = c3.text_input("출처 메모", value="", placeholder="예: 월말 앱 잔액")
     if st.button("잔액대조 실행", type="secondary"):
-        st.session_state["balance_reconciliation_report"] = service.get_balance_reconciliation_report(
+        st.session_state["balance_reconciliation_report"] = reconciliation_gateway.get_balance_reconciliation_report(
             baseline_at=baseline_at,
             source_note=source_note,
             tolerance=int(tolerance),
@@ -1696,10 +1703,10 @@ def page_imports(db_path: str) -> None:
         upload_path.write_bytes(uploaded.getbuffer())
         c1, c2 = st.columns(2)
         if c1.button("토스증권 Preview"):
-            res = service.preview_toss_securities_pdf_import(str(upload_path), db_path=db_path)
+            res = imports_gateway.preview_toss_securities_pdf_import(str(upload_path), db_path=db_path)
             st.session_state["toss_preview"] = res
         if c2.button("토스증권 Import", type="primary"):
-            res = service.import_toss_securities_pdf(str(upload_path), db_path=db_path)
+            res = imports_gateway.import_toss_securities_pdf(str(upload_path), db_path=db_path)
             clear_cache()
             st.session_state["toss_import_result"] = res
             st.rerun()
@@ -1730,10 +1737,10 @@ def page_imports(db_path: str) -> None:
         upload_path.write_bytes(toss_bank_uploaded.getbuffer())
         c1, c2 = st.columns(2)
         if c1.button("토스뱅크 Preview"):
-            res = service.preview_toss_bank_pdf_import(str(upload_path), db_path=db_path)
+            res = imports_gateway.preview_toss_bank_pdf_import(str(upload_path), db_path=db_path)
             st.session_state["toss_bank_preview"] = res
         if c2.button("토스뱅크 Import", type="primary"):
-            res = service.import_toss_bank_pdf(str(upload_path), db_path=db_path)
+            res = imports_gateway.import_toss_bank_pdf(str(upload_path), db_path=db_path)
             clear_cache()
             st.session_state["toss_bank_import_result"] = res
             st.rerun()
@@ -1755,7 +1762,7 @@ def page_imports(db_path: str) -> None:
 
     st.divider()
     render_section_label("🧾", "Import 이력")
-    batches = service.list_ledger_import_batches(limit=100, db_path=db_path).get("import_batches", [])
+    batches = overview_gateway.list_ledger_import_batches(limit=100, db_path=db_path).get("import_batches", [])
     st.dataframe(pd.DataFrame(batches), width="stretch", height=350)
     if batches:
         with st.form("rollback"):
@@ -1763,7 +1770,7 @@ def page_imports(db_path: str) -> None:
             hard = st.checkbox("Hard delete", value=False)
             submitted = st.form_submit_button("Rollback 실행", type="secondary")
         if submitted:
-            res = service.rollback_ledger_import_batch(batch_id, hard_delete=hard, db_path=db_path)
+            res = imports_gateway.rollback_ledger_import_batch(batch_id, hard_delete=hard, db_path=db_path)
             clear_cache()
             st.warning(res)
             st.rerun()
diff --git a/docs/batches/bat-012_20260703_import-pipeline-split/design.md b/docs/batches/bat-012_20260703_import-pipeline-split/design.md
new file mode 100644
index 0000000..215b2b2
--- /dev/null
+++ b/docs/batches/bat-012_20260703_import-pipeline-split/design.md
@@ -0,0 +1,23 @@
+# BAT-012 Design
+
+## Architecture summary
+- Introduce `jarca_ledger_mcp/application/imports_service.py` as the single application-layer home for import preview/import/rollback orchestration.
+- Introduce `jarca_ledger_mcp/infrastructure/parsers/` as the parser seam so application code no longer imports top-level parser implementations directly.
+- Keep current top-level parser modules for compatibility if needed, but route application/shared usage through the infrastructure seam.
+- Preserve `application/import_transfer_services.py` as the advanced workflow home; import-facing public functions become delegations into `imports_service.py`.
+
+## Changed areas
+- application import orchestration
+- infrastructure parser seam
+- service export surface / compatibility proxy
+- import regression verification docs
+
+## Key decisions
+1. Parser relocation is seam-first, not algorithm rewrite.
+2. Import account auto-linking/classification stays in the application layer, not in parser modules.
+3. Public API/tool names remain unchanged.
+
+## Verification focus
+- Import preview/import/rollback regression
+- Evidence field preservation
+- No breakage in MCP discovery and public exports
diff --git a/docs/batches/bat-012_20260703_import-pipeline-split/implementation.md b/docs/batches/bat-012_20260703_import-pipeline-split/implementation.md
new file mode 100644
index 0000000..fb1e9d8
--- /dev/null
+++ b/docs/batches/bat-012_20260703_import-pipeline-split/implementation.md
@@ -0,0 +1,46 @@
+# BAT-012 Implementation
+
+## Plan summary
+
+BAT-012 implements dcy-011 Phase 5 by extracting import preview/import/rollback orchestration into `application/imports_service.py`, introducing an `infrastructure/parsers/*` seam, and shrinking `application/import_transfer_services.py` toward advanced workflow ownership while preserving public contracts.
+
+## Changed files
+
+- `jarca_ledger_mcp/application/imports_service.py`
+- `jarca_ledger_mcp/application/import_transfer_services.py`
+- `jarca_ledger_mcp/application/shared.py`
+- `jarca_ledger_mcp/application/__init__.py`
+- `jarca_ledger_mcp/infrastructure/parsers/__init__.py`
+- `jarca_ledger_mcp/infrastructure/parsers/kb_bank_pdf.py`
+- `jarca_ledger_mcp/infrastructure/parsers/toss_bank_pdf.py`
+- `jarca_ledger_mcp/infrastructure/parsers/toss_securities_pdf.py`
+- `jarca_ledger_mcp/infrastructure/parsers/seoulpay_xlsx.py`
+- `jarca_ledger_mcp/service.py`
+- `docs/srs/index.md`
+- `docs/srs/Non-Functional/req-016_import-pipeline-application-infrastructure-split.md`
+- `docs/batches/index.md`
+- `docs/batches/bat-012_20260703_import-pipeline-split/*`
+- `docs/discovery/dcy-011_20260703_backend-refactor-phase4-write-import-workflow-adapter-narrowing.md`
+
+## Execution log
+
+- 2026-07-03 07:49 KST: dcy-011 Phase 5 범위로 `REQ-016`/`BAT-012`를 생성하고 승인 상태를 정리했다.
+- 2026-07-03 07:50 KST: `application/import_transfer_services.py`의 AST function ranges를 추출해 import-owned 함수와 advanced workflow 함수를 기계적으로 분리할 준비를 했다.
+- 2026-07-03 07:51 KST: `application/imports_service.py`를 생성하고 import preview/import/rollback + import-specific account/classification helpers를 이동했다.
+- 2026-07-03 07:51 KST: `application/import_transfer_services.py`를 advanced workflow 중심 모듈로 축소하고 import-facing public functions는 compatibility delegation으로 전환했다.
+- 2026-07-03 07:51 KST: `infrastructure/parsers/*` seam을 추가하고 `application/shared.py`가 top-level parser 구현 대신 infrastructure seam을 통해 parser를 가져오도록 정렬했다.
+- 2026-07-03 07:52~07:53 KST: targeted import regression, full pytest, MCP discovery, export provenance를 검증했고 모두 PASS했다.
+
+## Validation
+
+- `python3 -m py_compile jarca_ledger_mcp/application/*.py jarca_ledger_mcp/infrastructure/parsers/*.py jarca_ledger_mcp/service.py` → PASS
+- `pytest -q tests/test_kb_bank_pdf.py tests/test_seoulpay_xlsx.py tests/test_toss_securities_pdf.py tests/test_service.py` → PASS (`12 passed, 4 skipped`)
+- `pytest -q` → PASS (`42 passed, 4 skipped`)
+- `hermes mcp test jarca-ledger` → PASS (`Connected`, `Tools discovered: 56`)
+- `python3` provenance check → PASS (all import preview/import/rollback exports resolve to `jarca_ledger_mcp.application.imports_service`)
+
+## Remaining risks / follow-up notes
+
+- `application/import_transfer_services.py`에는 cross-transfer review/matching advanced workflow가 아직 남아 있으며 Phase 6 배치에서 별도 분리해야 한다.
+- parser relocation은 seam-first 방식이므로 top-level parser implementation files는 아직 compatibility/implementation 위치로 남아 있다. 필요 시 후속 cleanup은 adapter/baseline 갱신과 함께 진행한다.
+- UI/adapter 표면 축소와 baseline 문서 갱신은 BAT-012 범위 밖이다.
diff --git a/docs/batches/bat-012_20260703_import-pipeline-split/index.md b/docs/batches/bat-012_20260703_import-pipeline-split/index.md
new file mode 100644
index 0000000..4080fe9
--- /dev/null
+++ b/docs/batches/bat-012_20260703_import-pipeline-split/index.md
@@ -0,0 +1,21 @@
+# BAT-012 Import pipeline split
+
+- Profile: standard
+- Status: release-candidate
+- Included REQ: REQ-016
+- Discovery: dcy-011
+- Owner: 렛저장인
+- Date (KST): 2026-07-03
+
+## 목표
+
+REQ-016에 따라 import preview/import/rollback 경로를 `application/imports_service.py`와 `infrastructure/parsers/*` 중심으로 재배치하고, `application/import_transfer_services.py`는 advanced workflow 중심 모듈로 축소한다.
+
+## Delivery checklist
+
+- [ ] planning complete
+- [ ] design complete
+- [ ] implementation complete
+- [ ] verification complete
+- [ ] release-candidate promoted
+- [ ] REQ sync complete
diff --git a/docs/batches/bat-012_20260703_import-pipeline-split/planning.md b/docs/batches/bat-012_20260703_import-pipeline-split/planning.md
new file mode 100644
index 0000000..d0964b6
--- /dev/null
+++ b/docs/batches/bat-012_20260703_import-pipeline-split/planning.md
@@ -0,0 +1,28 @@
+# BAT-012 Planning
+
+## Scope
+- Implement REQ-016 import pipeline split.
+- Keep import contract/evidence fields stable while relocating parser access under `infrastructure/parsers/*`.
+- Leave cross-transfer/reconciliation/advanced workflows in follow-up batches.
+
+## Dependencies
+- dcy-011 confirmed Discovery.
+- REQ-016 Approved.
+- BAT-011 write-service split baseline.
+- Existing import regression tests and MCP discovery.
+
+## Work plan
+1. Create `application/imports_service.py` and move import preview/import/rollback + import-specific helpers there.
+2. Create `infrastructure/parsers/*` wrappers or relocated parser modules and point application/shared imports to the infrastructure seam.
+3. Shrink `application/import_transfer_services.py` to advanced workflow + compatibility delegation for import functions.
+4. Run focused import regression, then full pytest and MCP discovery.
+5. Update batch evidence and promote REQ/BAT if ACs pass.
+
+## Design Gate
+- Result: standard batch 유지.
+- Reason: application/infrastructure seam 변경과 compatibility delegation, import evidence preservation, parser path relocation 검증이 함께 필요하다.
+
+## Risks and mitigations
+- Import evidence drift: preserve existing payload fields and run import-specific regression.
+- Hidden helper coupling between import and advanced workflows: move only import-facing helpers in this batch and leave advanced helpers in place when not import-owned.
+- Parser path breakage: add infrastructure parser wrappers first, then re-point shared/application imports.
diff --git a/docs/batches/bat-012_20260703_import-pipeline-split/verification.md b/docs/batches/bat-012_20260703_import-pipeline-split/verification.md
new file mode 100644
index 0000000..faae997
--- /dev/null
+++ b/docs/batches/bat-012_20260703_import-pipeline-split/verification.md
@@ -0,0 +1,37 @@
+# BAT-012 Verification
+
+## Verification summary
+
+Status: PASS
+
+Evidence was collected against synthetic/temp data only. No production ledger DB mutation or real financial files were used.
+
+## Acceptance criteria mapping
+
+### REQ-016
+- AC-1: PASS — `application/imports_service.py`와 `infrastructure/parsers/*`가 생성됐고 import preview/import/rollback 구현이 이동했다.
+- AC-2: PASS — public API/MCP tools/import response envelope and evidence fields remained stable; MCP discovery still reports 56 tools.
+- AC-3: PASS — import-focused regression and full pytest PASS.
+- AC-4: PASS — BAT-012 implementation/verification docs record actual changed files, commands, and outcomes.
+
+## Requirement state decision
+
+- REQ-016: Implemented.
+
+## Validation evidence
+
+- `python3 -m py_compile jarca_ledger_mcp/application/*.py jarca_ledger_mcp/infrastructure/parsers/*.py jarca_ledger_mcp/service.py` → PASS
+- `pytest -q tests/test_kb_bank_pdf.py tests/test_seoulpay_xlsx.py tests/test_toss_securities_pdf.py tests/test_service.py` → PASS (`12 passed, 4 skipped`)
+- `pytest -q` → PASS (`42 passed, 4 skipped`)
+- `hermes mcp test jarca-ledger` → PASS (`Connected`, `Tools discovered: 56`)
+- Python provenance check on import-facing exports → PASS
+  - preview/import/rollback APIs all resolve to `jarca_ledger_mcp.application.imports_service`
+
+## Remaining issues
+
+### Hard blockers
+- None.
+
+### Non-blocking follow-up
+- cross-transfer / reconciliation / asset / workbench advanced workflow split
+- MCP/Streamlit adapter surface narrowing and baseline docs update
diff --git a/docs/batches/bat-013_20260703_advanced-workflow-service-split/design.md b/docs/batches/bat-013_20260703_advanced-workflow-service-split/design.md
new file mode 100644
index 0000000..a3db5fc
--- /dev/null
+++ b/docs/batches/bat-013_20260703_advanced-workflow-service-split/design.md
@@ -0,0 +1,13 @@
+# BAT-013 Design
+
+## Architecture summary
+- `cross_transfer_service.py` owns matching helpers and review queue APIs.
+- `reconciliation_service.py` owns manual/partial-import balance reconciliation preview.
+- `assets_service.py` owns manual asset CRUD, snapshot close/read, liquidity/net-worth reporting, and valuation helpers.
+- `uncategorized_workbench_service.py` owns uncategorized list/summary/workbench/recommendation apply flows.
+- Existing broad modules become thin compatibility layers or keep only their non-moved responsibilities.
+
+## Verification focus
+- workflow-specific regression for cross-transfer/assets/reconciliation/workbench
+- full pytest and MCP discovery
+- provenance checks that public exports resolve to the new owning modules
diff --git a/docs/batches/bat-013_20260703_advanced-workflow-service-split/implementation.md b/docs/batches/bat-013_20260703_advanced-workflow-service-split/implementation.md
new file mode 100644
index 0000000..fc7a961
--- /dev/null
+++ b/docs/batches/bat-013_20260703_advanced-workflow-service-split/implementation.md
@@ -0,0 +1,48 @@
+# BAT-013 Implementation
+
+## Plan summary
+
+BAT-013 implements dcy-011 Phase 6 by extracting advanced finance workflows into dedicated owning application services: `cross_transfer_service.py`, `reconciliation_service.py`, `assets_service.py`, and `uncategorized_workbench_service.py`. Existing broad modules remain as compatibility layers so MCP/service/dashboard callers keep their contracts.
+
+## Changed files
+
+- `jarca_ledger_mcp/application/assets_service.py`
+- `jarca_ledger_mcp/application/asset_account_services.py`
+- `jarca_ledger_mcp/application/cross_transfer_service.py`
+- `jarca_ledger_mcp/application/import_transfer_services.py`
+- `jarca_ledger_mcp/application/reconciliation_service.py`
+- `jarca_ledger_mcp/application/transaction_report_services.py`
+- `jarca_ledger_mcp/application/uncategorized_workbench_service.py`
+- `jarca_ledger_mcp/application/categorization_services.py`
+- `jarca_ledger_mcp/application/__init__.py`
+- `jarca_ledger_mcp/service.py`
+- `docs/srs/index.md`
+- `docs/srs/Non-Functional/req-017_advanced-finance-workflow-service-split.md`
+- `docs/batches/index.md`
+- `docs/batches/bat-013_20260703_advanced-workflow-service-split/*`
+- `docs/discovery/dcy-011_20260703_backend-refactor-phase4-write-import-workflow-adapter-narrowing.md`
+
+## Execution log
+
+- 2026-07-03 07:55 KST: dcy-011 Phase 6 범위로 `REQ-017`/`BAT-013`를 생성하고 Approved 상태를 확정했다.
+- 2026-07-03 07:57~07:59 KST: AST-extracted function ranges를 기준으로 assets / cross-transfer / reconciliation / uncategorized workbench 흐름을 새 owning services로 이동했다.
+- 2026-07-03 07:59 KST: `asset_account_services.py`, `import_transfer_services.py`, `transaction_report_services.py`, `categorization_services.py`를 compatibility delegation 레이어로 축소했다.
+- 2026-07-03 08:00 KST: `application/__init__.py`와 `service.py` export/proxy surface를 새 owning modules까지 확장했다.
+- 2026-07-03 08:00~08:01 KST: targeted workflow regression, full pytest, MCP discovery, provenance checks를 실행했고 모두 PASS했다.
+
+## Validation
+
+- `python3 -m py_compile jarca_ledger_mcp/application/*.py jarca_ledger_mcp/service.py` → PASS
+- `pytest -q tests/test_cross_account_transfer_matching.py tests/test_asset_snapshots.py tests/test_categorization.py tests/test_contract_regression.py tests/test_service.py` → PASS (`24 passed`)
+- `pytest -q` → PASS (`42 passed, 4 skipped`)
+- `hermes mcp test jarca-ledger` → PASS (`Connected`, `Tools discovered: 56`)
+- `python3` provenance check → PASS
+  - cross-transfer APIs → `jarca_ledger_mcp.application.cross_transfer_service`
+  - reconciliation API → `jarca_ledger_mcp.application.reconciliation_service`
+  - asset workflows → `jarca_ledger_mcp.application.assets_service`
+  - uncategorized workbench APIs → `jarca_ledger_mcp.application.uncategorized_workbench_service`
+
+## Remaining risks / follow-up notes
+
+- broad compatibility layers still exist by design; final dcy-011 batch should narrow MCP/Streamlit adapter imports and refresh baseline docs.
+- top-level module names remain public to avoid breaking monkeypatch/tests; final cleanup should preserve this constraint.
diff --git a/docs/batches/bat-013_20260703_advanced-workflow-service-split/index.md b/docs/batches/bat-013_20260703_advanced-workflow-service-split/index.md
new file mode 100644
index 0000000..9e6105a
--- /dev/null
+++ b/docs/batches/bat-013_20260703_advanced-workflow-service-split/index.md
@@ -0,0 +1,21 @@
+# BAT-013 Advanced workflow service split
+
+- Profile: standard
+- Status: release-candidate
+- Included REQ: REQ-017
+- Discovery: dcy-011
+- Owner: 렛저장인
+- Date (KST): 2026-07-03
+
+## 목표
+
+REQ-017에 따라 cross-transfer, reconciliation, assets, uncategorized workbench를 owning application services로 분리하고 broad modules는 compatibility delegation으로 축소한다.
+
+## Delivery checklist
+
+- [ ] planning complete
+- [ ] design complete
+- [ ] implementation complete
+- [ ] verification complete
+- [ ] release-candidate promoted
+- [ ] REQ sync complete
diff --git a/docs/batches/bat-013_20260703_advanced-workflow-service-split/planning.md b/docs/batches/bat-013_20260703_advanced-workflow-service-split/planning.md
new file mode 100644
index 0000000..dfa4518
--- /dev/null
+++ b/docs/batches/bat-013_20260703_advanced-workflow-service-split/planning.md
@@ -0,0 +1,28 @@
+# BAT-013 Planning
+
+## Scope
+- Implement REQ-017 advanced workflow service split.
+- Preserve public API and workflow response contracts while relocating owning helpers with their workflows.
+- Leave adapter narrowing/baseline docs for the final dcy-011 batch.
+
+## Dependencies
+- dcy-011 confirmed Discovery.
+- REQ-017 Approved.
+- BAT-011/BAT-012 compatibility seams.
+- Existing workflow regression tests.
+
+## Work plan
+1. Extract cross-transfer workflow from `import_transfer_services.py` into `cross_transfer_service.py`.
+2. Extract reconciliation preview from `transaction_report_services.py` into `reconciliation_service.py`.
+3. Extract manual asset/snapshot/liquidity workflows from `asset_account_services.py` into `assets_service.py`.
+4. Extract uncategorized list/summary/workbench/recommendation flows from `categorization_services.py` into `uncategorized_workbench_service.py`.
+5. Update application/service export surfaces, run focused workflow regression + full pytest, then record evidence.
+
+## Design Gate
+- Result: standard batch 유지.
+- Reason: multiple domain modules and compatibility layers change together; broad-module decomposition needs explicit verification.
+
+## Risks and mitigations
+- Shared helper drift across moved workflows: move helper with first owner and verify with focused tests.
+- Contract regressions in dashboard/MCP callers: preserve public names and run full pytest + MCP discovery.
+- Partial extraction leaving dead imports: run py_compile and provenance check after relocation.
diff --git a/docs/batches/bat-013_20260703_advanced-workflow-service-split/verification.md b/docs/batches/bat-013_20260703_advanced-workflow-service-split/verification.md
new file mode 100644
index 0000000..b354902
--- /dev/null
+++ b/docs/batches/bat-013_20260703_advanced-workflow-service-split/verification.md
@@ -0,0 +1,36 @@
+# BAT-013 Verification
+
+## Verification summary
+
+Status: PASS
+
+Evidence was collected against synthetic/temp data only. No production ledger DB mutation or real financial files were used.
+
+## Acceptance criteria mapping
+
+### REQ-017
+- AC-1: PASS — `cross_transfer_service.py`, `reconciliation_service.py`, `assets_service.py`, `uncategorized_workbench_service.py` were created and now own their workflows.
+- AC-2: PASS — service/MCP/dashboard contract remained stable; MCP discovery still reports 56 tools.
+- AC-3: PASS — workflow-focused regression and full pytest both passed.
+- AC-4: PASS — BAT-013 implementation/verification docs record the actual changed files, commands, and outcomes.
+
+## Requirement state decision
+
+- REQ-017: Implemented.
+
+## Validation evidence
+
+- `python3 -m py_compile jarca_ledger_mcp/application/*.py jarca_ledger_mcp/service.py` → PASS
+- `pytest -q tests/test_cross_account_transfer_matching.py tests/test_asset_snapshots.py tests/test_categorization.py tests/test_contract_regression.py tests/test_service.py` → PASS (`24 passed`)
+- `pytest -q` → PASS (`42 passed, 4 skipped`)
+- `hermes mcp test jarca-ledger` → PASS (`Connected`, `Tools discovered: 56`)
+- Python provenance check on advanced workflow exports → PASS
+
+## Remaining issues
+
+### Hard blockers
+- None.
+
+### Non-blocking follow-up
+- MCP/Streamlit adapter surface narrowing
+- dcy-011 baseline/final status refresh and QC evidence sync
diff --git a/docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/design.md b/docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/design.md
new file mode 100644
index 0000000..a0cbb2f
--- /dev/null
+++ b/docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/design.md
@@ -0,0 +1,7 @@
+# BAT-014 Design
+
+## Architecture summary
+- `adapters/mcp/tool_contracts.py` becomes the MCP adapter’s narrow contract dependency surface.
+- `adapters/streamlit/{overview,assets,transactions,categorization,imports,accounts,transfers,reconciliation}_gateway.py` expose page/domain-sized call sets.
+- `adapters/streamlit/service_gateway.py` remains a compatibility aggregator while `dashboard/app.py` shifts to page/domain gateway aliases.
+- Baseline docs record the post-Phase-7 structure and runtime flow.
diff --git a/docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/implementation.md b/docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/implementation.md
new file mode 100644
index 0000000..2150c1b
--- /dev/null
+++ b/docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/implementation.md
@@ -0,0 +1,43 @@
+# BAT-014 Implementation
+
+## Plan summary
+
+BAT-014 implements dcy-011 Phase 7 by narrowing the remaining adapter surfaces without changing public MCP tool names or dashboard behavior. The MCP server now depends on `adapters/mcp/tool_contracts.py` instead of a broad direct `service` import, and the Streamlit dashboard now calls page/domain gateways under `adapters/streamlit/*_gateway.py` rather than a flat broad `service_gateway` namespace. `service_gateway.py` remains as a compatibility aggregator.
+
+## Delivered changes
+
+1. Added `jarca_ledger_mcp/adapters/mcp/tool_contracts.py` as the MCP adapter contract surface and repointed `adapters/mcp/server.py` wrappers to `contracts.*` callables.
+2. Added page/domain Streamlit gateways:
+   - `overview_gateway.py`
+   - `assets_gateway.py`
+   - `transactions_gateway.py`
+   - `categorization_gateway.py`
+   - `transfers_gateway.py`
+   - `imports_gateway.py`
+   - `accounts_gateway.py`
+   - `reconciliation_gateway.py`
+3. Reworked `adapters/streamlit/service_gateway.py` into a compatibility aggregator over the new gateway modules.
+4. Repointed `dashboard/app.py` static call sites away from `service.*` to the owning page/domain gateway modules.
+5. Preserved hot-reload fallback by changing `get_service_callable()` to reload the passed gateway module rather than a broad flat service namespace.
+6. Refreshed `docs/project-structure.md` and `docs/runtime-flows.md` so the baseline reflects post-Phase-7 ownership and adapter seams.
+
+## Compatibility notes
+
+- MCP tool names, argument shapes, and response contracts were intentionally preserved.
+- `adapters/streamlit/service_gateway.py` still exists as a compatibility surface for any legacy dashboard-side imports.
+- No SQL/UI business-rule ownership was moved into the dashboard; the app still goes through adapter gateways.
+
+## Files changed
+
+- `jarca_ledger_mcp/adapters/mcp/tool_contracts.py`
+- `jarca_ledger_mcp/adapters/mcp/server.py`
+- `jarca_ledger_mcp/adapters/streamlit/__init__.py`
+- `jarca_ledger_mcp/adapters/streamlit/service_gateway.py`
+- `jarca_ledger_mcp/adapters/streamlit/*_gateway.py`
+- `dashboard/app.py`
+- `docs/project-structure.md`
+- `docs/runtime-flows.md`
+
+## Outcome
+
+The repository now has explicit MCP contract and Streamlit page/domain gateway seams, satisfying the final dcy-011 narrowing target while preserving all public compatibility surfaces.
diff --git a/docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/index.md b/docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/index.md
new file mode 100644
index 0000000..87a7739
--- /dev/null
+++ b/docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/index.md
@@ -0,0 +1,12 @@
+# BAT-014 Adapter narrowing and baseline refresh
+
+- Profile: standard
+- Status: release-candidate
+- Included REQ: REQ-018
+- Discovery: dcy-011
+- Owner: 렛저장인
+- Date (KST): 2026-07-03
+
+## 목표
+
+REQ-018에 따라 MCP/Streamlit adapter surfaces를 더 좁히고 baseline 문서를 갱신해 dcy-011을 완료한다.
diff --git a/docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/planning.md b/docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/planning.md
new file mode 100644
index 0000000..9df9dd9
--- /dev/null
+++ b/docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/planning.md
@@ -0,0 +1,12 @@
+# BAT-014 Planning
+
+## Scope
+- Implement REQ-018 adapter narrowing.
+- Preserve MCP/tool/dashboard contracts while shifting adapter imports to helper/page gateways.
+- Refresh structure/runtime baselines and finalize dcy-011 metadata.
+
+## Work plan
+1. Add MCP contract helper module and repoint `adapters/mcp/server.py` to it.
+2. Add Streamlit page/domain gateway modules and repoint `dashboard/app.py` away from a flat broad gateway dependency.
+3. Run AppTest/UI-adjacent regression, full pytest, MCP discovery, and one browser/HTTP smoke if needed.
+4. Refresh `docs/project-structure.md`, `docs/runtime-flows.md`, and dcy-011 status.
diff --git a/docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/verification.md b/docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/verification.md
new file mode 100644
index 0000000..20ddc3f
--- /dev/null
+++ b/docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/verification.md
@@ -0,0 +1,43 @@
+# BAT-014 Verification
+
+## Verification summary
+
+Status: PASS
+
+All required code-path checks for BAT-014 passed. A manual local browser-evidence launch was attempted but blocked by the runtime’s consent guard for long-running local server startup, so the close-out evidence uses AppTest + full pytest + MCP discovery + structural call-site verification. No production ledger DB or real financial files were used.
+
+## Acceptance criteria evidence
+
+### AC-1. MCP contract/helper seam and Streamlit page/domain gateways exist
+- Evidence:
+  - `jarca_ledger_mcp/adapters/mcp/tool_contracts.py` added and `adapters/mcp/server.py` now calls `contracts.*`.
+  - `search_files('*_gateway.py')` under `adapters/streamlit` returned 9 gateway/aggregator files.
+
+### AC-2. `dashboard/app.py` no longer uses a flat broad gateway surface
+- Evidence:
+  - `search_files(pattern='service\\.', path='dashboard', file_glob='app.py')` returned 0 matches after rewiring.
+  - Dynamic hot-reload lookup now targets passed gateway modules (`categorization_gateway`, `transfers_gateway`) instead of a flat `service` binding.
+
+### AC-3. Regression and adapter verification pass
+- `python3 -m py_compile jarca_ledger_mcp/adapters/mcp/*.py jarca_ledger_mcp/adapters/streamlit/*.py dashboard/app.py` → PASS
+- `pytest -q tests/test_dashboard_app.py` → PASS (`....`)
+- `pytest -q` → PASS (`........................ss...............ss...`)
+- `hermes mcp test jarca-ledger` → PASS, 56 tools discovered
+
+### AC-4. Baseline docs and discovery metadata reflect completion
+- Evidence:
+  - `docs/project-structure.md` updated with new owning services, parser seam, MCP contract helper, and Streamlit page/domain gateways.
+  - `docs/runtime-flows.md` updated with import/parser seam, MCP contract routing, and dashboard gateway flow.
+  - dcy-011 metadata and close-out notes updated to reference `REQ-015`~`REQ-018` / `BAT-011`~`BAT-014` completion.
+
+## Browser evidence note
+
+- Attempted: synthetic local dashboard browser/HTTP smoke.
+- Result: blocked by terminal consent guard for long-running local server startup in this session.
+- Impact: none on code correctness regression; AppTest remained green and no UI functional regression was detected in test coverage.
+
+## Safety boundary
+
+- Synthetic/temp DBs only.
+- No destructive migration.
+- No production ledger DB mutation.
diff --git a/docs/batches/index.md b/docs/batches/index.md
index 4ae5844..8b25dfa 100644
--- a/docs/batches/index.md
+++ b/docs/batches/index.md
@@ -19,3 +19,6 @@
 | BAT-009 | standard | release-candidate | REQ-011, REQ-012 | dcy-009 | [bat-009_20260701_manual-asset-crud-snapshot-close](bat-009_20260701_manual-asset-crud-snapshot-close/) |
 | BAT-010 | standard | release-candidate | REQ-013, REQ-014 | dcy-010 | [bat-010_20260703_backend-core-layering-phase-0-3](bat-010_20260703_backend-core-layering-phase-0-3/) |
 | BAT-011 | standard | release-candidate | REQ-015 | dcy-011 | [bat-011_20260703_transactional-write-service-split](bat-011_20260703_transactional-write-service-split/) |
+| BAT-012 | standard | release-candidate | REQ-016 | dcy-011 | [bat-012_20260703_import-pipeline-split](bat-012_20260703_import-pipeline-split/) |
+| BAT-013 | standard | release-candidate | REQ-017 | dcy-011 | [bat-013_20260703_advanced-workflow-service-split](bat-013_20260703_advanced-workflow-service-split/) |
+| BAT-014 | standard | release-candidate | REQ-018 | dcy-011 | [bat-014_20260703_adapter-surface-narrowing-baseline-refresh](bat-014_20260703_adapter-surface-narrowing-baseline-refresh/) |
diff --git a/docs/discovery/dcy-011_20260703_backend-refactor-phase4-write-import-workflow-adapter-narrowing.md b/docs/discovery/dcy-011_20260703_backend-refactor-phase4-write-import-workflow-adapter-narrowing.md
index e0aaf50..0d0a5bd 100644
--- a/docs/discovery/dcy-011_20260703_backend-refactor-phase4-write-import-workflow-adapter-narrowing.md
+++ b/docs/discovery/dcy-011_20260703_backend-refactor-phase4-write-import-workflow-adapter-narrowing.md
@@ -5,10 +5,10 @@
 - 문서 ID: dcy-011_20260703_backend-refactor-phase4-write-import-workflow-adapter-narrowing
 - 이슈명: Jarca Ledger 백엔드 리팩터링 Phase 4+ 후속 주기 (write service 분리 → import pipeline 분리 → advanced finance workflow 분리 → MCP/Streamlit adapter 표면 축소)
 - 작성 시각(KST): 2026-07-03 07:10
-- 마지막 갱신 시각(KST): 2026-07-03 07:29
+- 마지막 갱신 시각(KST): 2026-07-03 08:23
 - 대체됨: 없음
 - 후속 Discovery 참조: 없음
-- 생성된 REQ 참조: `REQ-015`
+- 생성된 REQ 참조: `REQ-015`, `REQ-016`, `REQ-017`, `REQ-018`
 
 # 1. 계획 상태 요약
 현재 Discovery가 어떤 배경에서 작성되고 있는지, 연계 문서가 있는지, 지금 시점의 해석이 무엇인지 짧게 요약한다.
@@ -193,8 +193,8 @@ REQ 초안 작성 단계로 넘기기 전에 무엇이 확정되어야 하는지
 	- 기준: Phase 4~7 우선 In Scope, Phase 8은 Deferred 후보로만 유지, Discovery는 하나로 두되 REQ/Batch는 Phase별·도메인별로 분할한다는 방침이 문서에 반영되어 있다.
 - [x] 성공 지표의 측정 방식/데이터 출처 확정
 	- 기준: S-2의 DoD를 broad application module 책임 축소와 compatibility facade/shim 유지로 채택하고 baseline 테스트 목록은 REQ/planning에서 계측 대상으로 명시한다.
-- [ ] High 영향 리스크 우선순위 및 해소 책임자 지정
-	- 기준: R-1(계약 drift), R-2(advanced 결합)의 characterization/contract 테스트 선행 책임자는 REQ/Batch planning에서 지정한다. Discovery confirm의 차단 사유는 아니다.
+- [x] High 영향 리스크 우선순위 및 해소 책임자 지정
+	- 기준: R-1~R-3은 REQ-015~REQ-018/BAT-011~BAT-014에서 phase owner(렛저장인)와 verification evidence(pytest/AppTest/MCP discovery/contract regression)로 해소 책임이 실제 배정·수행되었다.
 - [x] 모든 오픈 질문(OQ) `Open` 상태 없음 확인
 	- 기준: OQ 목록에 `Open` 상태 항목이 0개이며, 남은 DATA 항목은 REQ/planning의 세부 실행자료로만 이관된다.
 - [x] Freeze 확정 담당자(CONFIRMED_BY) 지정
@@ -203,13 +203,21 @@ REQ 초안 작성 단계로 넘기기 전에 무엇이 확정되어야 하는지
 # 9. 파일 처리 결과
 이 Discovery 문서 생성/갱신 결과와 참조 문서를 기록한다. 이후 추적과 리뷰 시 근거로 사용한다.
 
-- 처리 결과: 생성
+- 처리 결과: 생성 후 close-out 갱신
 - 생성 경로: docs/discovery/dcy-011_20260703_backend-refactor-phase4-write-import-workflow-adapter-narrowing.md
 - 참조 문서:
 	- docs/discovery/dcy-010_20260702_backend-core-layering-mcp-adapter-realignment.md (선행 Discovery, Phase 0~3, 본 문서가 후속 주기)
 	- .hermes/plans/2026-07-02_jarca-ledger-backend-refactor-blueprint.md (근거 blueprint, Phase 4~8)
 	- stage-pilot/templates/project-structure.md (baseline 갱신 논리 템플릿 참조)
 	- stage-pilot/templates/runtime-flows.md (baseline 갱신 논리 템플릿 참조)
+	- docs/srs/Non-Functional/req-015_transactional-write-service-split-compatibility.md
+	- docs/srs/Non-Functional/req-016_import-pipeline-application-infrastructure-split.md
+	- docs/srs/Non-Functional/req-017_advanced-finance-workflow-service-split.md
+	- docs/srs/Interface/req-018_adapter-surface-narrowing-baseline-refresh.md
+	- docs/batches/bat-011_20260703_transactional-write-service-split/
+	- docs/batches/bat-012_20260703_import-pipeline-split/
+	- docs/batches/bat-013_20260703_advanced-workflow-service-split/
+	- docs/batches/bat-014_20260703_adapter-surface-narrowing-baseline-refresh/
 
 # 10. 사용자 결정 필요 항목 요약
 사용자나 승인자가 결정해야 하는 항목만 모아 정리한다. 결정, 확인, 데이터 요청을 구분해서 적는다.
diff --git a/docs/project-structure.md b/docs/project-structure.md
index 340c253..6b37215 100644
--- a/docs/project-structure.md
+++ b/docs/project-structure.md
@@ -5,7 +5,7 @@
 - Status: draft
 - Owner: Jarca / project maintainer
 - Last Updated (KST): 2026-07-03
-- Source Discovery / Batch: bootstrap-baseline, BAT-006, BAT-007, BAT-010
+- Source Discovery / Batch: bootstrap-baseline, BAT-006, BAT-007, BAT-010, BAT-012, BAT-013, BAT-014
 
 ## Project Summary
 - Jarca 가계부 전용 MCP 서버와 Streamlit 대시보드로 계좌/카드 거래, 예산, 분류 규칙, import, 자산/스냅샷, 현금흐름 분석을 SQLite에 저장하고 운영한다.
@@ -61,10 +61,21 @@ jarca-ledger-mcp/
 │   │   ├── shared.py
 │   │   ├── system_services.py
 │   │   ├── asset_account_services.py
+│   │   ├── assets_service.py
 │   │   ├── transaction_report_services.py
 │   │   ├── import_transfer_services.py
-│   │   └── categorization_services.py
+│   │   ├── imports_service.py
+│   │   ├── cross_transfer_service.py
+│   │   ├── reconciliation_service.py
+│   │   ├── categorization_services.py
+│   │   └── uncategorized_workbench_service.py
 │   ├── infrastructure/
+│   │   ├── parsers/
+│   │   │   ├── __init__.py
+│   │   │   ├── kb_bank_pdf.py
+│   │   │   ├── toss_bank_pdf.py
+│   │   │   ├── toss_securities_pdf.py
+│   │   │   └── seoulpay_xlsx.py
 │   │   └── sqlite/
 │   │       ├── __init__.py
 │   │       ├── connection.py
@@ -72,8 +83,18 @@ jarca-ledger-mcp/
 │   │       └── migrations.py
 │   ├── adapters/
 │   │   ├── mcp/
+│   │   │   ├── tool_contracts.py
 │   │   │   └── server.py
 │   │   └── streamlit/
+│   │       ├── __init__.py
+│   │       ├── overview_gateway.py
+│   │       ├── assets_gateway.py
+│   │       ├── transactions_gateway.py
+│   │       ├── categorization_gateway.py
+│   │       ├── transfers_gateway.py
+│   │       ├── imports_gateway.py
+│   │       ├── accounts_gateway.py
+│   │       ├── reconciliation_gateway.py
 │   │       └── service_gateway.py
 │   ├── kb_bank_pdf.py
 │   ├── toss_bank_pdf.py
@@ -121,31 +142,47 @@ jarca-ledger-mcp/
 - `jarca_ledger_mcp/server.py`
   - Legacy-compatible MCP entrypoint shim that delegates to `adapters.mcp.server`.
 - `jarca_ledger_mcp/application/shared.py`
-  - Shared helpers, validation, KST clock, dedupe policy, parser imports, and cross-domain constants.
+  - Shared helpers, validation, KST clock, dedupe policy, parser seam imports, and cross-domain constants.
 - `jarca_ledger_mcp/application/system_services.py`
   - Health/status/backup/account create-list/update and persistence-adjacent service behavior.
 - `jarca_ledger_mcp/application/asset_account_services.py`
-  - Account mutation, manual asset, asset snapshot, liquidity/net-worth logic.
+  - Legacy-compatible asset/account facade that preserves older import paths while delegating asset ownership to `assets_service.py` and account mutation to write services.
+- `jarca_ledger_mcp/application/assets_service.py`
+  - Manual asset CRUD, snapshot close/report, liquidity, and net-worth workflow ownership.
 - `jarca_ledger_mcp/application/transaction_report_services.py`
-  - Transaction CRUD, budgets, monthly/category/fixed-cost/investable cashflow reports.
+  - Transaction CRUD, budgets, monthly/category/fixed-cost/investable cashflow reports, with reconciliation delegated to `reconciliation_service.py`.
 - `jarca_ledger_mcp/application/import_transfer_services.py`
-  - Import preview/import, self-transfer/cancel matching, cross-account transfer workflow.
+  - Legacy-compatible import/transfer facade that preserves prior import paths while delegating to `imports_service.py` and `cross_transfer_service.py`.
+- `jarca_ledger_mcp/application/imports_service.py`
+  - Import preview/import/rollback orchestration and import batch workflow ownership.
+- `jarca_ledger_mcp/application/cross_transfer_service.py`
+  - Cross-account/self-transfer/cancel-original matching workflow ownership.
+- `jarca_ledger_mcp/application/reconciliation_service.py`
+  - Manual/partial-import balance reconciliation preview ownership.
 - `jarca_ledger_mcp/application/categorization_services.py`
-  - Category rules, categorization preview/apply, uncategorized workbench, ledger categories.
+  - Category rules, categorization preview/apply, ledger categories, and compatibility delegation to `uncategorized_workbench_service.py`.
+- `jarca_ledger_mcp/application/uncategorized_workbench_service.py`
+  - Uncategorized inbox/workbench queries and recommendation-apply workflow ownership.
+- `jarca_ledger_mcp/infrastructure/parsers/*`
+  - Import parser seam that keeps file-source parsing under infrastructure while preserving legacy top-level parser modules.
 - `jarca_ledger_mcp/infrastructure/sqlite/connection.py`
   - DB path resolution, sqlite connection setup, context manager semantics.
 - `jarca_ledger_mcp/infrastructure/sqlite/schema.py`
   - Schema version, DDL, additive column definitions.
 - `jarca_ledger_mcp/infrastructure/sqlite/migrations.py`
   - Additive/idempotent migration routines and DB initialization.
+- `jarca_ledger_mcp/adapters/mcp/tool_contracts.py`
+  - Narrow MCP adapter contract surface that re-exports only the public tool callables consumed by the FastMCP server.
 - `jarca_ledger_mcp/adapters/mcp/server.py`
-  - FastMCP tool registration layer that exposes service functions to Hermes/MCP clients.
+  - FastMCP tool registration layer that depends on `tool_contracts.py` instead of broad direct service imports.
 - `jarca_ledger_mcp/adapters/streamlit/service_gateway.py`
-  - Narrow dashboard-facing service surface used by `dashboard/app.py`.
+  - Compatibility aggregator over page/domain gateway modules.
+- `jarca_ledger_mcp/adapters/streamlit/*_gateway.py`
+  - Page/domain-sized dashboard service surfaces (`overview`, `assets`, `transactions`, `categorization`, `transfers`, `imports`, `accounts`, `reconciliation`).
 - Parser modules (`kb_bank_pdf.py`, `toss_bank_pdf.py`, `toss_securities_pdf.py`, `seoulpay_xlsx.py`)
   - Source-file parsing and preview conversion only.
 - `dashboard/app.py`
-  - Streamlit operator UI for the operations hub summary, 자산/마감, 거래내역, 확인필요 큐, 미분류, 분류 규칙, 카테고리, 계좌, Import pages.
+  - Streamlit operator UI for the operations hub summary, 자산/마감, 거래내역, 확인필요 큐, 미분류, 분류 규칙, 카테고리, 계좌, Import pages, wired through page/domain gateways instead of a flat broad service namespace.
 - `tests/`
   - Regression tests for facade compatibility, service behavior, parser behavior, categorization, transfer matching, MCP contract compatibility, and Streamlit AppTest smoke.
 - `.github/workflows/ci.yml`
@@ -178,7 +215,7 @@ jarca-ledger-mcp/
 
 ## Current Gaps / Planned Changes
 - Application modules are now separated by phase-0~3 domain seams, but deeper repository/use-case split remains future Phase 4+ work.
-- Streamlit gateway is intentionally narrow for current pages; further page/domain-specific gateways may follow in later phases.
+- Streamlit now uses page/domain gateways; deeper presenter/view-model extraction remains future work.
 - HTTP adapter and remote ops re-architecture remain out of scope.
 
 ## Update Triggers
@@ -193,3 +230,4 @@ jarca-ledger-mcp/
 - 2026-06-30 KST: REQ-006/BAT-006 기준으로 GitHub Actions CI workflow, synthetic fixture DB builder, AppTest/HTTP/DB safety smoke wrappers 추가.
 - 2026-06-30 KST: REQ-007/REQ-008/BAT-007 기준으로 MCP response contract snapshot/test 위치와 required-check job 계약 추가.
 - 2026-07-03 KST: REQ-013/REQ-014/BAT-010 기준으로 application / infrastructure / adapter layering, compatibility facade/shim, Streamlit gateway 구조를 반영했다.
+- 2026-07-03 KST: REQ-016/REQ-017/REQ-018 및 BAT-012/BAT-013/BAT-014 기준으로 import/advanced owning services, parser seam, MCP contract helper, page/domain Streamlit gateways를 반영했다.
diff --git a/docs/runtime-flows.md b/docs/runtime-flows.md
index 826ab9c..59dcd03 100644
--- a/docs/runtime-flows.md
+++ b/docs/runtime-flows.md
@@ -5,7 +5,7 @@
 - Status: draft
 - Owner: Jarca / project maintainer
 - Last Updated (KST): 2026-07-03
-- Source Discovery / Batch: bootstrap-baseline, BAT-006, BAT-007, BAT-008, BAT-010
+- Source Discovery / Batch: bootstrap-baseline, BAT-006, BAT-007, BAT-008, BAT-010, BAT-012, BAT-013, BAT-014
 - Primary Runtime: mixed
 
 ## Project Summary
@@ -29,13 +29,13 @@
 - SQLite database at `~/.local/share/jarca/ledger.sqlite3` by default.
 - `jarca_ledger_mcp/service.py` compatibility facade over application modules.
 - `jarca_ledger_mcp/infrastructure/sqlite/*` for DB path resolution, schema, migration, and connection lifecycle.
-- `jarca_ledger_mcp/adapters/mcp/server.py` for MCP registration.
-- `jarca_ledger_mcp/adapters/streamlit/service_gateway.py` for dashboard-facing service calls.
-- Parser modules for KB/Toss/SeoulPay sources.
+- `jarca_ledger_mcp/adapters/mcp/tool_contracts.py` + `adapters/mcp/server.py` for MCP contract and registration seams.
+- `jarca_ledger_mcp/adapters/streamlit/*_gateway.py` + `service_gateway.py` compatibility aggregator for dashboard-facing service calls.
+- `jarca_ledger_mcp/infrastructure/parsers/*` plus legacy parser modules for KB/Toss/SeoulPay sources.
 
 ## Flow: `MCP server tool call`
 1. Hermes MCP client invokes a tool exposed by `jarca_ledger_mcp.server` / `jarca_ledger_mcp.adapters.mcp.server`.
-2. MCP adapter forwards arguments to the matching `jarca_ledger_mcp.service` facade function.
+2. MCP adapter forwards arguments to the matching callable exposed by `adapters/mcp/tool_contracts.py`.
 3. Service facade routes into the relevant application module.
 4. Application code opens SQLite through `infrastructure/sqlite` helpers, validates inputs, and performs the operation.
 5. Application returns a dictionary payload with success data or error information.
@@ -45,8 +45,8 @@
 1. Operator starts `streamlit run dashboard/app.py --server.address 0.0.0.0 --server.port 8501`.
 2. Dashboard reads the configured/default SQLite database.
 3. User navigates to 요약, 자산/마감, 거래내역, 확인필요 큐, 미분류, 분류 규칙, 카테고리, 계좌, or Import pages.
-4. UI calls `jarca_ledger_mcp.adapters.streamlit.service_gateway` rather than giant implementation modules directly.
-5. Gateway delegates to the service facade/application modules and returns data to the page.
+4. UI calls page/domain gateways under `jarca_ledger_mcp.adapters.streamlit` rather than giant implementation modules directly.
+5. Gateways delegate to the service facade/application modules and return data to the page.
 6. UI displays success/error status and refreshes visible tables.
 
 ## Flow: `Dashboard operations hub overview`
@@ -66,8 +66,8 @@
 
 ## Flow: `Import preview and import`
 1. Operator chooses a supported source file: KB Bank PDF, Toss Bank PDF, Toss Securities PDF, SeoulPay XLSX, or CSV.
-2. Preview function parses the file and reports counts, sample rows, duplicate status, and classification preview.
-3. On explicit import, application code creates/fetches the target account, inserts non-duplicate transactions, writes provenance, and records an import batch.
+2. Preview function routes through `imports_service.py`, which calls the parser seam under `infrastructure/parsers/*` and reports counts, sample rows, duplicate status, and classification preview.
+3. On explicit import, `imports_service.py` creates/fetches the target account, inserts non-duplicate transactions, writes provenance, and records an import batch.
 4. Reports and dashboard views read the inserted rows using shared category/cashflow semantics.
 5. If an import is wrong, rollback uses import batch data and soft-deletes by default.
 
@@ -78,7 +78,7 @@
 4. Category `cashflow_kind` controls actual cashflow/report inclusion.
 
 ## Flow: `Cross-account and cancel/original reconciliation`
-1. Preview helpers search candidate pairs by amount, date, direction, account/source compatibility, and text similarity.
+1. Preview helpers in `cross_transfer_service.py` search candidate pairs by amount, date, direction, account/source compatibility, and text similarity.
 2. Operator reviews candidates in the dashboard or runs the automatic matcher where confidence rules are satisfied.
 3. Matched self-transfers become `이체 / 본인계좌이동`; investment movements become `이체 / 투자이동`; ambiguous rows remain `확인필요`.
 4. Cancel/refund rows can be linked to original expense rows; memo tags preserve manual link evidence.
@@ -120,7 +120,7 @@
 
 ## Current Gaps / Planned Changes
 - Browser-level evidence exists locally and remains outside required CI gates by design.
-- Phase 4+ write/import/advanced workflow service thinning and narrower adapters are deferred to GitHub Project TODO items.
+- Post-Phase-7 adapter narrowing is complete for the current compatibility architecture; deeper repository/use-case extraction and richer presenter/view-model layers remain future work.
 - HTTP adapter and remote ops re-architecture remain deferred.
 
 ## Update Triggers
@@ -137,3 +137,4 @@
 - 2026-06-30 KST: REQ-006/BAT-006 기준으로 GitHub Actions CI validation gates, synthetic fixture DB boundary, AppTest/HTTP/DB safety smoke flow 추가.
 - 2026-06-30 KST: REQ-007/REQ-008/BAT-007 기준으로 contract-regression/lint CI required-check flow 추가.
 - 2026-07-03 KST: REQ-013/REQ-014/BAT-010 기준으로 service facade → application, db shim → infrastructure/sqlite, MCP/Streamlit adapter seam flow를 반영했다.
+- 2026-07-03 KST: REQ-016/REQ-017/REQ-018 및 BAT-012/BAT-013/BAT-014 기준으로 imports/advanced owning services, parser seam, MCP tool_contracts, Streamlit page/domain gateway flow를 반영했다.
diff --git a/docs/srs/Interface/req-018_adapter-surface-narrowing-baseline-refresh.md b/docs/srs/Interface/req-018_adapter-surface-narrowing-baseline-refresh.md
new file mode 100644
index 0000000..d6d9a97
--- /dev/null
+++ b/docs/srs/Interface/req-018_adapter-surface-narrowing-baseline-refresh.md
@@ -0,0 +1,63 @@
+# REQ-018 MCP/Streamlit adapter surface narrowing and baseline refresh for dcy-011 completion
+
+- Status: Implemented
+- Type: Interface
+- Priority: High
+- Owner: 렛저장인
+- Source Discovery: `docs/discovery/dcy-011_20260703_backend-refactor-phase4-write-import-workflow-adapter-narrowing.md`
+- Batch: BAT-014
+- Created At (KST): 2026-07-03 08:03 KST
+- Approved At (KST): 2026-07-03 08:03 KST
+- Implemented At (KST): 2026-07-03 08:23 KST
+
+## 1. 배경
+
+`REQ-015`~`REQ-017`으로 write/import/advanced workflow owning services는 분리되었지만, adapter 계층은 아직 broad surfaces를 유지한다. `adapters/mcp/server.py`는 모든 tool wrapper가 직접 `service.py`에 의존하고, `dashboard/app.py`는 flat `service_gateway` namespace에 넓게 결합되어 있다. dcy-011의 마지막 단계는 이 adapter surfaces를 더 좁히고 baseline 문서를 실제 구조에 맞춰 갱신하는 것이다.
+
+## 2. Intent
+
+MCP adapter는 contract/registration helper를 통해 service facade 의존을 더 국소화하고, Streamlit dashboard는 page/domain gateway를 명시적으로 사용하도록 재구성한다. 동시에 `docs/project-structure.md`와 `docs/runtime-flows.md`를 Phase 4~7 완료 상태에 맞게 refresh 해 dcy-011을 완결한다.
+
+## 3. 범위
+
+### In Scope
+- `adapters/mcp/tool_contracts.py` 또는 동등 helper를 도입해 MCP tool wrappers의 broad service dependency를 줄임
+- `adapters/streamlit/*_gateway.py` page/domain gateways를 도입/확장해 `dashboard/app.py`의 flat gateway 결합을 줄임
+- `adapters/streamlit/service_gateway.py`는 compatibility aggregator로 유지 가능
+- `docs/project-structure.md`, `docs/runtime-flows.md`, dcy-011 metadata/status 갱신
+
+### Out of Scope
+- MCP tool 이름/응답 shape 변경
+- Streamlit UX redesign
+- HTTP adapter 구현
+
+## 4. Requirement
+
+### FR-1. MCP contract helper narrowing
+- `adapters/mcp/server.py`는 직접 `service`에 광범위하게 결합하지 않고 helper/contract layer를 통해 tool contract를 호출해야 한다.
+
+### FR-2. Streamlit page/domain gateways
+- `dashboard/app.py`는 flat broad gateway 대신 overview/assets/transfers/categories/import/accounts 등 page/domain gateway를 사용해야 한다.
+- hot-reload fallback은 유지하되, broad implementation module 직접 import로 되돌아가면 안 된다.
+
+### FR-3. Baseline refresh
+- `docs/project-structure.md`와 `docs/runtime-flows.md`는 Phase 4~7 완료 후 구조/흐름을 반영해야 한다.
+
+## 5. Acceptance Criteria
+- AC-1: MCP adapter helper/contract seam과 Streamlit page/domain gateway modules가 존재한다.
+- AC-2: `dashboard/app.py`가 flat broad gateway 의존을 줄이고 page/domain gateway 호출로 동작한다.
+- AC-3: `pytest tests/test_dashboard_app.py`, full pytest, MCP discovery, 필요 시 browser/HTTP smoke가 PASS 한다.
+- AC-4: baseline docs와 dcy-011 status/links가 완료 상태를 반영한다.
+
+## 6. Impacted Area
+- `jarca_ledger_mcp/adapters/mcp/server.py`
+- `jarca_ledger_mcp/adapters/mcp/tool_contracts.py`
+- `jarca_ledger_mcp/adapters/streamlit/*.py`
+- `dashboard/app.py`
+- `docs/project-structure.md`
+- `docs/runtime-flows.md`
+- `docs/discovery/dcy-011_20260703_backend-refactor-phase4-write-import-workflow-adapter-narrowing.md`
+
+## 7. Change Log
+- 2026-07-03 08:03 KST: dcy-011 Phase 7 adapter narrowing + baseline refresh 범위를 바탕으로 REQ-018을 생성하고 BAT-014 입력으로 즉시 승인했다.
+- 2026-07-03 08:23 KST: MCP tool_contracts, Streamlit page/domain gateways, baseline 문서 refresh, AppTest/full pytest/MCP discovery evidence를 근거로 REQ-018을 Implemented로 전환했다.
diff --git a/docs/srs/Non-Functional/req-016_import-pipeline-application-infrastructure-split.md b/docs/srs/Non-Functional/req-016_import-pipeline-application-infrastructure-split.md
new file mode 100644
index 0000000..45f74b2
--- /dev/null
+++ b/docs/srs/Non-Functional/req-016_import-pipeline-application-infrastructure-split.md
@@ -0,0 +1,80 @@
+# REQ-016 Import pipeline application/infrastructure split and evidence-preserving parser relocation
+
+- Status: Implemented
+- Type: Non-Functional
+- Priority: High
+- Owner: 렛저장인
+- Source Discovery: `docs/discovery/dcy-011_20260703_backend-refactor-phase4-write-import-workflow-adapter-narrowing.md`
+- Batch: BAT-012
+- Created At (KST): 2026-07-03 07:49 KST
+- Approved At (KST): 2026-07-03 07:49 KST
+- Implemented At (KST): 2026-07-03 07:53 KST
+
+## 1. 배경
+
+`REQ-015`/BAT-011로 transactional write path는 좁은 write-oriented application module로 분리됐지만, import preview/import/rollback과 account auto-linking/classification/orchestration은 여전히 `application/import_transfer_services.py`에 크게 남아 있다. 또한 parser entrypoint는 top-level `kb_bank_pdf.py`, `toss_bank_pdf.py`, `toss_securities_pdf.py`, `seoulpay_xlsx.py`에 흩어져 있어 application layer가 parser 구현 위치를 직접 아는 구조다.
+
+## 2. Intent
+
+이 REQ의 목적은 import preview/import/rollback을 전용 `imports_service`로 분리하고, parser access path를 `infrastructure/parsers/*`로 정리해 parser(파싱/정규화)와 application import orchestration(dedupe/account auto-link/import_batch persistence/classification/post-import bookkeeping)을 구분하는 것이다. 기존 MCP tool 이름, response envelope, import evidence fields는 유지해야 한다.
+
+## 3. 범위
+
+### In Scope
+- `preview_ledger_csv_import`
+- `preview/import_kb_bank_pdf`
+- `preview/import_seoulpay_xlsx`
+- `preview/import_toss_bank_pdf`
+- `preview/import_toss_securities_pdf`
+- `list_ledger_import_batches`
+- `rollback_ledger_import_batch`
+- import용 account auto-linking/classification/helper를 `imports_service` 경계로 이동
+- parser import 경로를 `jarca_ledger_mcp/infrastructure/parsers/*`로 정리
+- 기존 `application/import_transfer_services.py`는 advanced workflow 중심으로 남기고 import-facing 함수는 compatibility delegation으로 유지
+
+### Out of Scope
+- cross-transfer/reconciliation split
+- asset/net worth/workbench split
+- MCP/Streamlit adapter surface narrowing
+- parser 알고리즘/분류 semantics 변경
+- destructive migration
+
+## 4. Requirement
+
+### FR-1. Import orchestration service extraction
+- import preview/import/rollback과 account auto-linking/classification helper는 `application/imports_service.py`로 이동해야 한다.
+- `service.py`, MCP adapter, 기존 호출 surface는 함수명/인자/응답 envelope을 유지해야 한다.
+
+### FR-2. Parser relocation by infrastructure seam
+- application layer는 parser 구현을 top-level 모듈에서 직접 끌어오지 않고 `infrastructure/parsers/*` 경로를 통해 접근해야 한다.
+- parser는 parsing/normalization/preview helper를 제공하고, persistence/import_batch 생성/transaction insert는 application service에 남아야 한다.
+
+### FR-3. Evidence-preserving contract safety
+- `raw_summary`, `counterparty`, `raw_payload`, `external_id`, `import_batch_id`, duplicate counting, validation payload 등 import evidence 필드는 의도적으로 변경되지 않아야 한다.
+- import regression tests, rollback behavior, MCP discovery가 PASS 해야 한다.
+
+## 5. Acceptance Criteria
+- AC-1: `application/imports_service.py`와 `infrastructure/parsers/*`가 생성되고 import preview/import/rollback 구현이 그쪽으로 이동한다.
+- AC-2: 기존 public API/MCP tools/import response envelope과 evidence fields가 유지된다.
+- AC-3: import-related regression과 full pytest가 PASS 한다.
+- AC-4: BAT-012 implementation/verification 문서가 실제 changed files, commands, outcomes를 기록한다.
+
+## 6. Impacted Area
+- `jarca_ledger_mcp/application/import_transfer_services.py`
+- `jarca_ledger_mcp/application/imports_service.py`
+- `jarca_ledger_mcp/application/shared.py`
+- `jarca_ledger_mcp/infrastructure/parsers/*`
+- `jarca_ledger_mcp/service.py`
+- `jarca_ledger_mcp/adapters/mcp/server.py`
+- `tests/test_kb_bank_pdf.py`
+- `tests/test_seoulpay_xlsx.py`
+- `tests/test_toss_securities_pdf.py`
+- `tests/test_service.py`
+
+## 7. Notes
+- 이 REQ는 package boundary와 maintainability 중심의 refactor이므로 `Non-Functional`로 관리한다.
+- advanced workflow split은 후속 REQ/BAT로 이어진다.
+
+## 8. Change Log
+- 2026-07-03 07:49 KST: dcy-011 Phase 5 import pipeline split 범위를 바탕으로 REQ-016을 생성하고 BAT-012 입력으로 즉시 승인했다.
+- 2026-07-03 07:53 KST: imports_service/parsers seam 분리, import-focused regression, full pytest, MCP discovery evidence를 근거로 Implemented로 전환했다.
diff --git a/docs/srs/Non-Functional/req-017_advanced-finance-workflow-service-split.md b/docs/srs/Non-Functional/req-017_advanced-finance-workflow-service-split.md
new file mode 100644
index 0000000..136c1ba
--- /dev/null
+++ b/docs/srs/Non-Functional/req-017_advanced-finance-workflow-service-split.md
@@ -0,0 +1,70 @@
+# REQ-017 Advanced finance workflow service split and compatibility-preserving domain isolation
+
+- Status: Implemented
+- Type: Non-Functional
+- Priority: High
+- Owner: 렛저장인
+- Source Discovery: `docs/discovery/dcy-011_20260703_backend-refactor-phase4-write-import-workflow-adapter-narrowing.md`
+- Batch: BAT-013
+- Created At (KST): 2026-07-03 07:55 KST
+- Approved At (KST): 2026-07-03 07:55 KST
+- Implemented At (KST): 2026-07-03 08:01 KST
+
+## 1. 배경
+
+`REQ-015`와 `REQ-016`으로 transactional write path와 import pipeline seam은 좁아졌지만, cross-transfer matching/review, balance reconciliation, manual asset/snapshot/net-worth workflows, uncategorized workbench는 여전히 broad application modules(`import_transfer_services.py`, `transaction_report_services.py`, `asset_account_services.py`, `categorization_services.py`)에 흩어져 있다.
+
+## 2. Intent
+
+이 REQ의 목적은 advanced finance workflows를 domain-oriented application modules로 추가 분리해 import/write/read service와 책임이 겹치지 않게 하는 것이다. public API, MCP contracts, dashboard behavior는 유지해야 하며, domain-heavy helpers는 owning workflow service로 이동해야 한다.
+
+## 3. 범위
+
+### In Scope
+- cross-transfer matching/review queue를 `application/cross_transfer_service.py`로 이동
+- balance reconciliation을 `application/reconciliation_service.py`로 이동
+- manual asset CRUD / snapshot / liquidity / net-worth workflows를 `application/assets_service.py`로 이동
+- uncategorized list/summary/workbench/recommendation apply 흐름을 `application/uncategorized_workbench_service.py`로 이동
+- 기존 broad modules는 compatibility delegation을 유지
+
+### Out of Scope
+- MCP/Streamlit adapter surface narrowing
+- baseline docs final update
+- tool contract rename or semantic changes
+- destructive migration
+
+## 4. Requirement
+
+### FR-1. Advanced workflow ownership split
+- cross-transfer, reconciliation, assets, uncategorized workbench는 각각 owning application module로 이동해야 한다.
+- helper 함수도 첫 소비자와 함께 이동해 broad module 내부 숨은 결합을 줄여야 한다.
+
+### FR-2. Compatibility-preserving delegation
+- `service.py`, MCP tool names, dashboard call signatures, response envelope은 유지해야 한다.
+- 기존 broad modules는 필요 시 thin delegation/re-export layer로 남을 수 있다.
+
+### FR-3. Regression-safe domain isolation
+- cross-transfer tests, asset snapshot/manual asset tests, reconciliation tests, uncategorized workflow tests, full pytest가 PASS 해야 한다.
+
+## 5. Acceptance Criteria
+- AC-1: `cross_transfer_service.py`, `reconciliation_service.py`, `assets_service.py`, `uncategorized_workbench_service.py`가 생성되고 owning workflow 구현이 이동한다.
+- AC-2: public API/MCP/dashboard contract가 유지된다.
+- AC-3: workflow-focused regression과 full pytest가 PASS 한다.
+- AC-4: BAT-013 implementation/verification 문서가 실제 변경 파일과 검증 결과를 반영한다.
+
+## 6. Impacted Area
+- `jarca_ledger_mcp/application/import_transfer_services.py`
+- `jarca_ledger_mcp/application/transaction_report_services.py`
+- `jarca_ledger_mcp/application/asset_account_services.py`
+- `jarca_ledger_mcp/application/categorization_services.py`
+- `jarca_ledger_mcp/application/{cross_transfer,reconciliation,assets,uncategorized_workbench}_service.py`
+- `jarca_ledger_mcp/service.py`
+- `tests/test_cross_account_transfer_matching.py`
+- `tests/test_asset_snapshots.py`
+- `tests/test_contract_regression.py`
+- `tests/test_categorization.py`
+- `tests/test_service.py`
+
+## 7. Change Log
+- 2026-07-03 07:55 KST: dcy-011 Phase 6 advanced workflow split 범위를 바탕으로 REQ-017을 생성하고 BAT-013 입력으로 즉시 승인했다.
+- 2026-07-03 08:01 KST: advanced workflow service extraction, full pytest, MCP discovery evidence를 근거로 Implemented로 전환했다.
diff --git a/docs/srs/index.md b/docs/srs/index.md
index 04457c6..7f65b23 100644
--- a/docs/srs/index.md
+++ b/docs/srs/index.md
@@ -33,7 +33,7 @@
 
 ## Next Requirement ID
 
-- Current: `REQ-016`
+- Current: `REQ-019`
 - Rule: `Requirement Register`에 있는 가장 큰 REQ 번호보다 정확히 1 커야 한다.
 
 ## Requirement Register
@@ -55,6 +55,9 @@
 | REQ-013 | Backend core layering / SQLite infrastructure split and service facade compatibility | Non-Functional | Implemented | High | 렛저장인 | [req-013_backend-core-layering-sqlite-infrastructure-service-facade-compatibility.md](Non-Functional/req-013_backend-core-layering-sqlite-infrastructure-service-facade-compatibility.md) |
 | REQ-014 | MCP/Streamlit adapter seam realignment and contract-preserving gateway | Interface | Implemented | High | 렛저장인 | [req-014_mcp-streamlit-adapter-seam-realignment-contract-preserving-gateway.md](Interface/req-014_mcp-streamlit-adapter-seam-realignment-contract-preserving-gateway.md) |
 | REQ-015 | Transactional write service split and compatibility-preserving domain extraction | Non-Functional | Implemented | High | 렛저장인 | [req-015_transactional-write-service-split-compatibility.md](Non-Functional/req-015_transactional-write-service-split-compatibility.md) |
+| REQ-016 | Import pipeline application/infrastructure split and evidence-preserving parser relocation | Non-Functional | Implemented | High | 렛저장인 | [req-016_import-pipeline-application-infrastructure-split.md](Non-Functional/req-016_import-pipeline-application-infrastructure-split.md) |
+| REQ-017 | Advanced finance workflow service split and compatibility-preserving domain isolation | Non-Functional | Implemented | High | 렛저장인 | [req-017_advanced-finance-workflow-service-split.md](Non-Functional/req-017_advanced-finance-workflow-service-split.md) |
+| REQ-018 | MCP/Streamlit adapter surface narrowing and baseline refresh for dcy-011 completion | Interface | Implemented | High | 렛저장인 | [req-018_adapter-surface-narrowing-baseline-refresh.md](Interface/req-018_adapter-surface-narrowing-baseline-refresh.md) |
 
 ## Recent Change Log Summary
 
@@ -83,3 +86,9 @@
 - 2026-07-03 00:18 KST: BAT-010 구현·검증 evidence, full pytest PASS, MCP tool discovery, browser smoke를 근거로 `REQ-013`, `REQ-014`를 Implemented로 전환했다.
 - 2026-07-03 07:30 KST: dcy-011 confirmed 범위를 `REQ-015` Proposed로 생성하고 첫 delivery 입력으로 Approved 상태로 전환했다.
 - 2026-07-03 07:39 KST: BAT-011 write-service split 구현과 full pytest/MCP discovery evidence를 근거로 `REQ-015`를 Implemented로 전환했다.
+- 2026-07-03 07:49 KST: dcy-011 Phase 5 import pipeline split 범위를 `REQ-016`으로 생성하고 BAT-012 입력으로 Approved 상태를 확정했다.
+- 2026-07-03 07:53 KST: BAT-012 import pipeline split 구현과 full pytest/MCP discovery evidence를 근거로 `REQ-016`을 Implemented로 전환했다.
+- 2026-07-03 07:55 KST: dcy-011 Phase 6 advanced workflow split 범위를 `REQ-017`로 생성하고 BAT-013 입력으로 Approved 상태를 확정했다.
+- 2026-07-03 08:01 KST: BAT-013 advanced workflow split 구현과 full pytest/MCP discovery evidence를 근거로 `REQ-017`을 Implemented로 전환했다.
+- 2026-07-03 08:03 KST: dcy-011 Phase 7 adapter narrowing + baseline refresh 범위를 `REQ-018`로 생성하고 BAT-014 입력으로 Approved 상태를 확정했다.
+- 2026-07-03 08:23 KST: BAT-014 adapter narrowing + baseline refresh 구현과 AppTest/full pytest/MCP discovery evidence를 근거로 `REQ-018`을 Implemented로 전환했다.
diff --git a/jarca_ledger_mcp/adapters/mcp/server.py b/jarca_ledger_mcp/adapters/mcp/server.py
index 9be3a3e..bd6b9b0 100644
--- a/jarca_ledger_mcp/adapters/mcp/server.py
+++ b/jarca_ledger_mcp/adapters/mcp/server.py
@@ -4,7 +4,7 @@ from pathlib import Path
 from dotenv import load_dotenv
 from mcp.server.fastmcp import FastMCP
 
-from ... import service
+from . import tool_contracts as contracts
 
 load_dotenv(Path(__file__).resolve().parents[3] / ".env")
 mcp = FastMCP("jarca-ledger-mcp")
@@ -12,282 +12,282 @@ mcp = FastMCP("jarca-ledger-mcp")
 @mcp.tool()
 def ledger_health_check() -> dict:
     """Verify the Jarca Ledger MCP server and SQLite database are reachable."""
-    return service.health_check()
+    return contracts.health_check()
 
 @mcp.tool()
 def get_ledger_db_status(include_hash: bool = True, include_integrity: bool = True) -> dict:
     """Return non-sensitive DB status evidence: path, size, mtime, checksum, table row counts, and aggregate totals."""
-    return service.get_ledger_db_status(include_hash=include_hash, include_integrity=include_integrity)
+    return contracts.get_ledger_db_status(include_hash=include_hash, include_integrity=include_integrity)
 
 @mcp.tool()
 def create_ledger_backup(backup_dir: str = "", retention_days: int = 30, retention_count: int = 20) -> dict:
     """Create a timestamped ledger DB backup. Restore is intentionally not exposed as an MCP tool."""
-    return service.create_ledger_backup(backup_dir=backup_dir or None, retention_days=retention_days, retention_count=retention_count)
+    return contracts.create_ledger_backup(backup_dir=backup_dir or None, retention_days=retention_days, retention_count=retention_count)
 
 @mcp.tool()
 def list_ledger_backups(backup_dir: str = "", include_hash: bool = False, limit: int = 50) -> dict:
     """List ledger DB backups with metadata-only evidence."""
-    return service.list_ledger_backups(backup_dir=backup_dir or None, include_hash=include_hash, limit=limit)
+    return contracts.list_ledger_backups(backup_dir=backup_dir or None, include_hash=include_hash, limit=limit)
 
 @mcp.tool()
 def get_ledger_backup_status(file_path: str, include_hash: bool = True, include_integrity: bool = True) -> dict:
     """Return metadata-only status and integrity evidence for one backup file."""
-    return service.get_ledger_backup_status(file_path, include_hash=include_hash, include_integrity=include_integrity)
+    return contracts.get_ledger_backup_status(file_path, include_hash=include_hash, include_integrity=include_integrity)
 
 @mcp.tool()
 def create_ledger_account(name: str, type: str = "bank", institution: str = "", currency: str = "KRW", active: bool = True, memo: str = "", account_number_masked: str = "", holder_name: str = "", last_balance: int | None = None, last_synced_at: str = "", statement_status: str = "unknown", liquidity_bucket: str = "other") -> dict:
     """Create an account/card/cash/investment/loan ledger account."""
-    return service.create_ledger_account(name, type, institution, currency, active, memo, account_number_masked, holder_name, last_balance, last_synced_at, statement_status, liquidity_bucket)
+    return contracts.create_ledger_account(name, type, institution, currency, active, memo, account_number_masked, holder_name, last_balance, last_synced_at, statement_status, liquidity_bucket)
 
 @mcp.tool()
 def list_ledger_accounts(active_only: bool = True) -> dict:
     """List ledger accounts and cards."""
-    return service.list_ledger_accounts(active_only)
+    return contracts.list_ledger_accounts(active_only)
 
 @mcp.tool()
 def update_ledger_account(account_id: str, name: str | None = None, type: str | None = None, institution: str | None = None, currency: str | None = None, active: bool | None = None, memo: str | None = None, account_number_masked: str | None = None, holder_name: str | None = None, last_balance: int | None = None, last_synced_at: str | None = None, statement_status: str | None = None, liquidity_bucket: str | None = None) -> dict:
     """Update selected account fields, including statement_status/liquidity_bucket for manual reconciliation."""
-    return service.update_ledger_account(account_id, name=name, type=type, institution=institution, currency=currency, active=active, memo=memo, account_number_masked=account_number_masked, holder_name=holder_name, last_balance=last_balance, last_synced_at=last_synced_at, statement_status=statement_status, liquidity_bucket=liquidity_bucket)
+    return contracts.update_ledger_account(account_id, name=name, type=type, institution=institution, currency=currency, active=active, memo=memo, account_number_masked=account_number_masked, holder_name=holder_name, last_balance=last_balance, last_synced_at=last_synced_at, statement_status=statement_status, liquidity_bucket=liquidity_bucket)
 
 @mcp.tool()
 def get_asset_liquidity_report() -> dict:
     """Return asset liquidity, debt, and net worth from shared account/manual valuation sources."""
-    return service.get_asset_liquidity_report()
+    return contracts.get_asset_liquidity_report()
 
 @mcp.tool()
 def list_manual_assets(active_only: bool = True) -> dict:
     """List manual non-transaction assets/liabilities used for net-worth valuation."""
-    return service.list_manual_assets(active_only=active_only)
+    return contracts.list_manual_assets(active_only=active_only)
 
 @mcp.tool()
 def create_manual_asset(name: str, valuation_amount: int, valuation_date: str, liquidity_bucket: str, source_note: str, memo: str = "", currency: str = "KRW", active: bool = True) -> dict:
     """Create a manual non-transaction asset/liability valuation row."""
-    return service.create_manual_asset(name, valuation_amount, valuation_date, liquidity_bucket, source_note=source_note, memo=memo, currency=currency, active=active)
+    return contracts.create_manual_asset(name, valuation_amount, valuation_date, liquidity_bucket, source_note=source_note, memo=memo, currency=currency, active=active)
 
 @mcp.tool()
 def update_manual_asset(asset_id: str, name: str | None = None, valuation_amount: int | None = None, valuation_date: str | None = None, liquidity_bucket: str | None = None, source_note: str | None = None, memo: str | None = None, currency: str | None = None, active: bool | None = None) -> dict:
     """Update selected fields on a manual asset/liability valuation row."""
-    return service.update_manual_asset(asset_id, name=name, valuation_amount=valuation_amount, valuation_date=valuation_date, liquidity_bucket=liquidity_bucket, source_note=source_note, memo=memo, currency=currency, active=active)
+    return contracts.update_manual_asset(asset_id, name=name, valuation_amount=valuation_amount, valuation_date=valuation_date, liquidity_bucket=liquidity_bucket, source_note=source_note, memo=memo, currency=currency, active=active)
 
 @mcp.tool()
 def deactivate_manual_asset(asset_id: str, hard_delete: bool = False) -> dict:
     """Deactivate a manual asset/liability by default; hard delete only for exceptional cleanup."""
-    return service.deactivate_manual_asset(asset_id, hard_delete=hard_delete)
+    return contracts.deactivate_manual_asset(asset_id, hard_delete=hard_delete)
 
 @mcp.tool()
 def preview_asset_snapshot_close(basis_date: str = "") -> dict:
     """Preview a month-end asset snapshot close without mutating the database."""
-    return service.preview_asset_snapshot_close(basis_date=basis_date)
+    return contracts.preview_asset_snapshot_close(basis_date=basis_date)
 
 @mcp.tool()
 def close_asset_snapshot(basis_date: str = "", source_note: str = "", memo: str = "", correction_reason: str = "") -> dict:
     """Close an asset snapshot for a basis date; same-date re-close appends correction rows."""
-    return service.close_asset_snapshot(basis_date=basis_date, source_note=source_note, memo=memo, correction_reason=correction_reason)
+    return contracts.close_asset_snapshot(basis_date=basis_date, source_note=source_note, memo=memo, correction_reason=correction_reason)
 
 @mcp.tool()
 def get_asset_snapshot_report(basis_date: str = "") -> dict:
     """Return basis-date asset snapshot/net-worth report with live-authority fallback."""
-    return service.get_asset_snapshot_report(basis_date=basis_date)
+    return contracts.get_asset_snapshot_report(basis_date=basis_date)
 
 @mcp.tool()
 def get_net_worth_timeseries(start_date: str = "", end_date: str = "", months: int = 12) -> dict:
     """Return net-worth timeseries from stored asset snapshot rows."""
-    return service.get_net_worth_timeseries(start_date=start_date, end_date=end_date, months=months)
+    return contracts.get_net_worth_timeseries(start_date=start_date, end_date=end_date, months=months)
 
 @mcp.tool()
 def get_balance_reconciliation_report(account_id: str = "", baseline_balance: int | None = None, baseline_at: str = "", source_note: str = "", memo: str = "", tolerance: int = 1000) -> dict:
     """Preview balance reconciliation for manual_only/partial_import accounts without mutating the DB."""
-    return service.get_balance_reconciliation_report(account_id, baseline_balance, baseline_at, source_note, memo, tolerance)
+    return contracts.get_balance_reconciliation_report(account_id, baseline_balance, baseline_at, source_note, memo, tolerance)
 
 @mcp.tool()
 def create_ledger_transaction(transaction_date: str = "", type: str = "expense", amount: int = 0, category: str = "", subcategory: str = "", merchant: str = "", description: str = "", account_id: str = "", payment_method: str = "", person: str = "", project: str = "", tags: list[str] | str | None = None, is_fixed: bool = False, currency: str = "KRW", posted_date: str = "", source: str = "manual", external_id: str = "", memo: str = "", dedupe: bool = True) -> dict:
     """Create an income, expense, transfer, refund, or adjustment transaction. Amount is positive KRW minor unit; type determines direction."""
-    return service.create_ledger_transaction(transaction_date, type, amount, category, subcategory, merchant, description, account_id, payment_method, person, project, tags, is_fixed, currency, posted_date, source, external_id, memo, dedupe)
+    return contracts.create_ledger_transaction(transaction_date, type, amount, category, subcategory, merchant, description, account_id, payment_method, person, project, tags, is_fixed, currency, posted_date, source, external_id, memo, dedupe)
 
 @mcp.tool()
 def list_ledger_transactions(start_date: str = "", end_date: str = "", type: str = "", category: str = "", person: str = "", account_id: str = "", include_deleted: bool = False, limit: int = 100) -> dict:
     """List ledger transactions with date/category/person/account filters."""
-    return service.list_ledger_transactions(start_date, end_date, type, category, person, account_id, include_deleted, limit)
+    return contracts.list_ledger_transactions(start_date, end_date, type, category, person, account_id, include_deleted, limit)
 
 @mcp.tool()
 def update_ledger_transaction(transaction_id: str, transaction_date: str | None = None, type: str | None = None, amount: int | None = None, category: str | None = None, subcategory: str | None = None, merchant: str | None = None, description: str | None = None, account_id: str | None = None, payment_method: str | None = None, person: str | None = None, project: str | None = None, tags: list[str] | str | None = None, is_fixed: bool | None = None, memo: str | None = None, status: str | None = None) -> dict:
     """Update selected fields on a ledger transaction."""
-    return service.update_ledger_transaction(transaction_id, transaction_date=transaction_date, type=type, amount=amount, category=category, subcategory=subcategory, merchant=merchant, description=description, account_id=account_id, payment_method=payment_method, person=person, project=project, tags=tags, is_fixed=is_fixed, memo=memo, status=status)
+    return contracts.update_ledger_transaction(transaction_id, transaction_date=transaction_date, type=type, amount=amount, category=category, subcategory=subcategory, merchant=merchant, description=description, account_id=account_id, payment_method=payment_method, person=person, project=project, tags=tags, is_fixed=is_fixed, memo=memo, status=status)
 
 @mcp.tool()
 def delete_ledger_transaction(transaction_id: str, hard_delete: bool = False) -> dict:
     """Soft-delete a ledger transaction by default, or permanently delete with hard_delete=true."""
-    return service.delete_ledger_transaction(transaction_id, hard_delete)
+    return contracts.delete_ledger_transaction(transaction_id, hard_delete)
 
 @mcp.tool()
 def create_ledger_budget(month: str, category: str, amount: int, subcategory: str = "", person: str = "", currency: str = "KRW", memo: str = "") -> dict:
     """Create or update a monthly category/person budget."""
-    return service.create_ledger_budget(month, category, amount, subcategory, person, currency, memo)
+    return contracts.create_ledger_budget(month, category, amount, subcategory, person, currency, memo)
 
 @mcp.tool()
 def list_ledger_budgets(month: str = "") -> dict:
     """List ledger budgets, optionally for a month YYYY-MM."""
-    return service.list_ledger_budgets(month)
+    return contracts.list_ledger_budgets(month)
 
 @mcp.tool()
 def get_monthly_cashflow_report(month: str = "") -> dict:
     """Return income, expenses, net cashflow, and savings rate for a month."""
-    return service.get_monthly_cashflow_report(month)
+    return contracts.get_monthly_cashflow_report(month)
 
 @mcp.tool()
 def get_category_spending_report(month: str = "", start_date: str = "", end_date: str = "", person: str = "") -> dict:
     """Return category/subcategory spending for a month or date range."""
-    return service.get_category_spending_report(month, start_date, end_date, person)
+    return contracts.get_category_spending_report(month, start_date, end_date, person)
 
 @mcp.tool()
 def get_fixed_cost_report(month: str = "") -> dict:
     """Return fixed expense transactions and total fixed cost for a month."""
-    return service.get_fixed_cost_report(month)
+    return contracts.get_fixed_cost_report(month)
 
 @mcp.tool()
 def get_budget_vs_actual_report(month: str = "") -> dict:
     """Compare monthly budgets against actual spending."""
-    return service.get_budget_vs_actual_report(month)
+    return contracts.get_budget_vs_actual_report(month)
 
 @mcp.tool()
 def get_investable_cashflow_report(month: str = "", minimum_cash_buffer: int = 0, planned_extra_expense: int = 0) -> dict:
     """Estimate cashflow available for investing after buffer and planned extra expenses."""
-    return service.get_investable_cashflow_report(month, minimum_cash_buffer, planned_extra_expense)
+    return contracts.get_investable_cashflow_report(month, minimum_cash_buffer, planned_extra_expense)
 
 @mcp.tool()
 def preview_ledger_csv_import(file_path: str, source: str = "generic") -> dict:
     """Preview a UTF-8/UTF-8-BOM CSV import before saving transactions."""
-    return service.preview_ledger_csv_import(file_path, source)
+    return contracts.preview_ledger_csv_import(file_path, source)
 
 @mcp.tool()
 def preview_kb_bank_pdf_import(file_path: str, password: str, preview_limit: int = 20) -> dict:
     """Preview transactions from a password-protected KB Kookmin Bank account-history PDF without saving them."""
-    return service.preview_kb_bank_pdf_import(file_path, password, preview_limit)
+    return contracts.preview_kb_bank_pdf_import(file_path, password, preview_limit)
 
 @mcp.tool()
 def import_kb_bank_pdf(file_path: str, password: str, account_id: str = "", skip_duplicates: bool = True, skip_validation: bool = False) -> dict:
     """Import transactions from a KB Kookmin Bank account-history PDF into the ledger database."""
-    return service.import_kb_bank_pdf(file_path, password, account_id, skip_duplicates, skip_validation=skip_validation)
+    return contracts.import_kb_bank_pdf(file_path, password, account_id, skip_duplicates, skip_validation=skip_validation)
 
 @mcp.tool()
 def preview_seoulpay_xlsx_import(file_path: str, password: str = "", preview_limit: int = 20) -> dict:
     """Preview transactions from a SeoulPay usage-history XLSX before saving them."""
-    return service.preview_seoulpay_xlsx_import(file_path, password, preview_limit)
+    return contracts.preview_seoulpay_xlsx_import(file_path, password, preview_limit)
 
 @mcp.tool()
 def import_seoulpay_xlsx(file_path: str, password: str = "", account_id: str = "", skip_duplicates: bool = True) -> dict:
     """Import transactions from a SeoulPay usage-history XLSX into the ledger database."""
-    return service.import_seoulpay_xlsx(file_path, password, account_id, skip_duplicates)
+    return contracts.import_seoulpay_xlsx(file_path, password, account_id, skip_duplicates)
 
 @mcp.tool()
 def preview_toss_securities_pdf_import(file_path: str, preview_limit: int = 20) -> dict:
     """Preview transactions from a Toss Securities transaction-history PDF without saving them."""
-    return service.preview_toss_securities_pdf_import(file_path, preview_limit)
+    return contracts.preview_toss_securities_pdf_import(file_path, preview_limit)
 
 @mcp.tool()
 def import_toss_securities_pdf(file_path: str, account_id: str = "", skip_duplicates: bool = True) -> dict:
     """Import transactions from a Toss Securities transaction-history PDF into the ledger database."""
-    return service.import_toss_securities_pdf(file_path, account_id, skip_duplicates)
+    return contracts.import_toss_securities_pdf(file_path, account_id, skip_duplicates)
 
 @mcp.tool()
 def preview_toss_bank_pdf_import(file_path: str, preview_limit: int = 20) -> dict:
     """Preview transactions from a Toss Bank transaction-history PDF without saving them."""
-    return service.preview_toss_bank_pdf_import(file_path, preview_limit)
+    return contracts.preview_toss_bank_pdf_import(file_path, preview_limit)
 
 @mcp.tool()
 def import_toss_bank_pdf(file_path: str, account_id: str = "", skip_duplicates: bool = True) -> dict:
     """Import transactions from a Toss Bank transaction-history PDF into the ledger database."""
-    return service.import_toss_bank_pdf(file_path, account_id, skip_duplicates)
+    return contracts.import_toss_bank_pdf(file_path, account_id, skip_duplicates)
 
 @mcp.tool()
 def list_ledger_import_batches(source: str = "", include_rolled_back: bool = True, limit: int = 50) -> dict:
     """List CSV/PDF/API import batches."""
-    return service.list_ledger_import_batches(source, include_rolled_back, limit)
+    return contracts.list_ledger_import_batches(source, include_rolled_back, limit)
 
 @mcp.tool()
 def rollback_ledger_import_batch(import_batch_id: str, hard_delete: bool = False) -> dict:
     """Roll back all transactions created by one import batch. Soft-deletes by default."""
-    return service.rollback_ledger_import_batch(import_batch_id, hard_delete)
+    return contracts.rollback_ledger_import_batch(import_batch_id, hard_delete)
 
 @mcp.tool()
 def create_category_rule(match_field: str, pattern: str, category: str, subcategory: str = "", transaction_type: str = "", match_type: str = "contains", priority: int = 100, person: str = "", is_fixed: bool | None = None, confidence: float = 1.0, memo: str = "", amount_min: int | None = None, amount_max: int | None = None, day_of_month: int | None = None) -> dict:
     """Create or update a ledger transaction categorization rule."""
-    return service.create_category_rule(match_field, pattern, category, subcategory, transaction_type, match_type, priority, person, is_fixed, confidence, memo, amount_min, amount_max, day_of_month)
+    return contracts.create_category_rule(match_field, pattern, category, subcategory, transaction_type, match_type, priority, person, is_fixed, confidence, memo, amount_min, amount_max, day_of_month)
 
 @mcp.tool()
 def seed_default_category_rules(overwrite: bool = False) -> dict:
     """Seed default Jarca category rules for common Korean household ledger patterns."""
-    return service.seed_default_category_rules(overwrite)
+    return contracts.seed_default_category_rules(overwrite)
 
 @mcp.tool()
 def list_category_rules(active_only: bool = True) -> dict:
     """List transaction categorization rules."""
-    return service.list_category_rules(active_only)
+    return contracts.list_category_rules(active_only)
 
 @mcp.tool()
 def update_category_rule(rule_id: str, match_field: str, pattern: str, category: str, subcategory: str = "", transaction_type: str = "", match_type: str = "contains", priority: int = 100, person: str = "", is_fixed: bool | None = None, confidence: float = 1.0, memo: str = "", amount_min: int | None = None, amount_max: int | None = None, day_of_month: int | None = None, active: bool = True) -> dict:
     """Update an existing ledger transaction categorization rule."""
-    return service.update_category_rule(rule_id, match_field, pattern, category, subcategory, transaction_type, match_type, priority, person, is_fixed, confidence, memo, amount_min, amount_max, day_of_month, active)
+    return contracts.update_category_rule(rule_id, match_field, pattern, category, subcategory, transaction_type, match_type, priority, person, is_fixed, confidence, memo, amount_min, amount_max, day_of_month, active)
 
 @mcp.tool()
 def delete_category_rule(rule_id: str) -> dict:
     """Delete a transaction categorization rule."""
-    return service.delete_category_rule(rule_id)
+    return contracts.delete_category_rule(rule_id)
 
 @mcp.tool()
 def preview_categorization(limit: int = 1000, only_uncategorized: bool = True) -> dict:
     """Preview category-rule matches before writing categories to transactions."""
-    return service.preview_categorization(limit, only_uncategorized)
+    return contracts.preview_categorization(limit, only_uncategorized)
 
 @mcp.tool()
 def apply_categorization_rules(limit: int = 1000, only_uncategorized: bool = True) -> dict:
     """Apply active category rules to matching ledger transactions."""
-    return service.apply_categorization_rules(limit, only_uncategorized)
+    return contracts.apply_categorization_rules(limit, only_uncategorized)
 
 @mcp.tool()
 def list_uncategorized_transactions(limit: int = 100, min_amount: int = 0) -> dict:
     """List active transactions that still have no category."""
-    return service.list_uncategorized_transactions(limit, min_amount)
+    return contracts.list_uncategorized_transactions(limit, min_amount)
 
 @mcp.tool()
 def get_uncategorized_merchant_summary(limit: int = 50) -> dict:
     """Summarize uncategorized transactions by merchant/raw_summary/type for user review."""
-    return service.get_uncategorized_merchant_summary(limit)
+    return contracts.get_uncategorized_merchant_summary(limit)
 
 @mcp.tool()
 def create_ledger_category(category: str, subcategory: str = "", sort_order: int = 100, memo: str = "", cashflow_kind: str = "") -> dict:
     """Create or reactivate a ledger category/subcategory used by the dashboard and transactions."""
-    return service.create_ledger_category(category, subcategory, sort_order, memo, cashflow_kind)
+    return contracts.create_ledger_category(category, subcategory, sort_order, memo, cashflow_kind)
 
 @mcp.tool()
 def seed_default_ledger_categories(overwrite: bool = False) -> dict:
     """Seed default ledger categories and subcategories."""
-    return service.seed_default_ledger_categories(overwrite)
+    return contracts.seed_default_ledger_categories(overwrite)
 
 @mcp.tool()
 def list_ledger_categories(active_only: bool = True, include_usage: bool = True) -> dict:
     """List ledger categories and subcategories, optionally with usage counts."""
-    return service.list_ledger_categories(active_only, include_usage)
+    return contracts.list_ledger_categories(active_only, include_usage)
 
 @mcp.tool()
 def list_cross_transfer_review_queue(limit: int = 300) -> dict:
     """List KB↔토스 계좌 이체 확인필요 큐 항목."""
-    return service.list_cross_transfer_review_queue(limit)
+    return contracts.list_cross_transfer_review_queue(limit)
 
 @mcp.tool()
 def preview_cross_account_transfers(limit: int = 500) -> dict:
     """Preview KB↔토스은행 이체 매칭 후보를 반환합니다."""
-    return service.preview_cross_account_transfers(limit)
+    return contracts.preview_cross_account_transfers(limit)
 
 @mcp.tool()
 def run_cross_account_transfer_match(limit: int = 500) -> dict:
     """Run KB↔토스은행 자동 이체 매칭 및 태그/분류 반영을 수행합니다."""
-    return service.run_cross_account_transfer_match(limit)
+    return contracts.run_cross_account_transfer_match(limit)
 
 @mcp.tool()
 def delete_ledger_category(category: str, subcategory: str = "", hard_delete: bool = False, force: bool = False) -> dict:
     """Deactivate or delete a ledger category/subcategory."""
-    return service.delete_ledger_category(category, subcategory, hard_delete, force)
+    return contracts.delete_ledger_category(category, subcategory, hard_delete, force)
 
 def main() -> None:
     mcp.run()
diff --git a/jarca_ledger_mcp/adapters/mcp/tool_contracts.py b/jarca_ledger_mcp/adapters/mcp/tool_contracts.py
new file mode 100644
index 0000000..cdeadc6
--- /dev/null
+++ b/jarca_ledger_mcp/adapters/mcp/tool_contracts.py
@@ -0,0 +1,60 @@
+from __future__ import annotations
+
+from ...service import (
+    apply_categorization_rules,
+    close_asset_snapshot,
+    create_category_rule,
+    create_ledger_account,
+    create_ledger_backup,
+    create_ledger_budget,
+    create_ledger_category,
+    create_ledger_transaction,
+    create_manual_asset,
+    deactivate_manual_asset,
+    delete_category_rule,
+    delete_ledger_category,
+    delete_ledger_transaction,
+    get_asset_liquidity_report,
+    get_asset_snapshot_report,
+    get_balance_reconciliation_report,
+    get_budget_vs_actual_report,
+    get_category_spending_report,
+    get_fixed_cost_report,
+    get_investable_cashflow_report,
+    get_ledger_backup_status,
+    get_ledger_db_status,
+    get_monthly_cashflow_report,
+    get_net_worth_timeseries,
+    health_check,
+    import_kb_bank_pdf,
+    import_seoulpay_xlsx,
+    import_toss_bank_pdf,
+    import_toss_securities_pdf,
+    list_category_rules,
+    list_cross_transfer_review_queue,
+    list_ledger_accounts,
+    list_ledger_backups,
+    list_ledger_budgets,
+    list_ledger_categories,
+    list_ledger_import_batches,
+    list_ledger_transactions,
+    list_manual_assets,
+    list_uncategorized_transactions,
+    preview_asset_snapshot_close,
+    preview_categorization,
+    preview_cross_account_transfers,
+    preview_kb_bank_pdf_import,
+    preview_ledger_csv_import,
+    preview_seoulpay_xlsx_import,
+    preview_toss_bank_pdf_import,
+    preview_toss_securities_pdf_import,
+    rollback_ledger_import_batch,
+    run_cross_account_transfer_match,
+    seed_default_category_rules,
+    seed_default_ledger_categories,
+    update_category_rule,
+    update_ledger_account,
+    update_ledger_transaction,
+    update_manual_asset,
+    get_uncategorized_merchant_summary,
+)
diff --git a/jarca_ledger_mcp/adapters/streamlit/__init__.py b/jarca_ledger_mcp/adapters/streamlit/__init__.py
index 444c7ed..23d5d4e 100644
--- a/jarca_ledger_mcp/adapters/streamlit/__init__.py
+++ b/jarca_ledger_mcp/adapters/streamlit/__init__.py
@@ -1 +1,11 @@
-from . import service_gateway
+from . import (
+    accounts_gateway,
+    assets_gateway,
+    categorization_gateway,
+    imports_gateway,
+    overview_gateway,
+    reconciliation_gateway,
+    service_gateway,
+    transactions_gateway,
+    transfers_gateway,
+)
diff --git a/jarca_ledger_mcp/adapters/streamlit/accounts_gateway.py b/jarca_ledger_mcp/adapters/streamlit/accounts_gateway.py
new file mode 100644
index 0000000..8c7c8c2
--- /dev/null
+++ b/jarca_ledger_mcp/adapters/streamlit/accounts_gateway.py
@@ -0,0 +1,6 @@
+from __future__ import annotations
+
+from ...service import (
+    list_ledger_accounts,
+    update_ledger_account,
+)
diff --git a/jarca_ledger_mcp/adapters/streamlit/assets_gateway.py b/jarca_ledger_mcp/adapters/streamlit/assets_gateway.py
new file mode 100644
index 0000000..808a6c2
--- /dev/null
+++ b/jarca_ledger_mcp/adapters/streamlit/assets_gateway.py
@@ -0,0 +1,10 @@
+from __future__ import annotations
+
+from ...service import (
+    close_asset_snapshot,
+    create_manual_asset,
+    deactivate_manual_asset,
+    list_manual_assets,
+    preview_asset_snapshot_close,
+    update_manual_asset,
+)
diff --git a/jarca_ledger_mcp/adapters/streamlit/categorization_gateway.py b/jarca_ledger_mcp/adapters/streamlit/categorization_gateway.py
new file mode 100644
index 0000000..39f7fcb
--- /dev/null
+++ b/jarca_ledger_mcp/adapters/streamlit/categorization_gateway.py
@@ -0,0 +1,20 @@
+from __future__ import annotations
+
+from ...service import (
+    apply_categorization_rules,
+    apply_category_rule,
+    apply_uncategorized_group_recommendation,
+    apply_uncategorized_recommendation,
+    create_category_rule,
+    create_ledger_category,
+    delete_ledger_category,
+    get_uncategorized_merchant_summary,
+    get_uncategorized_workbench,
+    list_category_rules,
+    list_ledger_categories,
+    list_uncategorized_transactions,
+    preview_categorization,
+    preview_category_rule_matches,
+    seed_default_category_rules,
+    seed_default_ledger_categories,
+)
diff --git a/jarca_ledger_mcp/adapters/streamlit/imports_gateway.py b/jarca_ledger_mcp/adapters/streamlit/imports_gateway.py
new file mode 100644
index 0000000..bcb1e82
--- /dev/null
+++ b/jarca_ledger_mcp/adapters/streamlit/imports_gateway.py
@@ -0,0 +1,10 @@
+from __future__ import annotations
+
+from ...service import (
+    import_toss_bank_pdf,
+    import_toss_securities_pdf,
+    list_ledger_import_batches,
+    preview_toss_bank_pdf_import,
+    preview_toss_securities_pdf_import,
+    rollback_ledger_import_batch,
+)
diff --git a/jarca_ledger_mcp/adapters/streamlit/overview_gateway.py b/jarca_ledger_mcp/adapters/streamlit/overview_gateway.py
new file mode 100644
index 0000000..20a900e
--- /dev/null
+++ b/jarca_ledger_mcp/adapters/streamlit/overview_gateway.py
@@ -0,0 +1,13 @@
+from __future__ import annotations
+
+from ...service import (
+    get_asset_liquidity_report,
+    get_balance_reconciliation_report,
+    get_budget_vs_actual_report,
+    get_monthly_cashflow_report,
+    health_check,
+    list_cross_transfer_review_queue,
+    list_ledger_backups,
+    list_ledger_import_batches,
+    list_uncategorized_transactions,
+)
diff --git a/jarca_ledger_mcp/adapters/streamlit/reconciliation_gateway.py b/jarca_ledger_mcp/adapters/streamlit/reconciliation_gateway.py
new file mode 100644
index 0000000..862dc83
--- /dev/null
+++ b/jarca_ledger_mcp/adapters/streamlit/reconciliation_gateway.py
@@ -0,0 +1,3 @@
+from __future__ import annotations
+
+from ...service import get_balance_reconciliation_report
diff --git a/jarca_ledger_mcp/adapters/streamlit/service_gateway.py b/jarca_ledger_mcp/adapters/streamlit/service_gateway.py
index f57fa23..d96afa6 100644
--- a/jarca_ledger_mcp/adapters/streamlit/service_gateway.py
+++ b/jarca_ledger_mcp/adapters/streamlit/service_gateway.py
@@ -1,44 +1,10 @@
 from __future__ import annotations
 
-from ...service import (
-    apply_categorization_rules,
-    apply_uncategorized_group_recommendation,
-    apply_uncategorized_recommendation,
-    close_asset_snapshot,
-    create_category_rule,
-    create_ledger_category,
-    create_manual_asset,
-    deactivate_manual_asset,
-    delete_ledger_category,
-    find_cancel_original_candidates,
-    find_self_transfer_candidates,
-    get_asset_liquidity_report,
-    get_balance_reconciliation_report,
-    get_budget_vs_actual_report,
-    get_monthly_cashflow_report,
-    health_check,
-    import_toss_bank_pdf,
-    import_toss_securities_pdf,
-    link_cancel_to_original,
-    link_self_transfer_pair,
-    list_category_rules,
-    list_cross_transfer_review_queue,
-    list_ledger_accounts,
-    list_ledger_backups,
-    list_ledger_categories,
-    list_ledger_import_batches,
-    list_manual_assets,
-    list_uncategorized_transactions,
-    preview_asset_snapshot_close,
-    preview_categorization,
-    preview_cross_account_transfers,
-    preview_toss_bank_pdf_import,
-    preview_toss_securities_pdf_import,
-    rollback_ledger_import_batch,
-    run_cross_account_transfer_match,
-    seed_default_category_rules,
-    seed_default_ledger_categories,
-    update_ledger_account,
-    update_ledger_transaction,
-    update_manual_asset,
-)
+from .accounts_gateway import *
+from .assets_gateway import *
+from .categorization_gateway import *
+from .imports_gateway import *
+from .overview_gateway import *
+from .reconciliation_gateway import *
+from .transactions_gateway import *
+from .transfers_gateway import *
diff --git a/jarca_ledger_mcp/adapters/streamlit/transactions_gateway.py b/jarca_ledger_mcp/adapters/streamlit/transactions_gateway.py
new file mode 100644
index 0000000..a0d23d8
--- /dev/null
+++ b/jarca_ledger_mcp/adapters/streamlit/transactions_gateway.py
@@ -0,0 +1,3 @@
+from __future__ import annotations
+
+from ...service import update_ledger_transaction
diff --git a/jarca_ledger_mcp/adapters/streamlit/transfers_gateway.py b/jarca_ledger_mcp/adapters/streamlit/transfers_gateway.py
new file mode 100644
index 0000000..e4a9991
--- /dev/null
+++ b/jarca_ledger_mcp/adapters/streamlit/transfers_gateway.py
@@ -0,0 +1,11 @@
+from __future__ import annotations
+
+from ...service import (
+    find_cancel_original_candidates,
+    find_self_transfer_candidates,
+    link_cancel_to_original,
+    link_self_transfer_pair,
+    list_cross_transfer_review_queue,
+    preview_cross_account_transfers,
+    run_cross_account_transfer_match,
+)
diff --git a/jarca_ledger_mcp/application/__init__.py b/jarca_ledger_mcp/application/__init__.py
index ffbeaa4..af708b1 100644
--- a/jarca_ledger_mcp/application/__init__.py
+++ b/jarca_ledger_mcp/application/__init__.py
@@ -1,10 +1,15 @@
 from . import shared
 from .accounts_write_services import *
 from .asset_account_services import *
+from .assets_service import *
 from .budgets_write_services import *
 from .category_rules_write_services import *
 from .categorization_services import *
+from .cross_transfer_service import *
+from .imports_service import *
 from .import_transfer_services import *
+from .reconciliation_service import *
 from .system_services import *
 from .transactions_write_services import *
 from .transaction_report_services import *
+from .uncategorized_workbench_service import *
diff --git a/jarca_ledger_mcp/application/asset_account_services.py b/jarca_ledger_mcp/application/asset_account_services.py
index f2c9095..9dbbb77 100644
--- a/jarca_ledger_mcp/application/asset_account_services.py
+++ b/jarca_ledger_mcp/application/asset_account_services.py
@@ -1,493 +1,15 @@
 from __future__ import annotations
 
-from .shared import *
-
-def _valuation_policy() -> dict[str, Any]:
-    return {
-        "authority_sources": ["accounts.last_balance", "manual_assets"],
-        "snapshot_role": "historical_record_layer",
-        "formula": "asset_buckets - loan",
-        "asset_buckets": ASSET_BUCKETS,
-        "liability_buckets": LIABILITY_BUCKETS,
-        "missing_valuation_policy": "exclude_without_transaction_inference",
-    }
-
-
-def _account_valuation_items(conn: sqlite3.Connection) -> tuple[list[dict[str, Any]], list[dict[str, Any]], list[dict[str, Any]]]:
-    rows = conn.execute("SELECT * FROM accounts WHERE active=1 ORDER BY name").fetchall()
-    included: list[dict[str, Any]] = []
-    excluded: list[dict[str, Any]] = []
-    invalid: list[dict[str, Any]] = []
-    for row in rows:
-        account = row_to_dict(row)
-        raw_bucket = str(account.get("liquidity_bucket") or "other")
-        try:
-            bucket = normalize_liquidity_bucket(raw_bucket)
-        except ValueError:
-            bucket = "other"
-            invalid.append({"id": account.get("id", ""), "name": account.get("name", ""), "liquidity_bucket": raw_bucket})
-        base = {
-            "id": account.get("id", ""),
-            "source_id": account.get("id", ""),
-            "source_type": "account",
-            "authority_source": "accounts.last_balance",
-            "name": account.get("name", ""),
-            "type": account.get("type", ""),
-            "liquidity_bucket": bucket,
-            "valuation_date": account.get("last_synced_at", ""),
-            "last_synced_at": account.get("last_synced_at", ""),
-            "source_note": account.get("memo", ""),
-            "memo": account.get("memo", ""),
-        }
-        balance = account.get("last_balance")
-        if balance is None:
-            excluded.append({**base, "reason": "missing_last_balance"})
-            continue
-        amount = int(balance)
-        included.append({**base, "amount": amount, "last_balance": amount})
-    return included, excluded, invalid
-
-
-def _manual_asset_valuation_items(conn: sqlite3.Connection, active_only: bool = True) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]:
-    where = "WHERE active=1" if active_only else ""
-    rows = conn.execute(f"SELECT * FROM manual_assets {where} ORDER BY name").fetchall()
-    included: list[dict[str, Any]] = []
-    excluded: list[dict[str, Any]] = []
-    for row in rows:
-        asset = row_to_dict(row)
-        try:
-            bucket = normalize_liquidity_bucket(asset.get("liquidity_bucket"))
-        except ValueError:
-            bucket = "other"
-        base = {
-            "id": asset.get("id", ""),
-            "source_id": asset.get("id", ""),
-            "source_type": "manual_asset",
-            "authority_source": "manual_assets",
-            "name": asset.get("name", ""),
-            "liquidity_bucket": bucket,
-            "valuation_date": asset.get("valuation_date", ""),
-            "source_note": asset.get("source_note", ""),
-            "memo": asset.get("memo", ""),
-            "active": asset.get("active", True),
-        }
-        value = asset.get("valuation_amount")
-        if value is None:
-            excluded.append({**base, "reason": "missing_valuation_amount"})
-            continue
-        included.append({**base, "amount": int(value), "valuation_amount": int(value)})
-    return included, excluded
-
-
-def _snapshot_rows_for_latest_run(conn: sqlite3.Connection, basis_date: str) -> tuple[list[sqlite3.Row], dict[str, Any]]:
-    rows = conn.execute(
-        "SELECT rowid AS _rowid, * FROM asset_snapshots WHERE basis_date=? ORDER BY created_at ASC, name ASC",
-        (basis_date,),
-    ).fetchall()
-    if not rows:
-        return [], {}
-    run_rows = [row for row in rows if "snapshot_run_id" in row.keys() and row["snapshot_run_id"]]
-    if not run_rows:
-        return rows, {"snapshot_run_id": "", "is_correction": False, "correction_of_run_id": "", "correction_reason": "", "total_snapshot_rows_for_basis_date": len(rows)}
-    latest = sorted(run_rows, key=lambda row: (int(row["_rowid"]), row["closed_at"] or row["created_at"] or "", row["snapshot_run_id"] or ""))[-1]
-    latest_run_id = latest["snapshot_run_id"]
-    selected = [row for row in run_rows if row["snapshot_run_id"] == latest_run_id]
-    metadata = {
-        "snapshot_run_id": latest_run_id,
-        "is_correction": bool(latest["is_correction"]),
-        "correction_of_run_id": latest["correction_of_run_id"] or "",
-        "correction_reason": latest["correction_reason"] or "",
-        "closed_at": latest["closed_at"] or latest["created_at"] or "",
-        "total_snapshot_rows_for_basis_date": len(rows),
-    }
-    return selected, metadata
-
-
-def _snapshot_items(conn: sqlite3.Connection, basis_date: str) -> list[dict[str, Any]]:
-    rows, _metadata = _snapshot_rows_for_latest_run(conn, basis_date)
-    items: list[dict[str, Any]] = []
-    for row in rows:
-        snap = row_to_dict(row)
-        try:
-            bucket = normalize_liquidity_bucket(snap.get("liquidity_bucket"))
-        except ValueError:
-            bucket = "other"
-        items.append(
-            {
-                "id": snap.get("id", ""),
-                "source_id": snap.get("source_id", ""),
-                "source_type": snap.get("source_type", ""),
-                "authority_source": "asset_snapshots",
-                "valuation_source": snap.get("valuation_source", ""),
-                "name": snap.get("name", ""),
-                "liquidity_bucket": bucket,
-                "amount": int(snap.get("amount") or 0),
-                "valuation_date": snap.get("basis_date", ""),
-                "source_note": snap.get("source_note", ""),
-                "memo": snap.get("memo", ""),
-                "snapshot_run_id": snap.get("snapshot_run_id", ""),
-                "is_correction": bool(snap.get("is_correction", False)),
-                "correction_of_run_id": snap.get("correction_of_run_id", ""),
-                "correction_reason": snap.get("correction_reason", ""),
-            }
-        )
-    return items
-
-
-def _summarize_valuation_items(included_items: list[dict[str, Any]], excluded_items: list[dict[str, Any]] | None = None) -> dict[str, Any]:
-    bucket_totals = {bucket: 0 for bucket in ASSET_BUCKET_ORDER}
-    bucket_counts = {bucket: 0 for bucket in ASSET_BUCKET_ORDER}
-    for item in included_items:
-        bucket = normalize_liquidity_bucket(item.get("liquidity_bucket"))
-        amount = int(item.get("amount") or 0)
-        bucket_counts[bucket] += 1
-        bucket_totals[bucket] += amount
-    total_assets = sum(bucket_totals[bucket] for bucket in ASSET_BUCKETS)
-    total_liabilities = sum(abs(bucket_totals[bucket]) for bucket in LIABILITY_BUCKETS)
-    net_worth = total_assets - total_liabilities
-    buckets = []
-    for bucket in ASSET_BUCKET_ORDER:
-        amount = bucket_totals[bucket]
-        denominator = total_liabilities if bucket in LIABILITY_BUCKETS else total_assets
-        buckets.append(
-            {
-                "liquidity_bucket": bucket,
-                "amount": amount,
-                "account_count": sum(1 for item in included_items if item.get("source_type") == "account" and item.get("liquidity_bucket") == bucket),
-                "manual_asset_count": sum(1 for item in included_items if item.get("source_type") == "manual_asset" and item.get("liquidity_bucket") == bucket),
-                "item_count": bucket_counts[bucket],
-                "asset_share": round(amount / total_assets, 4) if bucket not in LIABILITY_BUCKETS and total_assets else None,
-                "liability_share": round(abs(amount) / total_liabilities, 4) if bucket in LIABILITY_BUCKETS and total_liabilities else None,
-                "share": round(abs(amount) / denominator, 4) if denominator else None,
-            }
-        )
-    return {
-        "buckets": buckets,
-        "bucket_totals": bucket_totals,
-        "total_assets": total_assets,
-        "total_liabilities": total_liabilities,
-        "net_worth": net_worth,
-        "included_items": included_items,
-        "excluded_items": excluded_items or [],
-        "valuation_policy": _valuation_policy(),
-    }
-
-
-def create_manual_asset(name: str, valuation_amount: int | None, valuation_date: str, liquidity_bucket: str, source_note: str = "", memo: str = "", currency: str = "KRW", active: bool = True, db_path: str | None = None) -> dict[str, Any]:
-    if not name.strip():
-        return fail("validation_error", "manual asset name is required")
-    try:
-        bucket = normalize_liquidity_bucket(liquidity_bucket)
-        amount = int(valuation_amount) if valuation_amount is not None else None
-        vdate = normalize_date(valuation_date, default_today=True)
-    except Exception as e:
-        return fail("validation_error", str(e))
-    ts = now_iso()
-    asset_id = new_id("asset")
-    with db(db_path) as conn:
-        conn.execute(
-            """INSERT INTO manual_assets(id,name,liquidity_bucket,valuation_amount,valuation_date,currency,source_note,memo,active,created_at,updated_at)
-               VALUES(?,?,?,?,?,?,?,?,?,?,?)""",
-            (asset_id, name.strip(), bucket, amount, vdate, currency, source_note, memo, int(active), ts, ts),
-        )
-        row = conn.execute("SELECT * FROM manual_assets WHERE id=?", (asset_id,)).fetchone()
-    return ok(manual_asset={**row_to_dict(row), "authority_source": "manual_assets"})
-
-
-def list_manual_assets(active_only: bool = True, db_path: str | None = None) -> dict[str, Any]:
-    where = "WHERE active=1" if active_only else ""
-    with db(db_path) as conn:
-        rows = conn.execute(f"SELECT * FROM manual_assets {where} ORDER BY name").fetchall()
-    assets = [{**row_to_dict(row), "authority_source": "manual_assets"} for row in rows]
-    return ok(manual_assets=assets, count=len(assets), valuation_policy=_valuation_policy())
-
-
-def update_manual_asset(asset_id: str, name: str | None = None, valuation_amount: int | None = None, valuation_date: str | None = None, liquidity_bucket: str | None = None, source_note: str | None = None, memo: str | None = None, currency: str | None = None, active: bool | None = None, db_path: str | None = None) -> dict[str, Any]:
-    allowed_updates: list[str] = []
-    params: list[Any] = []
-    try:
-        if name is not None:
-            if not name.strip():
-                return fail("validation_error", "manual asset name is required")
-            allowed_updates.append("name=?")
-            params.append(name.strip())
-        if valuation_amount is not None:
-            allowed_updates.append("valuation_amount=?")
-            params.append(int(valuation_amount))
-        if valuation_date is not None:
-            allowed_updates.append("valuation_date=?")
-            params.append(normalize_date(valuation_date, default_today=True))
-        if liquidity_bucket is not None:
-            allowed_updates.append("liquidity_bucket=?")
-            params.append(normalize_liquidity_bucket(liquidity_bucket))
-        if source_note is not None:
-            allowed_updates.append("source_note=?")
-            params.append(source_note)
-        if memo is not None:
-            allowed_updates.append("memo=?")
-            params.append(memo)
-        if currency is not None:
-            allowed_updates.append("currency=?")
-            params.append(currency or "KRW")
-        if active is not None:
-            allowed_updates.append("active=?")
-            params.append(int(bool(active)))
-    except Exception as e:
-        return fail("validation_error", str(e))
-    if not allowed_updates:
-        return fail("validation_error", "no fields to update")
-    allowed_updates.append("updated_at=?")
-    params.extend([now_iso(), asset_id])
-    with db(db_path) as conn:
-        cur = conn.execute(f"UPDATE manual_assets SET {', '.join(allowed_updates)} WHERE id=?", params)
-        if cur.rowcount == 0:
-            return fail("not_found", "manual asset not found")
-        row = conn.execute("SELECT * FROM manual_assets WHERE id=?", (asset_id,)).fetchone()
-    return ok(manual_asset={**row_to_dict(row), "authority_source": "manual_assets"})
-
-
-def deactivate_manual_asset(asset_id: str, hard_delete: bool = False, db_path: str | None = None) -> dict[str, Any]:
-    with db(db_path) as conn:
-        if hard_delete:
-            cur = conn.execute("DELETE FROM manual_assets WHERE id=?", (asset_id,))
-        else:
-            cur = conn.execute("UPDATE manual_assets SET active=0, updated_at=? WHERE id=?", (now_iso(), asset_id))
-        if cur.rowcount == 0:
-            return fail("not_found", "manual asset not found")
-        row = None if hard_delete else conn.execute("SELECT * FROM manual_assets WHERE id=?", (asset_id,)).fetchone()
-    payload = {"manual_asset_id": asset_id, "deactivated": True, "hard_delete": hard_delete}
-    if row is not None:
-        payload["manual_asset"] = {**row_to_dict(row), "authority_source": "manual_assets"}
-    return ok(**payload)
-
-
-def _snapshot_preview_from_conn(conn: sqlite3.Connection, basis: str) -> dict[str, Any]:
-    account_items, account_excluded, invalid_bucket_accounts = _account_valuation_items(conn)
-    manual_items, manual_excluded = _manual_asset_valuation_items(conn)
-    included = account_items + manual_items
-    excluded = account_excluded + manual_excluded
-    summary = _summarize_valuation_items(included, excluded)
-    existing_rows = conn.execute("SELECT COUNT(*) FROM asset_snapshots WHERE basis_date=?", (basis,)).fetchone()[0]
-    warnings: list[dict[str, Any]] = []
-    for item in excluded:
-        warnings.append({"code": item.get("reason", "excluded_item"), "source_type": item.get("source_type", ""), "source_id": item.get("source_id", ""), "name": item.get("name", "")})
-    for item in included:
-        item_date = str(item.get("valuation_date") or "")[:10]
-        if item_date and item_date < basis:
-            warnings.append({"code": "stale_valuation", "source_type": item.get("source_type", ""), "source_id": item.get("source_id", ""), "name": item.get("name", ""), "valuation_date": item_date, "basis_date": basis})
-    for item in invalid_bucket_accounts:
-        warnings.append({"code": "invalid_liquidity_bucket", **item})
-    if existing_rows:
-        warnings.append({"code": "existing_snapshot_rows", "basis_date": basis, "snapshot_row_count": int(existing_rows), "policy": "append_only_correction_required"})
-    return {
-        "basis_date": basis,
-        "source": "live_authority",
-        "is_closed_snapshot": False,
-        "snapshot_count": 0,
-        "existing_snapshot_row_count": int(existing_rows),
-        "would_be_correction": bool(existing_rows),
-        "warnings": warnings,
-        **summary,
-    }
-
-
-def preview_asset_snapshot_close(basis_date: str = "", db_path: str | None = None) -> dict[str, Any]:
-    try:
-        basis = normalize_date(basis_date, default_today=True) if basis_date else current_kst_date().isoformat()
-    except Exception as e:
-        return fail("validation_error", str(e))
-    with db(db_path) as conn:
-        preview = _snapshot_preview_from_conn(conn, basis)
-    return ok(**preview)
-
-
-def close_asset_snapshot(basis_date: str = "", source_note: str = "", memo: str = "", correction_reason: str = "", db_path: str | None = None) -> dict[str, Any]:
-    try:
-        basis = normalize_date(basis_date, default_today=True) if basis_date else current_kst_date().isoformat()
-    except Exception as e:
-        return fail("validation_error", str(e))
-    ts = now_iso()
-    run_id = new_id("snaprun")
-    with db(db_path) as conn:
-        preview = _snapshot_preview_from_conn(conn, basis)
-        included = preview["included_items"]
-        latest_rows, latest_meta = _snapshot_rows_for_latest_run(conn, basis)
-        is_correction = bool(preview["existing_snapshot_row_count"])
-        if is_correction and not correction_reason.strip():
-            return fail("validation_error", "correction_reason is required when closing a basis_date that already has snapshot rows", basis_date=basis, existing_snapshot_row_count=preview["existing_snapshot_row_count"])
-        correction_of_run_id = latest_meta.get("snapshot_run_id", "") if latest_rows else ""
-        snapshot_ids: list[str] = []
-        for item in included:
-            snapshot_id = new_id("snap")
-            row_note = source_note or item.get("source_note", "") or "month-end close"
-            row_memo = memo or item.get("memo", "")
-            conn.execute(
-                """INSERT INTO asset_snapshots(id,basis_date,source_type,source_id,name,amount,currency,liquidity_bucket,valuation_source,source_note,memo,snapshot_run_id,correction_of_run_id,correction_reason,is_correction,closed_at,created_at)
-                   VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
-                (
-                    snapshot_id,
-                    basis,
-                    item.get("source_type", ""),
-                    item.get("source_id", item.get("id", "")),
-                    item.get("name", ""),
-                    int(item.get("amount") or 0),
-                    item.get("currency", "KRW"),
-                    normalize_liquidity_bucket(item.get("liquidity_bucket")),
-                    item.get("authority_source", item.get("source_type", "")),
-                    row_note,
-                    row_memo,
-                    run_id,
-                    correction_of_run_id,
-                    correction_reason.strip(),
-                    int(is_correction),
-                    ts,
-                    ts,
-                ),
-            )
-            snapshot_ids.append(snapshot_id)
-        report_items = _snapshot_items(conn, basis)
-        summary = _summarize_valuation_items(report_items, [])
-    return ok(
-        basis_date=basis,
-        source="snapshot",
-        is_closed_snapshot=True,
-        snapshot_run_id=run_id,
-        correction_of_run_id=correction_of_run_id,
-        correction_reason=correction_reason.strip(),
-        is_correction=is_correction,
-        closed_at=ts,
-        inserted_count=len(snapshot_ids),
-        inserted_snapshot_ids=snapshot_ids,
-        warnings=preview["warnings"],
-        **summary,
-    )
-
-
-def create_asset_snapshot(basis_date: str, source_type: str, source_id: str, name: str, amount: int | None, liquidity_bucket: str, valuation_source: str = "", source_note: str = "", memo: str = "", currency: str = "KRW", db_path: str | None = None) -> dict[str, Any]:
-    if source_type not in {"account", "manual_asset"}:
-        return fail("validation_error", "source_type must be one of ['account', 'manual_asset']")
-    if not name.strip():
-        return fail("validation_error", "snapshot item name is required")
-    try:
-        basis = normalize_date(basis_date, default_today=True)
-        bucket = normalize_liquidity_bucket(liquidity_bucket)
-        amount_int = int(amount) if amount is not None else None
-    except Exception as e:
-        return fail("validation_error", str(e))
-    ts = now_iso()
-    snapshot_id = new_id("snap")
-    with db(db_path) as conn:
-        conn.execute(
-            """INSERT INTO asset_snapshots(id,basis_date,source_type,source_id,name,amount,currency,liquidity_bucket,valuation_source,source_note,memo,created_at)
-               VALUES(?,?,?,?,?,?,?,?,?,?,?,?)""",
-            (snapshot_id, basis, source_type, source_id, name.strip(), amount_int, currency, bucket, valuation_source or source_type, source_note, memo, ts),
-        )
-        row = conn.execute("SELECT * FROM asset_snapshots WHERE id=?", (snapshot_id,)).fetchone()
-    return ok(asset_snapshot=row_to_dict(row))
-
-
-def get_asset_snapshot_report(basis_date: str = "", db_path: str | None = None) -> dict[str, Any]:
-    basis = normalize_date(basis_date, default_today=True) if basis_date else current_kst_date().isoformat()
-    with db(db_path) as conn:
-        rows, close_metadata = _snapshot_rows_for_latest_run(conn, basis)
-        snapshots = _snapshot_items(conn, basis) if rows else []
-        if snapshots:
-            summary = _summarize_valuation_items(snapshots, [])
-            return ok(
-                basis_date=basis,
-                source="asset_snapshots",
-                is_closed_snapshot=True,
-                snapshot_count=len(snapshots),
-                close_metadata=close_metadata,
-                **summary,
-            )
-        account_items, account_excluded, invalid_bucket_accounts = _account_valuation_items(conn)
-        manual_items, manual_excluded = _manual_asset_valuation_items(conn)
-    included = account_items + manual_items
-    excluded = account_excluded + manual_excluded
-    summary = _summarize_valuation_items(included, excluded)
-    return ok(
-        basis_date=basis,
-        source="live_authority",
-        is_closed_snapshot=False,
-        snapshot_count=0,
-        invalid_bucket_accounts=invalid_bucket_accounts,
-        **summary,
-    )
-
-
-def get_net_worth_timeseries(start_date: str = "", end_date: str = "", months: int = 12, db_path: str | None = None) -> dict[str, Any]:
-    try:
-        months_int = max(1, min(int(months), 120))
-        end_norm = normalize_date(end_date, default_today=True) if end_date else current_kst_date().isoformat()
-        if start_date:
-            start_norm = normalize_date(start_date)
-        else:
-            end_dt = date.fromisoformat(end_norm)
-            start_month = end_dt.month - months_int + 1
-            year = end_dt.year + (start_month - 1) // 12
-            month = (start_month - 1) % 12 + 1
-            start_norm = date(year, month, 1).isoformat()
-    except Exception as e:
-        return fail("validation_error", str(e))
-    with db(db_path) as conn:
-        basis_rows = conn.execute(
-            """SELECT DISTINCT basis_date FROM asset_snapshots
-                 WHERE basis_date>=? AND basis_date<=?
-                 ORDER BY basis_date ASC""",
-            (start_norm, end_norm),
-        ).fetchall()
-        points = []
-        for row in basis_rows:
-            basis = row["basis_date"]
-            items = _snapshot_items(conn, basis)
-            summary = _summarize_valuation_items(items, [])
-            points.append(
-                {
-                    "basis_date": basis,
-                    "total_assets": summary["total_assets"],
-                    "total_liabilities": summary["total_liabilities"],
-                    "net_worth": summary["net_worth"],
-                    "source": "asset_snapshots",
-                    "item_count": len(items),
-                }
-            )
-    return ok(start_date=start_norm, end_date=end_norm, months=months_int, points=points, count=len(points), valuation_policy=_valuation_policy())
-
-
-def get_asset_liquidity_report(db_path: str | None = None) -> dict[str, Any]:
-    """Summarize assets/liabilities from shared valuation authority sources.
-
-    Existing REQ-005 keys are preserved. When manual_assets is empty, totals and
-    source remain equivalent to the original accounts.last_balance behavior.
-    """
-    with db(db_path) as conn:
-        account_items, account_excluded, invalid_bucket_accounts = _account_valuation_items(conn)
-        manual_items, manual_excluded = _manual_asset_valuation_items(conn)
-    included = account_items + manual_items
-    excluded = account_excluded + manual_excluded
-    summary = _summarize_valuation_items(included, excluded)
-    source_accounts = [item for item in included if item.get("source_type") == "account"]
-    source = "accounts.last_balance" if not manual_items and not manual_excluded else "accounts.last_balance+manual_assets"
-    return ok(
-        buckets=summary["buckets"],
-        accounts=source_accounts,
-        manual_assets=manual_items,
-        total_assets=summary["total_assets"],
-        total_liabilities=summary["total_liabilities"],
-        net_worth=summary["net_worth"],
-        missing_balance_accounts=[
-            {"id": item.get("id", ""), "name": item.get("name", ""), "liquidity_bucket": item.get("liquidity_bucket", "")}
-            for item in account_excluded
-        ],
-        missing_valuation_items=excluded,
-        invalid_bucket_accounts=invalid_bucket_accounts,
-        manual_asset_count=len(manual_items),
-        included_items=included,
-        excluded_items=excluded,
-        valuation_policy=summary["valuation_policy"],
-        source=source,
-    )
-
-
+from .assets_service import (
+    close_asset_snapshot,
+    create_asset_snapshot,
+    create_manual_asset,
+    deactivate_manual_asset,
+    get_asset_liquidity_report,
+    get_asset_snapshot_report,
+    get_net_worth_timeseries,
+    list_manual_assets,
+    preview_asset_snapshot_close,
+    update_manual_asset,
+)
 from .accounts_write_services import update_ledger_account
diff --git a/jarca_ledger_mcp/application/assets_service.py b/jarca_ledger_mcp/application/assets_service.py
new file mode 100644
index 0000000..5c15920
--- /dev/null
+++ b/jarca_ledger_mcp/application/assets_service.py
@@ -0,0 +1,490 @@
+from __future__ import annotations
+
+from .shared import *
+
+def _valuation_policy() -> dict[str, Any]:
+    return {
+        "authority_sources": ["accounts.last_balance", "manual_assets"],
+        "snapshot_role": "historical_record_layer",
+        "formula": "asset_buckets - loan",
+        "asset_buckets": ASSET_BUCKETS,
+        "liability_buckets": LIABILITY_BUCKETS,
+        "missing_valuation_policy": "exclude_without_transaction_inference",
+    }
+
+
+def _account_valuation_items(conn: sqlite3.Connection) -> tuple[list[dict[str, Any]], list[dict[str, Any]], list[dict[str, Any]]]:
+    rows = conn.execute("SELECT * FROM accounts WHERE active=1 ORDER BY name").fetchall()
+    included: list[dict[str, Any]] = []
+    excluded: list[dict[str, Any]] = []
+    invalid: list[dict[str, Any]] = []
+    for row in rows:
+        account = row_to_dict(row)
+        raw_bucket = str(account.get("liquidity_bucket") or "other")
+        try:
+            bucket = normalize_liquidity_bucket(raw_bucket)
+        except ValueError:
+            bucket = "other"
+            invalid.append({"id": account.get("id", ""), "name": account.get("name", ""), "liquidity_bucket": raw_bucket})
+        base = {
+            "id": account.get("id", ""),
+            "source_id": account.get("id", ""),
+            "source_type": "account",
+            "authority_source": "accounts.last_balance",
+            "name": account.get("name", ""),
+            "type": account.get("type", ""),
+            "liquidity_bucket": bucket,
+            "valuation_date": account.get("last_synced_at", ""),
+            "last_synced_at": account.get("last_synced_at", ""),
+            "source_note": account.get("memo", ""),
+            "memo": account.get("memo", ""),
+        }
+        balance = account.get("last_balance")
+        if balance is None:
+            excluded.append({**base, "reason": "missing_last_balance"})
+            continue
+        amount = int(balance)
+        included.append({**base, "amount": amount, "last_balance": amount})
+    return included, excluded, invalid
+
+
+def _manual_asset_valuation_items(conn: sqlite3.Connection, active_only: bool = True) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]:
+    where = "WHERE active=1" if active_only else ""
+    rows = conn.execute(f"SELECT * FROM manual_assets {where} ORDER BY name").fetchall()
+    included: list[dict[str, Any]] = []
+    excluded: list[dict[str, Any]] = []
+    for row in rows:
+        asset = row_to_dict(row)
+        try:
+            bucket = normalize_liquidity_bucket(asset.get("liquidity_bucket"))
+        except ValueError:
+            bucket = "other"
+        base = {
+            "id": asset.get("id", ""),
+            "source_id": asset.get("id", ""),
+            "source_type": "manual_asset",
+            "authority_source": "manual_assets",
+            "name": asset.get("name", ""),
+            "liquidity_bucket": bucket,
+            "valuation_date": asset.get("valuation_date", ""),
+            "source_note": asset.get("source_note", ""),
+            "memo": asset.get("memo", ""),
+            "active": asset.get("active", True),
+        }
+        value = asset.get("valuation_amount")
+        if value is None:
+            excluded.append({**base, "reason": "missing_valuation_amount"})
+            continue
+        included.append({**base, "amount": int(value), "valuation_amount": int(value)})
+    return included, excluded
+
+
+def _snapshot_rows_for_latest_run(conn: sqlite3.Connection, basis_date: str) -> tuple[list[sqlite3.Row], dict[str, Any]]:
+    rows = conn.execute(
+        "SELECT rowid AS _rowid, * FROM asset_snapshots WHERE basis_date=? ORDER BY created_at ASC, name ASC",
+        (basis_date,),
+    ).fetchall()
+    if not rows:
+        return [], {}
+    run_rows = [row for row in rows if "snapshot_run_id" in row.keys() and row["snapshot_run_id"]]
+    if not run_rows:
+        return rows, {"snapshot_run_id": "", "is_correction": False, "correction_of_run_id": "", "correction_reason": "", "total_snapshot_rows_for_basis_date": len(rows)}
+    latest = sorted(run_rows, key=lambda row: (int(row["_rowid"]), row["closed_at"] or row["created_at"] or "", row["snapshot_run_id"] or ""))[-1]
+    latest_run_id = latest["snapshot_run_id"]
+    selected = [row for row in run_rows if row["snapshot_run_id"] == latest_run_id]
+    metadata = {
+        "snapshot_run_id": latest_run_id,
+        "is_correction": bool(latest["is_correction"]),
+        "correction_of_run_id": latest["correction_of_run_id"] or "",
+        "correction_reason": latest["correction_reason"] or "",
+        "closed_at": latest["closed_at"] or latest["created_at"] or "",
+        "total_snapshot_rows_for_basis_date": len(rows),
+    }
+    return selected, metadata
+
+
+def _snapshot_items(conn: sqlite3.Connection, basis_date: str) -> list[dict[str, Any]]:
+    rows, _metadata = _snapshot_rows_for_latest_run(conn, basis_date)
+    items: list[dict[str, Any]] = []
+    for row in rows:
+        snap = row_to_dict(row)
+        try:
+            bucket = normalize_liquidity_bucket(snap.get("liquidity_bucket"))
+        except ValueError:
+            bucket = "other"
+        items.append(
+            {
+                "id": snap.get("id", ""),
+                "source_id": snap.get("source_id", ""),
+                "source_type": snap.get("source_type", ""),
+                "authority_source": "asset_snapshots",
+                "valuation_source": snap.get("valuation_source", ""),
+                "name": snap.get("name", ""),
+                "liquidity_bucket": bucket,
+                "amount": int(snap.get("amount") or 0),
+                "valuation_date": snap.get("basis_date", ""),
+                "source_note": snap.get("source_note", ""),
+                "memo": snap.get("memo", ""),
+                "snapshot_run_id": snap.get("snapshot_run_id", ""),
+                "is_correction": bool(snap.get("is_correction", False)),
+                "correction_of_run_id": snap.get("correction_of_run_id", ""),
+                "correction_reason": snap.get("correction_reason", ""),
+            }
+        )
+    return items
+
+
+def _summarize_valuation_items(included_items: list[dict[str, Any]], excluded_items: list[dict[str, Any]] | None = None) -> dict[str, Any]:
+    bucket_totals = {bucket: 0 for bucket in ASSET_BUCKET_ORDER}
+    bucket_counts = {bucket: 0 for bucket in ASSET_BUCKET_ORDER}
+    for item in included_items:
+        bucket = normalize_liquidity_bucket(item.get("liquidity_bucket"))
+        amount = int(item.get("amount") or 0)
+        bucket_counts[bucket] += 1
+        bucket_totals[bucket] += amount
+    total_assets = sum(bucket_totals[bucket] for bucket in ASSET_BUCKETS)
+    total_liabilities = sum(abs(bucket_totals[bucket]) for bucket in LIABILITY_BUCKETS)
+    net_worth = total_assets - total_liabilities
+    buckets = []
+    for bucket in ASSET_BUCKET_ORDER:
+        amount = bucket_totals[bucket]
+        denominator = total_liabilities if bucket in LIABILITY_BUCKETS else total_assets
+        buckets.append(
+            {
+                "liquidity_bucket": bucket,
+                "amount": amount,
+                "account_count": sum(1 for item in included_items if item.get("source_type") == "account" and item.get("liquidity_bucket") == bucket),
+                "manual_asset_count": sum(1 for item in included_items if item.get("source_type") == "manual_asset" and item.get("liquidity_bucket") == bucket),
+                "item_count": bucket_counts[bucket],
+                "asset_share": round(amount / total_assets, 4) if bucket not in LIABILITY_BUCKETS and total_assets else None,
+                "liability_share": round(abs(amount) / total_liabilities, 4) if bucket in LIABILITY_BUCKETS and total_liabilities else None,
+                "share": round(abs(amount) / denominator, 4) if denominator else None,
+            }
+        )
+    return {
+        "buckets": buckets,
+        "bucket_totals": bucket_totals,
+        "total_assets": total_assets,
+        "total_liabilities": total_liabilities,
+        "net_worth": net_worth,
+        "included_items": included_items,
+        "excluded_items": excluded_items or [],
+        "valuation_policy": _valuation_policy(),
+    }
+
+
+def create_manual_asset(name: str, valuation_amount: int | None, valuation_date: str, liquidity_bucket: str, source_note: str = "", memo: str = "", currency: str = "KRW", active: bool = True, db_path: str | None = None) -> dict[str, Any]:
+    if not name.strip():
+        return fail("validation_error", "manual asset name is required")
+    try:
+        bucket = normalize_liquidity_bucket(liquidity_bucket)
+        amount = int(valuation_amount) if valuation_amount is not None else None
+        vdate = normalize_date(valuation_date, default_today=True)
+    except Exception as e:
+        return fail("validation_error", str(e))
+    ts = now_iso()
+    asset_id = new_id("asset")
+    with db(db_path) as conn:
+        conn.execute(
+            """INSERT INTO manual_assets(id,name,liquidity_bucket,valuation_amount,valuation_date,currency,source_note,memo,active,created_at,updated_at)
+               VALUES(?,?,?,?,?,?,?,?,?,?,?)""",
+            (asset_id, name.strip(), bucket, amount, vdate, currency, source_note, memo, int(active), ts, ts),
+        )
+        row = conn.execute("SELECT * FROM manual_assets WHERE id=?", (asset_id,)).fetchone()
+    return ok(manual_asset={**row_to_dict(row), "authority_source": "manual_assets"})
+
+
+def list_manual_assets(active_only: bool = True, db_path: str | None = None) -> dict[str, Any]:
+    where = "WHERE active=1" if active_only else ""
+    with db(db_path) as conn:
+        rows = conn.execute(f"SELECT * FROM manual_assets {where} ORDER BY name").fetchall()
+    assets = [{**row_to_dict(row), "authority_source": "manual_assets"} for row in rows]
+    return ok(manual_assets=assets, count=len(assets), valuation_policy=_valuation_policy())
+
+
+def update_manual_asset(asset_id: str, name: str | None = None, valuation_amount: int | None = None, valuation_date: str | None = None, liquidity_bucket: str | None = None, source_note: str | None = None, memo: str | None = None, currency: str | None = None, active: bool | None = None, db_path: str | None = None) -> dict[str, Any]:
+    allowed_updates: list[str] = []
+    params: list[Any] = []
+    try:
+        if name is not None:
+            if not name.strip():
+                return fail("validation_error", "manual asset name is required")
+            allowed_updates.append("name=?")
+            params.append(name.strip())
+        if valuation_amount is not None:
+            allowed_updates.append("valuation_amount=?")
+            params.append(int(valuation_amount))
+        if valuation_date is not None:
+            allowed_updates.append("valuation_date=?")
+            params.append(normalize_date(valuation_date, default_today=True))
+        if liquidity_bucket is not None:
+            allowed_updates.append("liquidity_bucket=?")
+            params.append(normalize_liquidity_bucket(liquidity_bucket))
+        if source_note is not None:
+            allowed_updates.append("source_note=?")
+            params.append(source_note)
+        if memo is not None:
+            allowed_updates.append("memo=?")
+            params.append(memo)
+        if currency is not None:
+            allowed_updates.append("currency=?")
+            params.append(currency or "KRW")
+        if active is not None:
+            allowed_updates.append("active=?")
+            params.append(int(bool(active)))
+    except Exception as e:
+        return fail("validation_error", str(e))
+    if not allowed_updates:
+        return fail("validation_error", "no fields to update")
+    allowed_updates.append("updated_at=?")
+    params.extend([now_iso(), asset_id])
+    with db(db_path) as conn:
+        cur = conn.execute(f"UPDATE manual_assets SET {', '.join(allowed_updates)} WHERE id=?", params)
+        if cur.rowcount == 0:
+            return fail("not_found", "manual asset not found")
+        row = conn.execute("SELECT * FROM manual_assets WHERE id=?", (asset_id,)).fetchone()
+    return ok(manual_asset={**row_to_dict(row), "authority_source": "manual_assets"})
+
+
+def deactivate_manual_asset(asset_id: str, hard_delete: bool = False, db_path: str | None = None) -> dict[str, Any]:
+    with db(db_path) as conn:
+        if hard_delete:
+            cur = conn.execute("DELETE FROM manual_assets WHERE id=?", (asset_id,))
+        else:
+            cur = conn.execute("UPDATE manual_assets SET active=0, updated_at=? WHERE id=?", (now_iso(), asset_id))
+        if cur.rowcount == 0:
+            return fail("not_found", "manual asset not found")
+        row = None if hard_delete else conn.execute("SELECT * FROM manual_assets WHERE id=?", (asset_id,)).fetchone()
+    payload = {"manual_asset_id": asset_id, "deactivated": True, "hard_delete": hard_delete}
+    if row is not None:
+        payload["manual_asset"] = {**row_to_dict(row), "authority_source": "manual_assets"}
+    return ok(**payload)
+
+
+def _snapshot_preview_from_conn(conn: sqlite3.Connection, basis: str) -> dict[str, Any]:
+    account_items, account_excluded, invalid_bucket_accounts = _account_valuation_items(conn)
+    manual_items, manual_excluded = _manual_asset_valuation_items(conn)
+    included = account_items + manual_items
+    excluded = account_excluded + manual_excluded
+    summary = _summarize_valuation_items(included, excluded)
+    existing_rows = conn.execute("SELECT COUNT(*) FROM asset_snapshots WHERE basis_date=?", (basis,)).fetchone()[0]
+    warnings: list[dict[str, Any]] = []
+    for item in excluded:
+        warnings.append({"code": item.get("reason", "excluded_item"), "source_type": item.get("source_type", ""), "source_id": item.get("source_id", ""), "name": item.get("name", "")})
+    for item in included:
+        item_date = str(item.get("valuation_date") or "")[:10]
+        if item_date and item_date < basis:
+            warnings.append({"code": "stale_valuation", "source_type": item.get("source_type", ""), "source_id": item.get("source_id", ""), "name": item.get("name", ""), "valuation_date": item_date, "basis_date": basis})
+    for item in invalid_bucket_accounts:
+        warnings.append({"code": "invalid_liquidity_bucket", **item})
+    if existing_rows:
+        warnings.append({"code": "existing_snapshot_rows", "basis_date": basis, "snapshot_row_count": int(existing_rows), "policy": "append_only_correction_required"})
+    return {
+        "basis_date": basis,
+        "source": "live_authority",
+        "is_closed_snapshot": False,
+        "snapshot_count": 0,
+        "existing_snapshot_row_count": int(existing_rows),
+        "would_be_correction": bool(existing_rows),
+        "warnings": warnings,
+        **summary,
+    }
+
+
+def preview_asset_snapshot_close(basis_date: str = "", db_path: str | None = None) -> dict[str, Any]:
+    try:
+        basis = normalize_date(basis_date, default_today=True) if basis_date else current_kst_date().isoformat()
+    except Exception as e:
+        return fail("validation_error", str(e))
+    with db(db_path) as conn:
+        preview = _snapshot_preview_from_conn(conn, basis)
+    return ok(**preview)
+
+
+def close_asset_snapshot(basis_date: str = "", source_note: str = "", memo: str = "", correction_reason: str = "", db_path: str | None = None) -> dict[str, Any]:
+    try:
+        basis = normalize_date(basis_date, default_today=True) if basis_date else current_kst_date().isoformat()
+    except Exception as e:
+        return fail("validation_error", str(e))
+    ts = now_iso()
+    run_id = new_id("snaprun")
+    with db(db_path) as conn:
+        preview = _snapshot_preview_from_conn(conn, basis)
+        included = preview["included_items"]
+        latest_rows, latest_meta = _snapshot_rows_for_latest_run(conn, basis)
+        is_correction = bool(preview["existing_snapshot_row_count"])
+        if is_correction and not correction_reason.strip():
+            return fail("validation_error", "correction_reason is required when closing a basis_date that already has snapshot rows", basis_date=basis, existing_snapshot_row_count=preview["existing_snapshot_row_count"])
+        correction_of_run_id = latest_meta.get("snapshot_run_id", "") if latest_rows else ""
+        snapshot_ids: list[str] = []
+        for item in included:
+            snapshot_id = new_id("snap")
+            row_note = source_note or item.get("source_note", "") or "month-end close"
+            row_memo = memo or item.get("memo", "")
+            conn.execute(
+                """INSERT INTO asset_snapshots(id,basis_date,source_type,source_id,name,amount,currency,liquidity_bucket,valuation_source,source_note,memo,snapshot_run_id,correction_of_run_id,correction_reason,is_correction,closed_at,created_at)
+                   VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
+                (
+                    snapshot_id,
+                    basis,
+                    item.get("source_type", ""),
+                    item.get("source_id", item.get("id", "")),
+                    item.get("name", ""),
+                    int(item.get("amount") or 0),
+                    item.get("currency", "KRW"),
+                    normalize_liquidity_bucket(item.get("liquidity_bucket")),
+                    item.get("authority_source", item.get("source_type", "")),
+                    row_note,
+                    row_memo,
+                    run_id,
+                    correction_of_run_id,
+                    correction_reason.strip(),
+                    int(is_correction),
+                    ts,
+                    ts,
+                ),
+            )
+            snapshot_ids.append(snapshot_id)
+        report_items = _snapshot_items(conn, basis)
+        summary = _summarize_valuation_items(report_items, [])
+    return ok(
+        basis_date=basis,
+        source="snapshot",
+        is_closed_snapshot=True,
+        snapshot_run_id=run_id,
+        correction_of_run_id=correction_of_run_id,
+        correction_reason=correction_reason.strip(),
+        is_correction=is_correction,
+        closed_at=ts,
+        inserted_count=len(snapshot_ids),
+        inserted_snapshot_ids=snapshot_ids,
+        warnings=preview["warnings"],
+        **summary,
+    )
+
+
+def create_asset_snapshot(basis_date: str, source_type: str, source_id: str, name: str, amount: int | None, liquidity_bucket: str, valuation_source: str = "", source_note: str = "", memo: str = "", currency: str = "KRW", db_path: str | None = None) -> dict[str, Any]:
+    if source_type not in {"account", "manual_asset"}:
+        return fail("validation_error", "source_type must be one of ['account', 'manual_asset']")
+    if not name.strip():
+        return fail("validation_error", "snapshot item name is required")
+    try:
+        basis = normalize_date(basis_date, default_today=True)
+        bucket = normalize_liquidity_bucket(liquidity_bucket)
+        amount_int = int(amount) if amount is not None else None
+    except Exception as e:
+        return fail("validation_error", str(e))
+    ts = now_iso()
+    snapshot_id = new_id("snap")
+    with db(db_path) as conn:
+        conn.execute(
+            """INSERT INTO asset_snapshots(id,basis_date,source_type,source_id,name,amount,currency,liquidity_bucket,valuation_source,source_note,memo,created_at)
+               VALUES(?,?,?,?,?,?,?,?,?,?,?,?)""",
+            (snapshot_id, basis, source_type, source_id, name.strip(), amount_int, currency, bucket, valuation_source or source_type, source_note, memo, ts),
+        )
+        row = conn.execute("SELECT * FROM asset_snapshots WHERE id=?", (snapshot_id,)).fetchone()
+    return ok(asset_snapshot=row_to_dict(row))
+
+
+def get_asset_snapshot_report(basis_date: str = "", db_path: str | None = None) -> dict[str, Any]:
+    basis = normalize_date(basis_date, default_today=True) if basis_date else current_kst_date().isoformat()
+    with db(db_path) as conn:
+        rows, close_metadata = _snapshot_rows_for_latest_run(conn, basis)
+        snapshots = _snapshot_items(conn, basis) if rows else []
+        if snapshots:
+            summary = _summarize_valuation_items(snapshots, [])
+            return ok(
+                basis_date=basis,
+                source="asset_snapshots",
+                is_closed_snapshot=True,
+                snapshot_count=len(snapshots),
+                close_metadata=close_metadata,
+                **summary,
+            )
+        account_items, account_excluded, invalid_bucket_accounts = _account_valuation_items(conn)
+        manual_items, manual_excluded = _manual_asset_valuation_items(conn)
+    included = account_items + manual_items
+    excluded = account_excluded + manual_excluded
+    summary = _summarize_valuation_items(included, excluded)
+    return ok(
+        basis_date=basis,
+        source="live_authority",
+        is_closed_snapshot=False,
+        snapshot_count=0,
+        invalid_bucket_accounts=invalid_bucket_accounts,
+        **summary,
+    )
+
+
+def get_net_worth_timeseries(start_date: str = "", end_date: str = "", months: int = 12, db_path: str | None = None) -> dict[str, Any]:
+    try:
+        months_int = max(1, min(int(months), 120))
+        end_norm = normalize_date(end_date, default_today=True) if end_date else current_kst_date().isoformat()
+        if start_date:
+            start_norm = normalize_date(start_date)
+        else:
+            end_dt = date.fromisoformat(end_norm)
+            start_month = end_dt.month - months_int + 1
+            year = end_dt.year + (start_month - 1) // 12
+            month = (start_month - 1) % 12 + 1
+            start_norm = date(year, month, 1).isoformat()
+    except Exception as e:
+        return fail("validation_error", str(e))
+    with db(db_path) as conn:
+        basis_rows = conn.execute(
+            """SELECT DISTINCT basis_date FROM asset_snapshots
+                 WHERE basis_date>=? AND basis_date<=?
+                 ORDER BY basis_date ASC""",
+            (start_norm, end_norm),
+        ).fetchall()
+        points = []
+        for row in basis_rows:
+            basis = row["basis_date"]
+            items = _snapshot_items(conn, basis)
+            summary = _summarize_valuation_items(items, [])
+            points.append(
+                {
+                    "basis_date": basis,
+                    "total_assets": summary["total_assets"],
+                    "total_liabilities": summary["total_liabilities"],
+                    "net_worth": summary["net_worth"],
+                    "source": "asset_snapshots",
+                    "item_count": len(items),
+                }
+            )
+    return ok(start_date=start_norm, end_date=end_norm, months=months_int, points=points, count=len(points), valuation_policy=_valuation_policy())
+
+
+def get_asset_liquidity_report(db_path: str | None = None) -> dict[str, Any]:
+    """Summarize assets/liabilities from shared valuation authority sources.
+
+    Existing REQ-005 keys are preserved. When manual_assets is empty, totals and
+    source remain equivalent to the original accounts.last_balance behavior.
+    """
+    with db(db_path) as conn:
+        account_items, account_excluded, invalid_bucket_accounts = _account_valuation_items(conn)
+        manual_items, manual_excluded = _manual_asset_valuation_items(conn)
+    included = account_items + manual_items
+    excluded = account_excluded + manual_excluded
+    summary = _summarize_valuation_items(included, excluded)
+    source_accounts = [item for item in included if item.get("source_type") == "account"]
+    source = "accounts.last_balance" if not manual_items and not manual_excluded else "accounts.last_balance+manual_assets"
+    return ok(
+        buckets=summary["buckets"],
+        accounts=source_accounts,
+        manual_assets=manual_items,
+        total_assets=summary["total_assets"],
+        total_liabilities=summary["total_liabilities"],
+        net_worth=summary["net_worth"],
+        missing_balance_accounts=[
+            {"id": item.get("id", ""), "name": item.get("name", ""), "liquidity_bucket": item.get("liquidity_bucket", "")}
+            for item in account_excluded
+        ],
+        missing_valuation_items=excluded,
+        invalid_bucket_accounts=invalid_bucket_accounts,
+        manual_asset_count=len(manual_items),
+        included_items=included,
+        excluded_items=excluded,
+        valuation_policy=summary["valuation_policy"],
+        source=source,
+    )
diff --git a/jarca_ledger_mcp/application/categorization_services.py b/jarca_ledger_mcp/application/categorization_services.py
index 698a411..8452aeb 100644
--- a/jarca_ledger_mcp/application/categorization_services.py
+++ b/jarca_ledger_mcp/application/categorization_services.py
@@ -274,220 +274,13 @@ def apply_categorization_rules(limit: int = 1000, only_uncategorized: bool = Tru
             applied += 1
     return ok(applied_count=applied, preview=preview)
 
-
-def list_uncategorized_transactions(limit: int = 100, min_amount: int = 0, db_path: str | None = None) -> dict[str, Any]:
-    with db(db_path) as conn:
-        rows=conn.execute("""SELECT * FROM transactions WHERE status='active' AND COALESCE(category,'')='' AND amount>=? ORDER BY amount DESC, transaction_datetime DESC LIMIT ?""", (int(min_amount), max(1,min(int(limit),1000)))).fetchall()
-    return ok(transactions=[row_to_dict(r) for r in rows], count=len(rows))
-
-
-def get_uncategorized_merchant_summary(limit: int = 50, db_path: str | None = None) -> dict[str, Any]:
-    with db(db_path) as conn:
-        rows=conn.execute("""SELECT merchant, raw_summary, type, COUNT(*) count, SUM(amount) total_amount, MAX(amount) max_amount
-                             FROM transactions WHERE status='active' AND COALESCE(category,'')=''
-                             GROUP BY merchant, raw_summary, type ORDER BY total_amount DESC LIMIT ?""", (max(1,min(int(limit),500)),)).fetchall()
-    return ok(items=[row_to_dict(r) for r in rows], count=len(rows))
-
-
-def _recommend_category_for_transaction(tx: dict[str, Any], rules: list[dict[str, Any]], history: dict[tuple[str, str], list[dict[str, Any]]]) -> dict[str, Any]:
-    """Recommend a category for one transaction using deterministic, explainable signals."""
-    for rule in rules:
-        try:
-            if _rule_matches(rule, tx):
-                return {
-                    "category": rule.get("category", ""),
-                    "subcategory": rule.get("subcategory", ""),
-                    "confidence": round(float(rule.get("confidence") or 1.0), 2),
-                    "source": "rule",
-                    "reason": f"활성 규칙 `{rule.get('pattern', '')}` 매칭",
-                    "rule_id": rule.get("id", ""),
-                }
-        except re.error:
-            continue
-
-    key = (tx.get("merchant", "") or "", tx.get("type", "") or "")
-    candidates = history.get(key, [])
-    if candidates:
-        best = candidates[0]
-        count = int(best.get("count") or 0)
-        confidence = min(0.9, 0.55 + min(count, 5) * 0.07)
-        return {
-            "category": best.get("category", ""),
-            "subcategory": best.get("subcategory", ""),
-            "confidence": round(confidence, 2),
-            "source": "history",
-            "reason": f"같은 사용처의 기존 분류 {count}건",
-            "rule_id": "",
-        }
-
-    return {"category": "", "subcategory": "", "confidence": 0.0, "source": "none", "reason": "추천 근거 없음", "rule_id": ""}
-
-
-def get_uncategorized_workbench(
-    limit: int = 100,
-    min_amount: int = 0,
-    sort_by: str = "amount_desc",
-    db_path: str | None = None,
-) -> dict[str, Any]:
-    row_limit = max(1, min(int(limit), 1000))
-    minimum_amount = int(min_amount)
-    order_by_map = {
-        "amount_desc": "amount DESC, transaction_datetime DESC, transaction_date DESC, created_at DESC",
-        "date_desc": "transaction_datetime DESC, transaction_date DESC, created_at DESC, amount DESC",
-    }
-    order_by = order_by_map.get(sort_by, order_by_map["amount_desc"])
-    with db(db_path) as conn:
-        tx_rows = [row_to_dict(r) for r in conn.execute(
-            f"""SELECT id, transaction_date, transaction_datetime, type, amount, currency, merchant, raw_summary,
-                      counterparty, account_id, payment_method, person, is_fixed, source, memo, created_at
-                 FROM transactions
-                WHERE status='active' AND COALESCE(category,'')='' AND amount>=?
-                ORDER BY {order_by}
-                LIMIT ?""",
-            (minimum_amount, row_limit),
-        ).fetchall()]
-        total_row = conn.execute(
-            """SELECT COUNT(*) AS count, COALESCE(SUM(amount),0) AS total_amount,
-                      COUNT(DISTINCT COALESCE(NULLIF(merchant,''), raw_summary, counterparty, '')) AS merchant_group_count
-                 FROM transactions
-                WHERE status='active' AND COALESCE(category,'')='' AND amount>=?""",
-            (minimum_amount,),
-        ).fetchone()
-        history_rows = [row_to_dict(r) for r in conn.execute(
-            """SELECT merchant, type, category, subcategory, COUNT(*) AS count,
-                      SUM(amount) AS total_amount, MAX(transaction_date) AS latest_date
-                 FROM transactions
-                WHERE status='active' AND COALESCE(category,'')<>'' AND COALESCE(merchant,'')<>''
-                GROUP BY merchant, type, category, subcategory
-                ORDER BY merchant, type, count DESC, total_amount DESC"""
-        ).fetchall()]
-        rules = [row_to_dict(r) for r in conn.execute(
-            "SELECT * FROM category_rules WHERE active=1 ORDER BY priority, id"
-        ).fetchall()]
-
-    history: dict[tuple[str, str], list[dict[str, Any]]] = {}
-    for row in history_rows:
-        history.setdefault((row.get("merchant", "") or "", row.get("type", "") or ""), []).append(row)
-
-    txs: list[dict[str, Any]] = []
-    groups: dict[tuple[str, str, str], dict[str, Any]] = {}
-    for tx in tx_rows:
-        rec = _recommend_category_for_transaction(tx, rules, history)
-        public_tx = {
-            **tx,
-            "recommendation": rec,
-            "recommended_category": rec.get("category", ""),
-            "recommended_subcategory": rec.get("subcategory", ""),
-            "recommendation_confidence": rec.get("confidence", 0.0),
-            "recommendation_reason": rec.get("reason", ""),
-            "recommendation_source": rec.get("source", "none"),
-        }
-        txs.append(public_tx)
-        gkey = (tx.get("merchant", "") or "", tx.get("raw_summary", "") or "", tx.get("type", "") or "")
-        group = groups.setdefault(gkey, {
-            "merchant": gkey[0],
-            "raw_summary": gkey[1],
-            "type": gkey[2],
-            "count": 0,
-            "total_amount": 0,
-            "max_amount": 0,
-            "latest_date": "",
-            "sample_transaction_id": tx.get("id", ""),
-            "recommendation": rec,
-        })
-        group["count"] += 1
-        group["total_amount"] += int(tx.get("amount") or 0)
-        group["max_amount"] = max(int(group["max_amount"] or 0), int(tx.get("amount") or 0))
-        date_value = tx.get("transaction_datetime") or tx.get("transaction_date") or ""
-        if date_value > (group.get("latest_date") or ""):
-            group["latest_date"] = date_value
-        if not group["recommendation"].get("category") and rec.get("category"):
-            group["recommendation"] = rec
-
-    group_items = sorted(groups.values(), key=lambda g: (int(g["count"]), int(g["total_amount"])), reverse=True)
-    recommendable_count = sum(1 for tx in txs if tx.get("recommended_category"))
-    repeated_group_count = sum(1 for g in group_items if int(g.get("count") or 0) >= 2)
-    total = row_to_dict(total_row) if total_row else {"count": 0, "total_amount": 0, "merchant_group_count": 0}
-    return ok(
-        summary={
-            "uncategorized_count": int(total.get("count") or 0),
-            "uncategorized_total_amount": int(total.get("total_amount") or 0),
-            "merchant_group_count": int(total.get("merchant_group_count") or 0),
-            "recommendable_count": recommendable_count,
-            "repeated_group_count": repeated_group_count,
-        },
-        groups=group_items,
-        transactions=txs,
-        count=len(txs),
-    )
-
-
-def apply_uncategorized_recommendation(transaction_id: str, db_path: str | None = None) -> dict[str, Any]:
-    if not transaction_id:
-        return fail("validation_error", "transaction_id is required")
-    with db(db_path) as conn:
-        row = conn.execute("SELECT * FROM transactions WHERE id=? AND status='active'", (transaction_id,)).fetchone()
-        if not row:
-            return fail("not_found", "transaction not found")
-        tx = row_to_dict(row)
-        if tx.get("category"):
-            return fail("validation_error", "transaction is already categorized")
-        rules = [row_to_dict(r) for r in conn.execute("SELECT * FROM category_rules WHERE active=1 ORDER BY priority, id").fetchall()]
-        history_rows = [row_to_dict(r) for r in conn.execute(
-            """SELECT merchant, type, category, subcategory, COUNT(*) AS count, SUM(amount) AS total_amount
-                 FROM transactions
-                WHERE status='active' AND COALESCE(category,'')<>'' AND COALESCE(merchant,'')<>''
-                GROUP BY merchant, type, category, subcategory
-                ORDER BY count DESC, total_amount DESC"""
-        ).fetchall()]
-        history: dict[tuple[str, str], list[dict[str, Any]]] = {}
-        for h in history_rows:
-            history.setdefault((h.get("merchant", "") or "", h.get("type", "") or ""), []).append(h)
-        rec = _recommend_category_for_transaction(tx, rules, history)
-        if not rec.get("category"):
-            return fail("no_recommendation", "추천 분류가 없습니다")
-        cur = conn.execute(
-            """UPDATE transactions
-                  SET category=?, subcategory=?, updated_at=?
-                WHERE id=? AND status='active' AND COALESCE(category,'')=''""",
-            (rec.get("category", ""), rec.get("subcategory", ""), now_iso(), transaction_id),
-        )
-    return ok(applied_count=cur.rowcount, transaction_id=transaction_id, recommendation=rec)
-
-
-def apply_uncategorized_group_recommendation(merchant: str, raw_summary: str = "", transaction_type: str = "", db_path: str | None = None) -> dict[str, Any]:
-    with db(db_path) as conn:
-        sample = conn.execute(
-            """SELECT * FROM transactions
-                WHERE status='active' AND COALESCE(category,'')='' AND merchant=? AND raw_summary=? AND type=?
-                ORDER BY amount DESC, transaction_datetime DESC, transaction_date DESC LIMIT 1""",
-            (merchant, raw_summary, transaction_type),
-        ).fetchone()
-        if not sample:
-            return fail("not_found", "uncategorized group not found")
-        tx = row_to_dict(sample)
-        rules = [row_to_dict(r) for r in conn.execute("SELECT * FROM category_rules WHERE active=1 ORDER BY priority, id").fetchall()]
-        history_rows = [row_to_dict(r) for r in conn.execute(
-            """SELECT merchant, type, category, subcategory, COUNT(*) AS count, SUM(amount) AS total_amount
-                 FROM transactions
-                WHERE status='active' AND COALESCE(category,'')<>'' AND COALESCE(merchant,'')<>''
-                GROUP BY merchant, type, category, subcategory
-                ORDER BY count DESC, total_amount DESC"""
-        ).fetchall()]
-        history: dict[tuple[str, str], list[dict[str, Any]]] = {}
-        for h in history_rows:
-            history.setdefault((h.get("merchant", "") or "", h.get("type", "") or ""), []).append(h)
-        rec = _recommend_category_for_transaction(tx, rules, history)
-        if not rec.get("category"):
-            return fail("no_recommendation", "추천 분류가 없습니다")
-        cur = conn.execute(
-            """UPDATE transactions
-                  SET category=?, subcategory=?, updated_at=?
-                WHERE status='active' AND COALESCE(category,'')='' AND merchant=? AND raw_summary=? AND type=?""",
-            (rec.get("category", ""), rec.get("subcategory", ""), now_iso(), merchant, raw_summary, transaction_type),
-        )
-    return ok(applied_count=cur.rowcount, merchant=merchant, raw_summary=raw_summary, type=transaction_type, recommendation=rec)
-
+from .uncategorized_workbench_service import (
+    apply_uncategorized_group_recommendation,
+    apply_uncategorized_recommendation,
+    get_uncategorized_merchant_summary,
+    get_uncategorized_workbench,
+    list_uncategorized_transactions,
+)
 
 DEFAULT_LEDGER_CATEGORIES: list[dict[str, Any]] = [
     {"category": "수입", "subcategory": "급여", "sort_order": 10},
@@ -599,6 +392,3 @@ def list_ledger_categories(active_only: bool = True, include_usage: bool = True,
         if item["subcategory"] and item["subcategory"] not in grouped[item["category"]]:
             grouped[item["category"]].append(item["subcategory"])
     return ok(categories=items, grouped=grouped, count=len(items))
-
-
-
diff --git a/jarca_ledger_mcp/application/cross_transfer_service.py b/jarca_ledger_mcp/application/cross_transfer_service.py
new file mode 100644
index 0000000..f58846e
--- /dev/null
+++ b/jarca_ledger_mcp/application/cross_transfer_service.py
@@ -0,0 +1,593 @@
+from __future__ import annotations
+
+from .shared import *
+from .shared import _append_memo_tag, _memo_has_tag, _normalize_counterparty_for_match, _party_match_score
+
+def _build_cross_transfer_index(transactions: list[dict[str, Any]]) -> dict[tuple[str, int, str], list[dict[str, Any]]]:
+    index: dict[tuple[str, int, str], list[dict[str, Any]]] = {}
+    for tx in transactions:
+        if tx.get("status") != "active":
+            continue
+        key = (tx["transaction_date"], int(tx["amount"]), tx["type"])
+        index.setdefault(key, []).append(tx)
+    return index
+
+
+def _fetch_transfer_candidates(db_conn: Any, source: str) -> list[dict[str, Any]]:
+    rows = db_conn.execute(
+        """
+        SELECT t.id, t.transaction_datetime, t.transaction_date, t.type, t.amount, t.category, t.subcategory,
+               t.counterparty, t.memo, t.person, t.account_id, t.payment_method, t.status,
+               a.holder_name, t.merchant
+        FROM transactions t
+        LEFT JOIN accounts a ON a.id = t.account_id
+        WHERE t.source=?
+          AND t.type IN ('income','expense')
+          AND (
+            t.category='이체'
+            OR t.subcategory IN ('확인필요', '본인계좌이동', '투자이동')
+            OR t.payment_method IN ('내계좌간자동이전', '내계좌간자동이체', '오픈뱅킹', '펌뱅킹출금', '펌뱅킹입금', '입금', '출금', 'ATM출금', '적금개설')
+            OR t.merchant LIKE '%이체%'
+          )
+        """,
+        (source,),
+    ).fetchall()
+    return [dict(r) for r in rows]
+
+
+def _parse_tx_datetime(value: str | None) -> datetime | None:
+    if not value:
+        return None
+    text = value.strip()
+    for fmt in ("%Y-%m-%dT%H:%M:%S", "%Y-%m-%d %H:%M:%S"):
+        try:
+            return datetime.strptime(text, fmt)
+        except ValueError:
+            continue
+    try:
+        return datetime.fromisoformat(text)
+    except ValueError:
+        return None
+
+
+def _self_transfer_direction_score(source_type: str, candidate_type: str) -> int:
+    if source_type == "expense":
+        if candidate_type == "income":
+            return 3
+        if candidate_type == "transfer":
+            return 2
+    elif source_type == "income":
+        if candidate_type == "expense":
+            return 3
+        if candidate_type == "transfer":
+            return 2
+    elif source_type == "transfer":
+        if candidate_type in {"income", "expense"}:
+            return 2
+        if candidate_type == "transfer":
+            return 1
+    return 0
+
+
+def _normalize_cancel_lookup_text(value: str) -> str:
+    if not value:
+        return ""
+    text = value.lower().strip()
+    replacements = [
+        "kb체크취소", "승인취소", "취소", "환불", "환급", "refund", "reversal",
+        "카드입금", "카드환급", "매출취소", "입금", "출금", "체크카드", "신용카드",
+    ]
+    for token in replacements:
+        text = text.replace(token, " ")
+    text = re.sub(r"\s+", " ", text)
+    return _normalize_counterparty_for_match(text)
+
+
+def _cancel_channel(value: str) -> str:
+    text = (value or "").lower()
+    if any(token in text for token in ("카드", "check", "credit")):
+        return "card"
+    if any(token in text for token in ("이체", "오픈뱅킹", "출금", "입금", "bank")):
+        return "bank"
+    return "other"
+
+
+def _is_cancel_like_transaction(tx: dict[str, Any]) -> bool:
+    if str(tx.get("type") or "") == "refund":
+        return True
+    combined = " ".join(
+        str(tx.get(key) or "") for key in ("merchant", "raw_summary", "payment_method", "memo", "description")
+    ).lower()
+    keywords = ("취소", "환불", "환급", "refund", "reversal", "카드입금", "매출취소")
+    return str(tx.get("type") or "") == "income" and any(token in combined for token in keywords)
+
+
+def find_cancel_original_candidates(transaction_id: str, limit: int = 20, lookback_days: int = 365, db_path: str | None = None) -> dict[str, Any]:
+    limit = max(1, min(int(limit), 50))
+    lookback_days = max(1, min(int(lookback_days), 730))
+    with db(db_path) as conn:
+        source_row = conn.execute(
+            """
+            SELECT t.*, a.name AS account_name, a.institution, a.holder_name
+            FROM transactions t
+            LEFT JOIN accounts a ON a.id = t.account_id
+            WHERE t.id=?
+            """,
+            (transaction_id,),
+        ).fetchone()
+        if not source_row:
+            return fail("not_found", f"transaction not found: {transaction_id}")
+        source = row_to_dict(source_row)
+        explicit_cancel_like = _is_cancel_like_transaction(source)
+        source_type = str(source.get("type") or "")
+        if source_type not in {"income", "refund"} and not explicit_cancel_like:
+            return fail("not_cancel_like", "cancel/original lookup is intended for 취소·환급성 거래")
+
+        source_date = normalize_date(source.get("transaction_date"), default_today=False)
+        source_dt = _parse_tx_datetime(source.get("transaction_datetime"))
+        source_day = date.fromisoformat(source_date)
+        start_date = date.fromordinal(source_day.toordinal() - lookback_days).isoformat()
+        source_text = " ".join(str(source.get(key) or "") for key in ("merchant", "counterparty", "raw_summary", "payment_method"))
+        source_norm_text = _normalize_cancel_lookup_text(source_text)
+        source_channel = _cancel_channel(str(source.get("payment_method") or "") + " " + str(source.get("raw_summary") or "") + " " + str(source.get("merchant") or ""))
+        candidate_rows = conn.execute(
+            """
+            SELECT t.*, a.name AS account_name, a.institution, a.holder_name
+            FROM transactions t
+            LEFT JOIN accounts a ON a.id = t.account_id
+            WHERE t.status='active'
+              AND t.id<>?
+              AND t.amount=?
+              AND t.transaction_date BETWEEN ? AND ?
+              AND t.type IN ('expense', 'adjustment')
+            ORDER BY t.transaction_date DESC, t.transaction_datetime DESC, t.created_at DESC
+            """,
+            (transaction_id, int(source.get("amount") or 0), start_date, source_date),
+        ).fetchall()
+
+    candidates: list[dict[str, Any]] = []
+    for row in candidate_rows:
+        candidate = row_to_dict(row)
+        candidate_date = normalize_date(candidate.get("transaction_date"), default_today=False)
+        candidate_day = date.fromisoformat(candidate_date)
+        day_delta = (source_day - candidate_day).days
+        if day_delta < 0:
+            continue
+        candidate_dt = _parse_tx_datetime(candidate.get("transaction_datetime"))
+        time_delta_seconds = abs(int((source_dt - candidate_dt).total_seconds())) if source_dt and candidate_dt else None
+        same_account = bool(source.get("account_id") and source.get("account_id") == candidate.get("account_id"))
+        source_same = bool(source.get("source") and source.get("source") == candidate.get("source"))
+        candidate_text = " ".join(str(candidate.get(key) or "") for key in ("merchant", "counterparty", "raw_summary", "payment_method"))
+        candidate_norm_text = _normalize_cancel_lookup_text(candidate_text)
+        merchant_score = 0
+        if source_norm_text and candidate_norm_text:
+            if source_norm_text == candidate_norm_text:
+                merchant_score = 4
+            elif source_norm_text in candidate_norm_text or candidate_norm_text in source_norm_text:
+                merchant_score = 2
+        channel_score = 0
+        candidate_channel = _cancel_channel(str(candidate.get("payment_method") or "") + " " + str(candidate.get("raw_summary") or "") + " " + str(candidate.get("merchant") or ""))
+        if source_channel != "other" and source_channel == candidate_channel:
+            channel_score = 2
+        if not explicit_cancel_like and not (same_account and merchant_score >= 2 and channel_score >= 1):
+            continue
+        recency_score = 0
+        if day_delta == 0:
+            recency_score = 5
+        elif day_delta <= 7:
+            recency_score = 4
+        elif day_delta <= 30:
+            recency_score = 3
+        elif day_delta <= 90:
+            recency_score = 2
+        else:
+            recency_score = 1
+        time_score = 0
+        if time_delta_seconds is not None:
+            if time_delta_seconds <= 3600:
+                time_score = 2
+            elif time_delta_seconds <= 86400:
+                time_score = 1
+        score = merchant_score + channel_score + recency_score + time_score + (3 if same_account else 0) + (1 if source_same else 0)
+        reason_bits = [
+            f"same_account={int(same_account)}",
+            f"merchant={merchant_score}",
+            f"channel={channel_score}",
+            f"days={day_delta}",
+        ]
+        if time_delta_seconds is not None:
+            reason_bits.append(f"time_delta={time_delta_seconds}s")
+        candidates.append({
+            "id": candidate.get("id"),
+            "transaction_date": candidate.get("transaction_date"),
+            "transaction_datetime": candidate.get("transaction_datetime"),
+            "source": candidate.get("source"),
+            "type": candidate.get("type"),
+            "amount": candidate.get("amount"),
+            "category": candidate.get("category"),
+            "subcategory": candidate.get("subcategory"),
+            "merchant": candidate.get("merchant"),
+            "raw_summary": candidate.get("raw_summary"),
+            "counterparty": candidate.get("counterparty"),
+            "account_id": candidate.get("account_id"),
+            "account_name": candidate.get("account_name"),
+            "institution": candidate.get("institution"),
+            "payment_method": candidate.get("payment_method"),
+            "memo": candidate.get("memo"),
+            "score": score,
+            "day_delta": day_delta,
+            "time_delta_seconds": time_delta_seconds,
+            "reason": ", ".join(reason_bits),
+        })
+
+    candidates.sort(key=lambda item: (-int(item.get("score", 0)), int(item.get("day_delta", 9999)), int(item.get("time_delta_seconds") or 99999999), str(item.get("transaction_datetime") or "")))
+    return ok(source_tx=source, count=len(candidates[:limit]), candidates=candidates[:limit], source="cancel_original_candidate_search")
+
+
+def find_self_transfer_candidates(transaction_id: str, limit: int = 20, db_path: str | None = None) -> dict[str, Any]:
+    limit = max(1, min(int(limit), 50))
+    with db(db_path) as conn:
+        source_row = conn.execute(
+            """
+            SELECT t.*, a.name AS account_name, a.institution, a.holder_name
+            FROM transactions t
+            LEFT JOIN accounts a ON a.id = t.account_id
+            WHERE t.id=?
+            """,
+            (transaction_id,),
+        ).fetchone()
+        if not source_row:
+            return fail("not_found", f"transaction not found: {transaction_id}")
+
+        source = row_to_dict(source_row)
+        source_date = normalize_date(source.get("transaction_date"), default_today=False)
+        source_dt = _parse_tx_datetime(source.get("transaction_datetime"))
+        source_holder = _normalize_counterparty_for_match(source.get("holder_name", ""))
+        anchor_date = date.fromisoformat(source_date) if source_date else current_kst_date()
+        window_start = anchor_date.toordinal() - 1
+        window_end = anchor_date.toordinal() + 1
+        candidate_rows = conn.execute(
+            """
+            SELECT t.*, a.name AS account_name, a.institution, a.holder_name
+            FROM transactions t
+            LEFT JOIN accounts a ON a.id = t.account_id
+            WHERE t.status='active'
+              AND t.id<>?
+              AND COALESCE(t.account_id, '')<>COALESCE(?, '')
+              AND t.amount=?
+              AND t.transaction_date BETWEEN ? AND ?
+            ORDER BY t.transaction_date, t.transaction_datetime, t.created_at
+            """,
+            (
+                transaction_id,
+                source.get("account_id", ""),
+                int(source.get("amount") or 0),
+                date.fromordinal(window_start).isoformat(),
+                date.fromordinal(window_end).isoformat(),
+            ),
+        ).fetchall()
+
+    candidates: list[dict[str, Any]] = []
+    for row in candidate_rows:
+        candidate = row_to_dict(row)
+        direction_score = _self_transfer_direction_score(str(source.get("type", "")), str(candidate.get("type", "")))
+        if direction_score == 0:
+            continue
+
+        candidate_dt = _parse_tx_datetime(candidate.get("transaction_datetime"))
+        day_delta = abs((date.fromisoformat(candidate.get("transaction_date", source_date)) - date.fromisoformat(source_date)).days)
+        time_delta_seconds = abs(int((candidate_dt - source_dt).total_seconds())) if candidate_dt and source_dt else None
+        same_owner = bool(source_holder and source_holder == _normalize_counterparty_for_match(candidate.get("holder_name", "")))
+        party_score = max(
+            _party_match_score(str(source.get("counterparty", "")), str(candidate.get("counterparty", ""))),
+            _party_match_score(str(source.get("counterparty", "")), str(candidate.get("holder_name", ""))),
+            _party_match_score(str(source.get("merchant", "")), str(candidate.get("counterparty", ""))),
+            _party_match_score(str(source.get("merchant", "")), str(candidate.get("holder_name", ""))),
+        )
+        source_gap_score = 1 if source.get("source") != candidate.get("source") else 0
+        day_score = 3 if day_delta == 0 else 1
+        time_score = 0
+        if time_delta_seconds is not None:
+            if time_delta_seconds <= 120:
+                time_score = 3
+            elif time_delta_seconds <= 3600:
+                time_score = 2
+            elif time_delta_seconds <= 86400:
+                time_score = 1
+        score = direction_score + party_score + source_gap_score + day_score + time_score + (3 if same_owner else 0)
+        reason_bits = [
+            f"direction={direction_score}",
+            f"party={party_score}",
+            f"same_owner={int(same_owner)}",
+            f"same_source={int(source.get('source') == candidate.get('source'))}",
+            f"day_delta={day_delta}",
+        ]
+        if time_delta_seconds is not None:
+            reason_bits.append(f"time_delta={time_delta_seconds}s")
+        candidates.append(
+            {
+                "id": candidate.get("id"),
+                "transaction_date": candidate.get("transaction_date"),
+                "transaction_datetime": candidate.get("transaction_datetime"),
+                "source": candidate.get("source"),
+                "type": candidate.get("type"),
+                "amount": candidate.get("amount"),
+                "category": candidate.get("category"),
+                "subcategory": candidate.get("subcategory"),
+                "merchant": candidate.get("merchant"),
+                "counterparty": candidate.get("counterparty"),
+                "account_id": candidate.get("account_id"),
+                "account_name": candidate.get("account_name"),
+                "institution": candidate.get("institution"),
+                "holder_name": candidate.get("holder_name"),
+                "memo": candidate.get("memo"),
+                "score": score,
+                "same_owner": same_owner,
+                "day_delta": day_delta,
+                "time_delta_seconds": time_delta_seconds,
+                "reason": ", ".join(reason_bits),
+            }
+        )
+
+    candidates.sort(key=lambda item: (-int(item.get("score", 0)), int(item.get("day_delta", 99)), int(item.get("time_delta_seconds") or 999999), str(item.get("transaction_datetime") or "")))
+    return ok(source_tx=source, count=len(candidates[:limit]), candidates=candidates[:limit], source="self_transfer_candidate_search")
+
+
+def link_self_transfer_pair(source_transaction_id: str, candidate_transaction_id: str, db_path: str | None = None) -> dict[str, Any]:
+    if not source_transaction_id or not candidate_transaction_id:
+        return fail("validation_error", "both transaction ids are required")
+    if source_transaction_id == candidate_transaction_id:
+        return fail("validation_error", "source and candidate transaction must differ")
+
+    preview = find_self_transfer_candidates(source_transaction_id, limit=50, db_path=db_path)
+    if not preview.get("success"):
+        return preview
+    candidates = preview.get("candidates") or []
+    candidate = next((item for item in candidates if item.get("id") == candidate_transaction_id), None)
+    if candidate is None:
+        return fail("candidate_not_found", f"candidate not found for source transaction: {candidate_transaction_id}")
+
+    link_tag = f"manual_self_transfer_link:{source_transaction_id}|{candidate_transaction_id}"
+    ts = now_iso()
+    with db(db_path) as conn:
+        source_row = conn.execute("SELECT * FROM transactions WHERE id=?", (source_transaction_id,)).fetchone()
+        candidate_row = conn.execute("SELECT * FROM transactions WHERE id=?", (candidate_transaction_id,)).fetchone()
+        if not source_row or not candidate_row:
+            return fail("not_found", "source or candidate transaction not found")
+        source_memo = _append_memo_tag(source_row["memo"], link_tag)
+        candidate_memo = _append_memo_tag(candidate_row["memo"], link_tag)
+        conn.execute(
+            "UPDATE transactions SET category='이체', subcategory='본인계좌이동', memo=?, updated_at=? WHERE id=?",
+            (source_memo, ts, source_transaction_id),
+        )
+        conn.execute(
+            "UPDATE transactions SET category='이체', subcategory='본인계좌이동', memo=?, updated_at=? WHERE id=?",
+            (candidate_memo, ts, candidate_transaction_id),
+        )
+        updated_source = conn.execute("SELECT * FROM transactions WHERE id=?", (source_transaction_id,)).fetchone()
+        updated_candidate = conn.execute("SELECT * FROM transactions WHERE id=?", (candidate_transaction_id,)).fetchone()
+
+    return ok(
+        source_transaction=row_to_dict(updated_source),
+        candidate_transaction=row_to_dict(updated_candidate),
+        link_tag=link_tag,
+        score=candidate.get("score"),
+        reason=candidate.get("reason"),
+        source="manual_self_transfer_link",
+    )
+
+
+def link_cancel_to_original(cancel_transaction_id: str, original_transaction_id: str, db_path: str | None = None) -> dict[str, Any]:
+    if not cancel_transaction_id or not original_transaction_id:
+        return fail("validation_error", "both transaction ids are required")
+    if cancel_transaction_id == original_transaction_id:
+        return fail("validation_error", "cancel and original transaction must differ")
+
+    preview = find_cancel_original_candidates(cancel_transaction_id, limit=50, db_path=db_path)
+    if not preview.get("success"):
+        return preview
+    candidates = preview.get("candidates") or []
+    original = next((item for item in candidates if item.get("id") == original_transaction_id), None)
+    if original is None:
+        return fail("candidate_not_found", f"original transaction candidate not found: {original_transaction_id}")
+
+    link_tag = f"manual_cancel_link:cancel={cancel_transaction_id}|original={original_transaction_id}"
+    ts = now_iso()
+    with db(db_path) as conn:
+        cancel_row = conn.execute("SELECT * FROM transactions WHERE id=?", (cancel_transaction_id,)).fetchone()
+        original_row = conn.execute("SELECT * FROM transactions WHERE id=?", (original_transaction_id,)).fetchone()
+        if not cancel_row or not original_row:
+            return fail("not_found", "cancel or original transaction not found")
+
+        cancel_memo = _append_memo_tag(cancel_row["memo"], link_tag)
+        original_memo = _append_memo_tag(original_row["memo"], link_tag)
+        original_category = original_row["category"] or "수입"
+        original_subcategory = original_row["subcategory"] or "환급"
+        original_person = original_row["person"] or ""
+        conn.execute(
+            "UPDATE transactions SET type='refund', category=?, subcategory=?, person=?, memo=?, updated_at=? WHERE id=?",
+            (original_category, original_subcategory, original_person, cancel_memo, ts, cancel_transaction_id),
+        )
+        conn.execute(
+            "UPDATE transactions SET memo=?, updated_at=? WHERE id=?",
+            (original_memo, ts, original_transaction_id),
+        )
+        updated_cancel = conn.execute("SELECT * FROM transactions WHERE id=?", (cancel_transaction_id,)).fetchone()
+        updated_original = conn.execute("SELECT * FROM transactions WHERE id=?", (original_transaction_id,)).fetchone()
+
+    return ok(
+        cancel_transaction=row_to_dict(updated_cancel),
+        original_transaction=row_to_dict(updated_original),
+        link_tag=link_tag,
+        score=original.get("score"),
+        reason=original.get("reason"),
+        source="manual_cancel_link",
+    )
+
+
+def _tag_memos(conn: Any, tx_id: str, tag: str) -> None:
+    existing = conn.execute("SELECT memo FROM transactions WHERE id=?", (tx_id,)).fetchone()
+    if not existing:
+        return
+    memo = _append_memo_tag(existing["memo"], tag)
+    conn.execute("UPDATE transactions SET memo=?, updated_at=? WHERE id=?", (memo, now_iso(), tx_id))
+
+
+def _run_cross_account_transfer_match(limit: int = 500, dry_run: bool = True, db_path: str | None = None) -> dict[str, Any]:
+    limit = max(1, min(int(limit), 1000))
+    matched_pairs: list[dict[str, Any]] = []
+    review_items: list[dict[str, Any]] = []
+    matched_tx_ids: set[str] = set()
+
+    with db(db_path) as conn:
+        kb_rows = _fetch_transfer_candidates(conn, "kb_bank_pdf")
+        toss_rows = _fetch_transfer_candidates(conn, "toss_bank_pdf")
+        toss_index = _build_cross_transfer_index(toss_rows)
+
+        for kb in kb_rows:
+            if kb["status"] != "active":
+                continue
+            if _memo_has_tag(kb.get("memo", ""), CROSS_TRANSFER_MATCH_TAG):
+                continue
+            required_type = "income" if kb["type"] == "expense" else "expense"
+            candidates = [
+                t
+                for t in toss_index.get((kb["transaction_date"], int(kb["amount"]), required_type), [])
+                if t.get("status") == "active" and not _memo_has_tag(t.get("memo", ""), CROSS_TRANSFER_MATCH_TAG)
+            ]
+
+            if not candidates:
+                review_items.append(
+                    {
+                        "source_tx_id": kb["id"],
+                        "source": "kb_bank_pdf",
+                        "target_tx_id": "",
+                        "reason": "no_candidate_found",
+                        "type": kb["type"],
+                        "amount": kb["amount"],
+                        "transaction_date": kb["transaction_date"],
+                        "counterparty": kb["counterparty"],
+                    }
+                )
+                continue
+
+            scored = []
+            for candidate in candidates:
+                score = _party_match_score(kb.get("counterparty", ""), candidate.get("counterparty", ""))
+                scored.append((score, candidate["transaction_datetime"], candidate))
+            scored.sort(key=lambda item: (item[0], item[1]), reverse=True)
+            best_score = scored[0][0]
+            top = [c for c in scored if c[0] == best_score]
+
+            if best_score == 0 or len(top) != 1:
+                review_items.append(
+                    {
+                        "source_tx_id": kb["id"],
+                        "source": "kb_bank_pdf",
+                        "target_tx_id": "",
+                        "reason": "ambiguous_or_name_mismatch",
+                        "candidate_count": len(candidates),
+                        "best_name_score": best_score,
+                        "type": kb["type"],
+                        "amount": kb["amount"],
+                        "transaction_date": kb["transaction_date"],
+                        "counterparty": kb["counterparty"],
+                    }
+                )
+                continue
+
+            toss = top[0][2]
+            if toss["id"] in matched_tx_ids:
+                review_items.append(
+                    {
+                        "source_tx_id": kb["id"],
+                        "source": "kb_bank_pdf",
+                        "target_tx_id": toss["id"],
+                        "reason": "target_already_matched",
+                        "type": kb["type"],
+                        "amount": kb["amount"],
+                        "transaction_date": kb["transaction_date"],
+                        "counterparty": kb["counterparty"],
+                    }
+                )
+                continue
+
+            kb_party = _normalize_counterparty_for_match(kb.get("counterparty", ""))
+            toss_party = _normalize_counterparty_for_match(toss.get("counterparty", ""))
+            owner_candidates = {
+                _normalize_counterparty_for_match(kb.get("holder_name", "")),
+                _normalize_counterparty_for_match(toss.get("holder_name", "")),
+            }
+            if (kb_party and kb_party in owner_candidates) or (toss_party and toss_party in owner_candidates):
+                subcategory = "투자이동"
+            else:
+                subcategory = "확인필요"
+
+            matched_pairs.append(
+                {
+                    "kb_tx_id": kb["id"],
+                    "toss_tx_id": toss["id"],
+                    "transaction_date": kb["transaction_date"],
+                    "amount": kb["amount"],
+                    "counterparty_kb": kb["counterparty"],
+                    "counterparty_toss": toss["counterparty"],
+                    "subcategory": subcategory,
+                    "name_score": best_score,
+                    "reason": "paired_by_date_amount_party",
+                }
+            )
+            if not dry_run:
+                review_tag = f"{CROSS_TRANSFER_MATCH_TAG}kb={kb['id']}|toss={toss['id']}"
+                conn.execute(
+                    "UPDATE transactions SET category='이체', subcategory=?, memo=?, updated_at=? WHERE id=?",
+                    (subcategory, _append_memo_tag(kb.get("memo", ""), review_tag), now_iso(), kb["id"]),
+                )
+                conn.execute(
+                    "UPDATE transactions SET category='이체', subcategory=?, memo=?, updated_at=? WHERE id=?",
+                    (subcategory, _append_memo_tag(toss.get("memo", ""), review_tag), now_iso(), toss["id"]),
+                )
+
+            if not dry_run:
+                _tag_memos(conn, kb["id"], f"{CROSS_TRANSFER_REVIEW_TAG}paired")
+                _tag_memos(conn, toss["id"], f"{CROSS_TRANSFER_REVIEW_TAG}paired")
+            matched_tx_ids.add(kb["id"])
+            matched_tx_ids.add(toss["id"])
+
+            if len(matched_pairs) >= limit:
+                break
+
+        for review in review_items:
+            if review["source_tx_id"] in matched_tx_ids:
+                continue
+            if not dry_run:
+                _tag_memos(conn, review["source_tx_id"], f"{CROSS_TRANSFER_REVIEW_TAG}{review['reason']}")
+
+
+    return ok(
+        dry_run=dry_run,
+        matched_count=len(matched_pairs),
+        review_count=len(review_items),
+        matched_pairs=matched_pairs[:limit],
+        review_items=review_items[:limit],
+        source="kb_toss_cross_transfer_match",
+    )
+
+
+def preview_cross_account_transfers(limit: int = 500, db_path: str | None = None) -> dict[str, Any]:
+    return _run_cross_account_transfer_match(limit=limit, dry_run=True, db_path=db_path)
+
+
+def run_cross_account_transfer_match(limit: int = 500, db_path: str | None = None) -> dict[str, Any]:
+    return _run_cross_account_transfer_match(limit=limit, dry_run=False, db_path=db_path)
+
+
+def list_cross_transfer_review_queue(limit: int = 300, db_path: str | None = None) -> dict[str, Any]:
+    clauses = ["t.status='active'", "t.source IN ('kb_bank_pdf','toss_bank_pdf')"]
+    clauses.append("t.memo LIKE '%cross_transfer_review:%'")
+    where = " WHERE " + " AND ".join(clauses)
+    with db(db_path) as conn:
+        rows = conn.execute(
+            f"SELECT * FROM transactions t{where} ORDER BY t.transaction_date DESC, t.created_at DESC LIMIT ?",
+            (max(1, min(int(limit), 1000)),),
+        ).fetchall()
+    return ok(transactions=[row_to_dict(r) for r in rows], count=len(rows), source="cross_transfer_review_queue")
diff --git a/jarca_ledger_mcp/application/import_transfer_services.py b/jarca_ledger_mcp/application/import_transfer_services.py
index 7db4414..d79ba54 100644
--- a/jarca_ledger_mcp/application/import_transfer_services.py
+++ b/jarca_ledger_mcp/application/import_transfer_services.py
@@ -3,1511 +3,25 @@ from __future__ import annotations
 from .shared import *
 from .shared import _append_memo_tag, _memo_has_tag, _normalize_counterparty_for_match, _party_match_score
 
-def preview_ledger_csv_import(file_path: str, source: str = "generic", db_path: str | None = None) -> dict[str, Any]:
-    path=Path(file_path).expanduser()
-    if not path.exists():
-        return fail("not_found", f"CSV file not found: {path}")
-    with path.open("r", encoding="utf-8-sig", newline="") as f:
-        rows=list(csv.DictReader(f))
-    preview=[]
-    for r in rows[:20]:
-        preview.append(r)
-    return ok(source=source, filename=str(path), row_count=len(rows), preview=preview, required_columns=["transaction_date/date", "amount", "type optional", "category optional", "merchant/description optional"])
-
-
-def preview_kb_bank_pdf_import(file_path: str, password: str, preview_limit: int = 20, db_path: str | None = None) -> dict[str, Any]:
-    try:
-        parsed = parse_kb_bank_pdf(file_path, password)
-    except FileNotFoundError as e:
-        return fail("not_found", str(e))
-    except ValueError as e:
-        return fail("pdf_password_error", str(e))
-    except Exception as e:
-        return fail("pdf_parse_error", str(e))
-    transactions = parsed["transactions"]
-    validation = parsed["validation"]
-    source_hash = parsed["source_hash"]
-    duplicate_count = 0
-    with db(db_path) as conn:
-        for tx in transactions:
-            dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, tx.counterparty, "", tx.external_id(source_hash))
-            found = conn.execute("SELECT 1 FROM transactions WHERE dedupe_key=?", (dedupe_key,)).fetchone()
-            duplicate_count += 1 if found else 0
-    preview = [transaction_preview(tx, source_hash) for tx in transactions[: max(1, min(int(preview_limit), 100))]]
-    return ok(
-        source="kb_bank_pdf",
-        file_path=parsed["file_path"],
-        source_hash=source_hash,
-        page_count=parsed["page_count"],
-        metadata=parsed["metadata"],
-        row_count=len(transactions),
-        duplicate_count=duplicate_count,
-        validation=validation,
-        preview=preview,
-    )
-
-
-def _find_or_create_kb_account(conn: Any, metadata: dict[str, Any], account_id: str = "") -> str:
-    if account_id:
-        row = conn.execute("SELECT id FROM accounts WHERE id=?", (account_id,)).fetchone()
-        if row:
-            return account_id
-        raise ValueError(f"account_id not found: {account_id}")
-    account_number_masked = metadata.get("account_number_masked") or ""
-    row = None
-    if account_number_masked:
-        row = conn.execute(
-            "SELECT id FROM accounts WHERE institution=? AND account_number_masked=? AND active=1 ORDER BY created_at LIMIT 1",
-            ("KB국민은행", account_number_masked),
-        ).fetchone()
-    if row:
-        return row["id"]
-    ts = now_iso()
-    new_account_id = new_id("acct")
-    account_name = metadata.get("account_name") or "KB국민은행 계좌"
-    display_name = f"KB국민은행 {account_name} {account_number_masked}".strip()
-    conn.execute(
-        """INSERT INTO accounts(id,name,type,institution,currency,active,memo,account_number_masked,holder_name,last_balance,last_synced_at,created_at,updated_at)
-           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?)""",
-        (
-            new_account_id,
-            display_name,
-            "bank",
-            "KB국민은행",
-            "KRW",
-            1,
-            "국민은행 거래내역 PDF import로 자동 생성",
-            account_number_masked,
-            metadata.get("holder_name") or "",
-            metadata.get("final_balance"),
-            ts,
-            ts,
-            ts,
-        ),
-    )
-    return new_account_id
-
-
-def import_kb_bank_pdf(file_path: str, password: str, account_id: str = "", skip_duplicates: bool = True, db_path: str | None = None, skip_validation: bool = False) -> dict[str, Any]:
-    try:
-        parsed = parse_kb_bank_pdf(file_path, password)
-    except FileNotFoundError as e:
-        return fail("not_found", str(e))
-    except ValueError as e:
-        return fail("pdf_password_error", str(e))
-    except Exception as e:
-        return fail("pdf_parse_error", str(e))
-    transactions = parsed["transactions"]
-    validation = parsed["validation"]
-    if not validation.get("ok") and not skip_validation:
-        return fail("balance_validation_failed", "KB Bank PDF balance flow validation failed", validation=validation)
-    source_hash = parsed["source_hash"]
-    metadata = parsed["metadata"]
-    ts = now_iso()
-    batch_id = new_id("imp")
-    inserted = 0
-    skipped_duplicates = 0
-    inserted_ids: list[str] = []
-    try:
-        with db(db_path) as conn:
-            linked_account_id = _find_or_create_kb_account(conn, metadata, account_id)
-            conn.execute(
-                """INSERT INTO import_batches(id,source,filename,imported_at,row_count,status,memo)
-                   VALUES(?,?,?,?,?,?,?)""",
-                (
-                    batch_id,
-                    "kb_bank_pdf",
-                    parsed["file_path"],
-                    ts,
-                    0,
-                    "active",
-                    json.dumps(
-                        {
-                            "source_hash": source_hash,
-                            "page_count": parsed["page_count"],
-                            "metadata": metadata,
-                            "validation": validation,
-                            "parsed_row_count": len(transactions),
-                        },
-                        ensure_ascii=False,
-                    ),
-                ),
-            )
-            for tx in transactions:
-                tx_id = new_id("tx")
-                external_id = tx.external_id(source_hash)
-                dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, tx.counterparty, linked_account_id, external_id)
-                payload = tx.raw_payload()
-                description = f"국민은행 PDF import; 적요={tx.raw_summary}; 거래점={tx.branch}; 잔액={tx.balance_after}"
-                try:
-                    conn.execute(
-                        """INSERT INTO transactions(id,transaction_date,posted_date,type,amount,currency,category,subcategory,merchant,description,account_id,payment_method,person,project,tags,is_fixed,source,import_batch_id,external_id,dedupe_key,memo,transaction_datetime,counterparty,balance_after,branch,raw_summary,raw_payload,created_at,updated_at)
-                           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
-                        (
-                            tx_id,
-                            tx.transaction_date,
-                            "",
-                            tx.type,
-                            tx.amount,
-                            "KRW",
-                            "",
-                            "",
-                            tx.counterparty,
-                            description,
-                            linked_account_id,
-                            tx.raw_summary,
-                            "",
-                            "",
-                            normalize_tags(["kb_bank_pdf", tx.raw_summary] if tx.raw_summary else ["kb_bank_pdf"]),
-                            0,
-                            "kb_bank_pdf",
-                            batch_id,
-                            external_id,
-                            dedupe_key,
-                            tx.remittance_memo,
-                            tx.transaction_datetime,
-                            tx.counterparty,
-                            tx.balance_after,
-                            tx.branch,
-                            tx.raw_summary,
-                            payload,
-                            ts,
-                            ts,
-                        ),
-                    )
-                    inserted += 1
-                    inserted_ids.append(tx_id)
-                except Exception as e:
-                    if "UNIQUE constraint failed" in str(e) and skip_duplicates:
-                        skipped_duplicates += 1
-                        continue
-                    raise
-            conn.execute("UPDATE import_batches SET row_count=? WHERE id=?", (inserted, batch_id))
-            if transactions:
-                conn.execute(
-                    "UPDATE accounts SET last_balance=?, last_synced_at=?, updated_at=? WHERE id=?",
-                    (transactions[-1].balance_after, ts, ts, linked_account_id),
-                )
-    except ValueError as e:
-        return fail("validation_error", str(e))
-    except Exception as e:
-        return fail("db_error", str(e))
-    return ok(
-        import_batch_id=batch_id,
-        account_id=linked_account_id,
-        source="kb_bank_pdf",
-        parsed_row_count=len(transactions),
-        inserted_count=inserted,
-        skipped_duplicate_count=skipped_duplicates,
-        inserted_transaction_ids=inserted_ids[:20],
-        validation=validation,
-        metadata=metadata,
-    )
-
-
-def _find_or_create_seoulpay_account(conn: Any, metadata: dict[str, Any], account_id: str = "") -> str:
-    if account_id:
-        row = conn.execute("SELECT id FROM accounts WHERE id=?", (account_id,)).fetchone()
-        if row:
-            return account_id
-        raise ValueError(f"account_id not found: {account_id}")
-    holder_name = metadata.get("holder_name") or ""
-    row = conn.execute(
-        "SELECT id FROM accounts WHERE institution=? AND name=? AND active=1 ORDER BY created_at LIMIT 1",
-        ("서울페이", f"서울페이 이용내역 {holder_name}".strip()),
-    ).fetchone()
-    if row:
-        return row["id"]
-    ts = now_iso()
-    new_account_id = new_id("acct")
-    display_name = f"서울페이 이용내역 {holder_name}".strip()
-    conn.execute(
-        """INSERT INTO accounts(id,name,type,institution,currency,active,memo,account_number_masked,holder_name,last_balance,last_synced_at,created_at,updated_at)
-           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?)""",
-        (
-            new_account_id,
-            display_name,
-            "cash",
-            "서울페이",
-            "KRW",
-            1,
-            "서울페이 이용내역 XLSX import로 자동 생성",
-            "",
-            holder_name,
-            None,
-            ts,
-            ts,
-            ts,
-        ),
-    )
-    return new_account_id
-
-
-def _classify_seoulpay_transaction(tx: Any) -> tuple[str, str, str, bool]:
-    merchant = (tx.merchant or "").strip()
-    text = f"{merchant} {tx.transaction_kind} {tx.payment_kind}"
-    if tx.type == "refund":
-        return "수입", "환급", merchant, False
-    if "땡겨요" in text or "배달" in text:
-        return "식비", "배달", merchant, False
-    if any(keyword in text for keyword in ("교습소", "학원", "바이올린", "피아노")):
-        return "교육", "학원", merchant, False
-    if any(keyword in text for keyword in ("김영모", "베이커리", "제과", "카페", "커피")):
-        return "식비", "카페", merchant, False
-    if "온라인" in text:
-        return "쇼핑", "온라인", merchant, False
-    return "쇼핑", "생활용품", merchant, False
-
-
-def _find_or_create_toss_bank_account(conn: Any, metadata: dict[str, Any], account_id: str = "") -> str:
-    if account_id:
-        row = conn.execute("SELECT id FROM accounts WHERE id=?", (account_id,)).fetchone()
-        if row:
-            return account_id
-        raise ValueError(f"account_id not found: {account_id}")
-    account_number_masked = metadata.get("account_number_masked") or ""
-    row = None
-    if account_number_masked:
-        row = conn.execute(
-            "SELECT id FROM accounts WHERE institution=? AND account_number_masked=? AND active=1 ORDER BY created_at LIMIT 1",
-            ("토스뱅크", account_number_masked),
-        ).fetchone()
-    if row:
-        return row["id"]
-    ts = now_iso()
-    new_account_id = new_id("acct")
-    account_name = metadata.get("account_name") or "토스뱅크 계좌"
-    display_name = f"토스뱅크 {account_name} {account_number_masked}".strip()
-    conn.execute(
-        """INSERT INTO accounts(id,name,type,institution,currency,active,memo,account_number_masked,holder_name,last_balance,last_synced_at,created_at,updated_at)
-           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?)""",
-        (
-            new_account_id,
-            display_name,
-            "bank",
-            "토스뱅크",
-            "KRW",
-            1,
-            "토스뱅크 거래내역 PDF import로 자동 생성",
-            account_number_masked,
-            metadata.get("holder_name") or "",
-            metadata.get("final_balance"),
-            ts,
-            ts,
-            ts,
-        ),
-    )
-    return new_account_id
-
-
-def _classify_toss_bank_transaction(tx: Any, owner_name: str = "") -> tuple[str, str, str, bool]:
-    action = (tx.action or "").strip()
-    counterparty = (tx.counterparty or "").strip()
-    owner_name = (owner_name or "").strip()
-    if "이자" in action:
-        return "수입", "이자", counterparty, False
-    if "캐시백" in action:
-        return "수입", "캐시백", counterparty, False
-    if "체크카드결제" in action:
-        return "식비", "카드결제", counterparty, False
-    transfer_like = {
-        "이체",
-        "내계좌간자동이체",
-        "오픈뱅킹",
-        "펌뱅킹출금",
-        "펌뱅킹입금",
-        "적금 개설",
-        "출금",
-        "입금",
-        "ATM출금",
-    }
-    if any(keyword in action for keyword in transfer_like):
-        if counterparty and owner_name and counterparty != owner_name:
-            return "이체", "확인필요", counterparty, False
-        return "이체", "본인계좌이동", counterparty, False
-    if "캐시백" in action:
-        return "수입", "캐시백", counterparty, False
-    return "이체", "확인필요", counterparty, False
-
-
-def _build_cross_transfer_index(transactions: list[dict[str, Any]]) -> dict[tuple[str, int, str], list[dict[str, Any]]]:
-    index: dict[tuple[str, int, str], list[dict[str, Any]]] = {}
-    for tx in transactions:
-        if tx.get("status") != "active":
-            continue
-        key = (tx["transaction_date"], int(tx["amount"]), tx["type"])
-        index.setdefault(key, []).append(tx)
-    return index
-
-
-def _fetch_transfer_candidates(db_conn: Any, source: str) -> list[dict[str, Any]]:
-    rows = db_conn.execute(
-        """
-        SELECT t.id, t.transaction_datetime, t.transaction_date, t.type, t.amount, t.category, t.subcategory,
-               t.counterparty, t.memo, t.person, t.account_id, t.payment_method, t.status,
-               a.holder_name, t.merchant
-        FROM transactions t
-        LEFT JOIN accounts a ON a.id = t.account_id
-        WHERE t.source=?
-          AND t.type IN ('income','expense')
-          AND (
-            t.category='이체'
-            OR t.subcategory IN ('확인필요', '본인계좌이동', '투자이동')
-            OR t.payment_method IN ('내계좌간자동이전', '내계좌간자동이체', '오픈뱅킹', '펌뱅킹출금', '펌뱅킹입금', '입금', '출금', 'ATM출금', '적금개설')
-            OR t.merchant LIKE '%이체%'
-          )
-        """,
-        (source,),
-    ).fetchall()
-    return [dict(r) for r in rows]
-
-
-def _parse_tx_datetime(value: str | None) -> datetime | None:
-    if not value:
-        return None
-    text = value.strip()
-    for fmt in ("%Y-%m-%dT%H:%M:%S", "%Y-%m-%d %H:%M:%S"):
-        try:
-            return datetime.strptime(text, fmt)
-        except ValueError:
-            continue
-    try:
-        return datetime.fromisoformat(text)
-    except ValueError:
-        return None
-
-
-def _self_transfer_direction_score(source_type: str, candidate_type: str) -> int:
-    if source_type == "expense":
-        if candidate_type == "income":
-            return 3
-        if candidate_type == "transfer":
-            return 2
-    elif source_type == "income":
-        if candidate_type == "expense":
-            return 3
-        if candidate_type == "transfer":
-            return 2
-    elif source_type == "transfer":
-        if candidate_type in {"income", "expense"}:
-            return 2
-        if candidate_type == "transfer":
-            return 1
-    return 0
-
-
-def _normalize_cancel_lookup_text(value: str) -> str:
-    if not value:
-        return ""
-    text = value.lower().strip()
-    replacements = [
-        "kb체크취소", "승인취소", "취소", "환불", "환급", "refund", "reversal",
-        "카드입금", "카드환급", "매출취소", "입금", "출금", "체크카드", "신용카드",
-    ]
-    for token in replacements:
-        text = text.replace(token, " ")
-    text = re.sub(r"\s+", " ", text)
-    return _normalize_counterparty_for_match(text)
-
-
-def _cancel_channel(value: str) -> str:
-    text = (value or "").lower()
-    if any(token in text for token in ("카드", "check", "credit")):
-        return "card"
-    if any(token in text for token in ("이체", "오픈뱅킹", "출금", "입금", "bank")):
-        return "bank"
-    return "other"
-
-
-def _is_cancel_like_transaction(tx: dict[str, Any]) -> bool:
-    if str(tx.get("type") or "") == "refund":
-        return True
-    combined = " ".join(
-        str(tx.get(key) or "") for key in ("merchant", "raw_summary", "payment_method", "memo", "description")
-    ).lower()
-    keywords = ("취소", "환불", "환급", "refund", "reversal", "카드입금", "매출취소")
-    return str(tx.get("type") or "") == "income" and any(token in combined for token in keywords)
-
-
-def find_cancel_original_candidates(transaction_id: str, limit: int = 20, lookback_days: int = 365, db_path: str | None = None) -> dict[str, Any]:
-    limit = max(1, min(int(limit), 50))
-    lookback_days = max(1, min(int(lookback_days), 730))
-    with db(db_path) as conn:
-        source_row = conn.execute(
-            """
-            SELECT t.*, a.name AS account_name, a.institution, a.holder_name
-            FROM transactions t
-            LEFT JOIN accounts a ON a.id = t.account_id
-            WHERE t.id=?
-            """,
-            (transaction_id,),
-        ).fetchone()
-        if not source_row:
-            return fail("not_found", f"transaction not found: {transaction_id}")
-        source = row_to_dict(source_row)
-        explicit_cancel_like = _is_cancel_like_transaction(source)
-        source_type = str(source.get("type") or "")
-        if source_type not in {"income", "refund"} and not explicit_cancel_like:
-            return fail("not_cancel_like", "cancel/original lookup is intended for 취소·환급성 거래")
-
-        source_date = normalize_date(source.get("transaction_date"), default_today=False)
-        source_dt = _parse_tx_datetime(source.get("transaction_datetime"))
-        source_day = date.fromisoformat(source_date)
-        start_date = date.fromordinal(source_day.toordinal() - lookback_days).isoformat()
-        source_text = " ".join(str(source.get(key) or "") for key in ("merchant", "counterparty", "raw_summary", "payment_method"))
-        source_norm_text = _normalize_cancel_lookup_text(source_text)
-        source_channel = _cancel_channel(str(source.get("payment_method") or "") + " " + str(source.get("raw_summary") or "") + " " + str(source.get("merchant") or ""))
-        candidate_rows = conn.execute(
-            """
-            SELECT t.*, a.name AS account_name, a.institution, a.holder_name
-            FROM transactions t
-            LEFT JOIN accounts a ON a.id = t.account_id
-            WHERE t.status='active'
-              AND t.id<>?
-              AND t.amount=?
-              AND t.transaction_date BETWEEN ? AND ?
-              AND t.type IN ('expense', 'adjustment')
-            ORDER BY t.transaction_date DESC, t.transaction_datetime DESC, t.created_at DESC
-            """,
-            (transaction_id, int(source.get("amount") or 0), start_date, source_date),
-        ).fetchall()
-
-    candidates: list[dict[str, Any]] = []
-    for row in candidate_rows:
-        candidate = row_to_dict(row)
-        candidate_date = normalize_date(candidate.get("transaction_date"), default_today=False)
-        candidate_day = date.fromisoformat(candidate_date)
-        day_delta = (source_day - candidate_day).days
-        if day_delta < 0:
-            continue
-        candidate_dt = _parse_tx_datetime(candidate.get("transaction_datetime"))
-        time_delta_seconds = abs(int((source_dt - candidate_dt).total_seconds())) if source_dt and candidate_dt else None
-        same_account = bool(source.get("account_id") and source.get("account_id") == candidate.get("account_id"))
-        source_same = bool(source.get("source") and source.get("source") == candidate.get("source"))
-        candidate_text = " ".join(str(candidate.get(key) or "") for key in ("merchant", "counterparty", "raw_summary", "payment_method"))
-        candidate_norm_text = _normalize_cancel_lookup_text(candidate_text)
-        merchant_score = 0
-        if source_norm_text and candidate_norm_text:
-            if source_norm_text == candidate_norm_text:
-                merchant_score = 4
-            elif source_norm_text in candidate_norm_text or candidate_norm_text in source_norm_text:
-                merchant_score = 2
-        channel_score = 0
-        candidate_channel = _cancel_channel(str(candidate.get("payment_method") or "") + " " + str(candidate.get("raw_summary") or "") + " " + str(candidate.get("merchant") or ""))
-        if source_channel != "other" and source_channel == candidate_channel:
-            channel_score = 2
-        if not explicit_cancel_like and not (same_account and merchant_score >= 2 and channel_score >= 1):
-            continue
-        recency_score = 0
-        if day_delta == 0:
-            recency_score = 5
-        elif day_delta <= 7:
-            recency_score = 4
-        elif day_delta <= 30:
-            recency_score = 3
-        elif day_delta <= 90:
-            recency_score = 2
-        else:
-            recency_score = 1
-        time_score = 0
-        if time_delta_seconds is not None:
-            if time_delta_seconds <= 3600:
-                time_score = 2
-            elif time_delta_seconds <= 86400:
-                time_score = 1
-        score = merchant_score + channel_score + recency_score + time_score + (3 if same_account else 0) + (1 if source_same else 0)
-        reason_bits = [
-            f"same_account={int(same_account)}",
-            f"merchant={merchant_score}",
-            f"channel={channel_score}",
-            f"days={day_delta}",
-        ]
-        if time_delta_seconds is not None:
-            reason_bits.append(f"time_delta={time_delta_seconds}s")
-        candidates.append({
-            "id": candidate.get("id"),
-            "transaction_date": candidate.get("transaction_date"),
-            "transaction_datetime": candidate.get("transaction_datetime"),
-            "source": candidate.get("source"),
-            "type": candidate.get("type"),
-            "amount": candidate.get("amount"),
-            "category": candidate.get("category"),
-            "subcategory": candidate.get("subcategory"),
-            "merchant": candidate.get("merchant"),
-            "raw_summary": candidate.get("raw_summary"),
-            "counterparty": candidate.get("counterparty"),
-            "account_id": candidate.get("account_id"),
-            "account_name": candidate.get("account_name"),
-            "institution": candidate.get("institution"),
-            "payment_method": candidate.get("payment_method"),
-            "memo": candidate.get("memo"),
-            "score": score,
-            "day_delta": day_delta,
-            "time_delta_seconds": time_delta_seconds,
-            "reason": ", ".join(reason_bits),
-        })
-
-    candidates.sort(key=lambda item: (-int(item.get("score", 0)), int(item.get("day_delta", 9999)), int(item.get("time_delta_seconds") or 99999999), str(item.get("transaction_datetime") or "")))
-    return ok(source_tx=source, count=len(candidates[:limit]), candidates=candidates[:limit], source="cancel_original_candidate_search")
-
-
-def find_self_transfer_candidates(transaction_id: str, limit: int = 20, db_path: str | None = None) -> dict[str, Any]:
-    limit = max(1, min(int(limit), 50))
-    with db(db_path) as conn:
-        source_row = conn.execute(
-            """
-            SELECT t.*, a.name AS account_name, a.institution, a.holder_name
-            FROM transactions t
-            LEFT JOIN accounts a ON a.id = t.account_id
-            WHERE t.id=?
-            """,
-            (transaction_id,),
-        ).fetchone()
-        if not source_row:
-            return fail("not_found", f"transaction not found: {transaction_id}")
-
-        source = row_to_dict(source_row)
-        source_date = normalize_date(source.get("transaction_date"), default_today=False)
-        source_dt = _parse_tx_datetime(source.get("transaction_datetime"))
-        source_holder = _normalize_counterparty_for_match(source.get("holder_name", ""))
-        anchor_date = date.fromisoformat(source_date) if source_date else current_kst_date()
-        window_start = anchor_date.toordinal() - 1
-        window_end = anchor_date.toordinal() + 1
-        candidate_rows = conn.execute(
-            """
-            SELECT t.*, a.name AS account_name, a.institution, a.holder_name
-            FROM transactions t
-            LEFT JOIN accounts a ON a.id = t.account_id
-            WHERE t.status='active'
-              AND t.id<>?
-              AND COALESCE(t.account_id, '')<>COALESCE(?, '')
-              AND t.amount=?
-              AND t.transaction_date BETWEEN ? AND ?
-            ORDER BY t.transaction_date, t.transaction_datetime, t.created_at
-            """,
-            (
-                transaction_id,
-                source.get("account_id", ""),
-                int(source.get("amount") or 0),
-                date.fromordinal(window_start).isoformat(),
-                date.fromordinal(window_end).isoformat(),
-            ),
-        ).fetchall()
-
-    candidates: list[dict[str, Any]] = []
-    for row in candidate_rows:
-        candidate = row_to_dict(row)
-        direction_score = _self_transfer_direction_score(str(source.get("type", "")), str(candidate.get("type", "")))
-        if direction_score == 0:
-            continue
-
-        candidate_dt = _parse_tx_datetime(candidate.get("transaction_datetime"))
-        day_delta = abs((date.fromisoformat(candidate.get("transaction_date", source_date)) - date.fromisoformat(source_date)).days)
-        time_delta_seconds = abs(int((candidate_dt - source_dt).total_seconds())) if candidate_dt and source_dt else None
-        same_owner = bool(source_holder and source_holder == _normalize_counterparty_for_match(candidate.get("holder_name", "")))
-        party_score = max(
-            _party_match_score(str(source.get("counterparty", "")), str(candidate.get("counterparty", ""))),
-            _party_match_score(str(source.get("counterparty", "")), str(candidate.get("holder_name", ""))),
-            _party_match_score(str(source.get("merchant", "")), str(candidate.get("counterparty", ""))),
-            _party_match_score(str(source.get("merchant", "")), str(candidate.get("holder_name", ""))),
-        )
-        source_gap_score = 1 if source.get("source") != candidate.get("source") else 0
-        day_score = 3 if day_delta == 0 else 1
-        time_score = 0
-        if time_delta_seconds is not None:
-            if time_delta_seconds <= 120:
-                time_score = 3
-            elif time_delta_seconds <= 3600:
-                time_score = 2
-            elif time_delta_seconds <= 86400:
-                time_score = 1
-        score = direction_score + party_score + source_gap_score + day_score + time_score + (3 if same_owner else 0)
-        reason_bits = [
-            f"direction={direction_score}",
-            f"party={party_score}",
-            f"same_owner={int(same_owner)}",
-            f"same_source={int(source.get('source') == candidate.get('source'))}",
-            f"day_delta={day_delta}",
-        ]
-        if time_delta_seconds is not None:
-            reason_bits.append(f"time_delta={time_delta_seconds}s")
-        candidates.append(
-            {
-                "id": candidate.get("id"),
-                "transaction_date": candidate.get("transaction_date"),
-                "transaction_datetime": candidate.get("transaction_datetime"),
-                "source": candidate.get("source"),
-                "type": candidate.get("type"),
-                "amount": candidate.get("amount"),
-                "category": candidate.get("category"),
-                "subcategory": candidate.get("subcategory"),
-                "merchant": candidate.get("merchant"),
-                "counterparty": candidate.get("counterparty"),
-                "account_id": candidate.get("account_id"),
-                "account_name": candidate.get("account_name"),
-                "institution": candidate.get("institution"),
-                "holder_name": candidate.get("holder_name"),
-                "memo": candidate.get("memo"),
-                "score": score,
-                "same_owner": same_owner,
-                "day_delta": day_delta,
-                "time_delta_seconds": time_delta_seconds,
-                "reason": ", ".join(reason_bits),
-            }
-        )
-
-    candidates.sort(key=lambda item: (-int(item.get("score", 0)), int(item.get("day_delta", 99)), int(item.get("time_delta_seconds") or 999999), str(item.get("transaction_datetime") or "")))
-    return ok(source_tx=source, count=len(candidates[:limit]), candidates=candidates[:limit], source="self_transfer_candidate_search")
-
-
-def link_self_transfer_pair(source_transaction_id: str, candidate_transaction_id: str, db_path: str | None = None) -> dict[str, Any]:
-    if not source_transaction_id or not candidate_transaction_id:
-        return fail("validation_error", "both transaction ids are required")
-    if source_transaction_id == candidate_transaction_id:
-        return fail("validation_error", "source and candidate transaction must differ")
-
-    preview = find_self_transfer_candidates(source_transaction_id, limit=50, db_path=db_path)
-    if not preview.get("success"):
-        return preview
-    candidates = preview.get("candidates") or []
-    candidate = next((item for item in candidates if item.get("id") == candidate_transaction_id), None)
-    if candidate is None:
-        return fail("candidate_not_found", f"candidate not found for source transaction: {candidate_transaction_id}")
-
-    link_tag = f"manual_self_transfer_link:{source_transaction_id}|{candidate_transaction_id}"
-    ts = now_iso()
-    with db(db_path) as conn:
-        source_row = conn.execute("SELECT * FROM transactions WHERE id=?", (source_transaction_id,)).fetchone()
-        candidate_row = conn.execute("SELECT * FROM transactions WHERE id=?", (candidate_transaction_id,)).fetchone()
-        if not source_row or not candidate_row:
-            return fail("not_found", "source or candidate transaction not found")
-        source_memo = _append_memo_tag(source_row["memo"], link_tag)
-        candidate_memo = _append_memo_tag(candidate_row["memo"], link_tag)
-        conn.execute(
-            "UPDATE transactions SET category='이체', subcategory='본인계좌이동', memo=?, updated_at=? WHERE id=?",
-            (source_memo, ts, source_transaction_id),
-        )
-        conn.execute(
-            "UPDATE transactions SET category='이체', subcategory='본인계좌이동', memo=?, updated_at=? WHERE id=?",
-            (candidate_memo, ts, candidate_transaction_id),
-        )
-        updated_source = conn.execute("SELECT * FROM transactions WHERE id=?", (source_transaction_id,)).fetchone()
-        updated_candidate = conn.execute("SELECT * FROM transactions WHERE id=?", (candidate_transaction_id,)).fetchone()
-
-    return ok(
-        source_transaction=row_to_dict(updated_source),
-        candidate_transaction=row_to_dict(updated_candidate),
-        link_tag=link_tag,
-        score=candidate.get("score"),
-        reason=candidate.get("reason"),
-        source="manual_self_transfer_link",
-    )
-
-
-def link_cancel_to_original(cancel_transaction_id: str, original_transaction_id: str, db_path: str | None = None) -> dict[str, Any]:
-    if not cancel_transaction_id or not original_transaction_id:
-        return fail("validation_error", "both transaction ids are required")
-    if cancel_transaction_id == original_transaction_id:
-        return fail("validation_error", "cancel and original transaction must differ")
-
-    preview = find_cancel_original_candidates(cancel_transaction_id, limit=50, db_path=db_path)
-    if not preview.get("success"):
-        return preview
-    candidates = preview.get("candidates") or []
-    original = next((item for item in candidates if item.get("id") == original_transaction_id), None)
-    if original is None:
-        return fail("candidate_not_found", f"original transaction candidate not found: {original_transaction_id}")
-
-    link_tag = f"manual_cancel_link:cancel={cancel_transaction_id}|original={original_transaction_id}"
-    ts = now_iso()
-    with db(db_path) as conn:
-        cancel_row = conn.execute("SELECT * FROM transactions WHERE id=?", (cancel_transaction_id,)).fetchone()
-        original_row = conn.execute("SELECT * FROM transactions WHERE id=?", (original_transaction_id,)).fetchone()
-        if not cancel_row or not original_row:
-            return fail("not_found", "cancel or original transaction not found")
-
-        cancel_memo = _append_memo_tag(cancel_row["memo"], link_tag)
-        original_memo = _append_memo_tag(original_row["memo"], link_tag)
-        original_category = original_row["category"] or "수입"
-        original_subcategory = original_row["subcategory"] or "환급"
-        original_person = original_row["person"] or ""
-        conn.execute(
-            "UPDATE transactions SET type='refund', category=?, subcategory=?, person=?, memo=?, updated_at=? WHERE id=?",
-            (original_category, original_subcategory, original_person, cancel_memo, ts, cancel_transaction_id),
-        )
-        conn.execute(
-            "UPDATE transactions SET memo=?, updated_at=? WHERE id=?",
-            (original_memo, ts, original_transaction_id),
-        )
-        updated_cancel = conn.execute("SELECT * FROM transactions WHERE id=?", (cancel_transaction_id,)).fetchone()
-        updated_original = conn.execute("SELECT * FROM transactions WHERE id=?", (original_transaction_id,)).fetchone()
-
-    return ok(
-        cancel_transaction=row_to_dict(updated_cancel),
-        original_transaction=row_to_dict(updated_original),
-        link_tag=link_tag,
-        score=original.get("score"),
-        reason=original.get("reason"),
-        source="manual_cancel_link",
-    )
-
-
-def _tag_memos(conn: Any, tx_id: str, tag: str) -> None:
-    existing = conn.execute("SELECT memo FROM transactions WHERE id=?", (tx_id,)).fetchone()
-    if not existing:
-        return
-    memo = _append_memo_tag(existing["memo"], tag)
-    conn.execute("UPDATE transactions SET memo=?, updated_at=? WHERE id=?", (memo, now_iso(), tx_id))
-
-
-def _run_cross_account_transfer_match(limit: int = 500, dry_run: bool = True, db_path: str | None = None) -> dict[str, Any]:
-    limit = max(1, min(int(limit), 1000))
-    matched_pairs: list[dict[str, Any]] = []
-    review_items: list[dict[str, Any]] = []
-    matched_tx_ids: set[str] = set()
-
-    with db(db_path) as conn:
-        kb_rows = _fetch_transfer_candidates(conn, "kb_bank_pdf")
-        toss_rows = _fetch_transfer_candidates(conn, "toss_bank_pdf")
-        toss_index = _build_cross_transfer_index(toss_rows)
-
-        for kb in kb_rows:
-            if kb["status"] != "active":
-                continue
-            if _memo_has_tag(kb.get("memo", ""), CROSS_TRANSFER_MATCH_TAG):
-                continue
-            required_type = "income" if kb["type"] == "expense" else "expense"
-            candidates = [
-                t
-                for t in toss_index.get((kb["transaction_date"], int(kb["amount"]), required_type), [])
-                if t.get("status") == "active" and not _memo_has_tag(t.get("memo", ""), CROSS_TRANSFER_MATCH_TAG)
-            ]
-
-            if not candidates:
-                review_items.append(
-                    {
-                        "source_tx_id": kb["id"],
-                        "source": "kb_bank_pdf",
-                        "target_tx_id": "",
-                        "reason": "no_candidate_found",
-                        "type": kb["type"],
-                        "amount": kb["amount"],
-                        "transaction_date": kb["transaction_date"],
-                        "counterparty": kb["counterparty"],
-                    }
-                )
-                continue
-
-            scored = []
-            for candidate in candidates:
-                score = _party_match_score(kb.get("counterparty", ""), candidate.get("counterparty", ""))
-                scored.append((score, candidate["transaction_datetime"], candidate))
-            scored.sort(key=lambda item: (item[0], item[1]), reverse=True)
-            best_score = scored[0][0]
-            top = [c for c in scored if c[0] == best_score]
-
-            if best_score == 0 or len(top) != 1:
-                review_items.append(
-                    {
-                        "source_tx_id": kb["id"],
-                        "source": "kb_bank_pdf",
-                        "target_tx_id": "",
-                        "reason": "ambiguous_or_name_mismatch",
-                        "candidate_count": len(candidates),
-                        "best_name_score": best_score,
-                        "type": kb["type"],
-                        "amount": kb["amount"],
-                        "transaction_date": kb["transaction_date"],
-                        "counterparty": kb["counterparty"],
-                    }
-                )
-                continue
-
-            toss = top[0][2]
-            if toss["id"] in matched_tx_ids:
-                review_items.append(
-                    {
-                        "source_tx_id": kb["id"],
-                        "source": "kb_bank_pdf",
-                        "target_tx_id": toss["id"],
-                        "reason": "target_already_matched",
-                        "type": kb["type"],
-                        "amount": kb["amount"],
-                        "transaction_date": kb["transaction_date"],
-                        "counterparty": kb["counterparty"],
-                    }
-                )
-                continue
-
-            kb_party = _normalize_counterparty_for_match(kb.get("counterparty", ""))
-            toss_party = _normalize_counterparty_for_match(toss.get("counterparty", ""))
-            owner_candidates = {
-                _normalize_counterparty_for_match(kb.get("holder_name", "")),
-                _normalize_counterparty_for_match(toss.get("holder_name", "")),
-            }
-            if (kb_party and kb_party in owner_candidates) or (toss_party and toss_party in owner_candidates):
-                subcategory = "투자이동"
-            else:
-                subcategory = "확인필요"
-
-            matched_pairs.append(
-                {
-                    "kb_tx_id": kb["id"],
-                    "toss_tx_id": toss["id"],
-                    "transaction_date": kb["transaction_date"],
-                    "amount": kb["amount"],
-                    "counterparty_kb": kb["counterparty"],
-                    "counterparty_toss": toss["counterparty"],
-                    "subcategory": subcategory,
-                    "name_score": best_score,
-                    "reason": "paired_by_date_amount_party",
-                }
-            )
-            if not dry_run:
-                review_tag = f"{CROSS_TRANSFER_MATCH_TAG}kb={kb['id']}|toss={toss['id']}"
-                conn.execute(
-                    "UPDATE transactions SET category='이체', subcategory=?, memo=?, updated_at=? WHERE id=?",
-                    (subcategory, _append_memo_tag(kb.get("memo", ""), review_tag), now_iso(), kb["id"]),
-                )
-                conn.execute(
-                    "UPDATE transactions SET category='이체', subcategory=?, memo=?, updated_at=? WHERE id=?",
-                    (subcategory, _append_memo_tag(toss.get("memo", ""), review_tag), now_iso(), toss["id"]),
-                )
-
-            if not dry_run:
-                _tag_memos(conn, kb["id"], f"{CROSS_TRANSFER_REVIEW_TAG}paired")
-                _tag_memos(conn, toss["id"], f"{CROSS_TRANSFER_REVIEW_TAG}paired")
-            matched_tx_ids.add(kb["id"])
-            matched_tx_ids.add(toss["id"])
-
-            if len(matched_pairs) >= limit:
-                break
-
-        for review in review_items:
-            if review["source_tx_id"] in matched_tx_ids:
-                continue
-            if not dry_run:
-                _tag_memos(conn, review["source_tx_id"], f"{CROSS_TRANSFER_REVIEW_TAG}{review['reason']}")
-
-
-    return ok(
-        dry_run=dry_run,
-        matched_count=len(matched_pairs),
-        review_count=len(review_items),
-        matched_pairs=matched_pairs[:limit],
-        review_items=review_items[:limit],
-        source="kb_toss_cross_transfer_match",
-    )
-
-
-def preview_cross_account_transfers(limit: int = 500, db_path: str | None = None) -> dict[str, Any]:
-    return _run_cross_account_transfer_match(limit=limit, dry_run=True, db_path=db_path)
-
-
-def run_cross_account_transfer_match(limit: int = 500, db_path: str | None = None) -> dict[str, Any]:
-    return _run_cross_account_transfer_match(limit=limit, dry_run=False, db_path=db_path)
-
-
-def list_cross_transfer_review_queue(limit: int = 300, db_path: str | None = None) -> dict[str, Any]:
-    clauses = ["t.status='active'", "t.source IN ('kb_bank_pdf','toss_bank_pdf')"]
-    clauses.append("t.memo LIKE '%cross_transfer_review:%'")
-    where = " WHERE " + " AND ".join(clauses)
-    with db(db_path) as conn:
-        rows = conn.execute(
-            f"SELECT * FROM transactions t{where} ORDER BY t.transaction_date DESC, t.created_at DESC LIMIT ?",
-            (max(1, min(int(limit), 1000)),),
-        ).fetchall()
-    return ok(transactions=[row_to_dict(r) for r in rows], count=len(rows), source="cross_transfer_review_queue")
-
-
-def preview_seoulpay_xlsx_import(file_path: str, password: str = "", preview_limit: int = 20, db_path: str | None = None) -> dict[str, Any]:
-    try:
-        parsed = parse_seoulpay_xlsx(file_path, password=password)
-    except FileNotFoundError as e:
-        return fail("not_found", str(e))
-    except ValueError as e:
-        return fail("xlsx_parse_error", str(e))
-    except Exception as e:
-        return fail("xlsx_parse_error", str(e))
-
-    transactions = parsed["transactions"]
-    source_hash = parsed["source_hash"]
-    duplicate_count = 0
-    by_category: dict[str, dict[str, int]] = {}
-    with db(db_path) as conn:
-        linked_account_id = ""
-        metadata = parsed["metadata"]
-        holder_name = metadata.get("holder_name") or ""
-        if holder_name:
-            row = conn.execute(
-                "SELECT id FROM accounts WHERE institution=? AND name=? AND active=1 ORDER BY created_at LIMIT 1",
-                ("서울페이", f"서울페이 이용내역 {holder_name}".strip()),
-            ).fetchone()
-            linked_account_id = row["id"] if row else ""
-        for tx in transactions:
-            category, subcategory, _, _ = _classify_seoulpay_transaction(tx)
-            key = f"{category}/{subcategory}"
-            item = by_category.setdefault(key, {"count": 0, "amount": 0})
-            item["count"] += 1
-            item["amount"] += int(tx.amount)
-            merchant = f"서울페이 {tx.merchant}".strip()
-            dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, merchant, linked_account_id, tx.external_id(source_hash))
-            found = conn.execute("SELECT 1 FROM transactions WHERE dedupe_key=?", (dedupe_key,)).fetchone()
-            duplicate_count += 1 if found else 0
-
-    preview = [seoulpay_transaction_preview(tx, source_hash) for tx in transactions[: max(1, min(int(preview_limit), 100))]]
-    return ok(
-        source="seoulpay_xlsx",
-        file_path=parsed["file_path"],
-        source_hash=source_hash,
-        sheet_name=parsed["sheet_name"],
-        metadata=parsed["metadata"],
-        row_count=len(transactions),
-        duplicate_count=duplicate_count,
-        validation=parsed["validation"],
-        by_category=by_category,
-        preview=preview,
-    )
-
-
-def import_seoulpay_xlsx(file_path: str, password: str = "", account_id: str = "", skip_duplicates: bool = True, db_path: str | None = None) -> dict[str, Any]:
-    try:
-        parsed = parse_seoulpay_xlsx(file_path, password=password)
-    except FileNotFoundError as e:
-        return fail("not_found", str(e))
-    except ValueError as e:
-        return fail("xlsx_parse_error", str(e))
-    except Exception as e:
-        return fail("xlsx_parse_error", str(e))
-
-    transactions = parsed["transactions"]
-    validation = parsed["validation"]
-    if not validation.get("ok"):
-        return fail("xlsx_parse_error", "SeoulPay XLSX parsing produced invalid transactions", validation=validation)
-
-    source_hash = parsed["source_hash"]
-    metadata = parsed["metadata"]
-    ts = now_iso()
-    batch_id = new_id("imp")
-    inserted = 0
-    skipped_duplicates = 0
-    inserted_ids: list[str] = []
-
-    try:
-        with db(db_path) as conn:
-            linked_account_id = _find_or_create_seoulpay_account(conn, metadata, account_id)
-            conn.execute(
-                """INSERT INTO import_batches(id,source,filename,imported_at,row_count,status,memo)
-                   VALUES(?,?,?,?,?,?,?)""",
-                (
-                    batch_id,
-                    "seoulpay_xlsx",
-                    parsed["file_path"],
-                    ts,
-                    0,
-                    "active",
-                    json.dumps(
-                        {
-                            "source_hash": source_hash,
-                            "sheet_name": parsed["sheet_name"],
-                            "metadata": metadata,
-                            "validation": validation,
-                            "parsed_row_count": len(transactions),
-                        },
-                        ensure_ascii=False,
-                    ),
-                ),
-            )
-            for tx in transactions:
-                tx_id = new_id("tx")
-                external_id = tx.external_id(source_hash)
-                merchant = f"서울페이 {tx.merchant}".strip()
-                category, subcategory, person, is_fixed = _classify_seoulpay_transaction(tx)
-                dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, merchant, linked_account_id, external_id)
-                description = f"서울페이 XLSX import; 거래유형={tx.transaction_kind}; 결제구분={tx.payment_kind}; 거래번호={tx.transaction_no}; 가맹점ID={tx.merchant_id}"
-                try:
-                    conn.execute(
-                        """INSERT INTO transactions(id,transaction_date,posted_date,type,amount,currency,category,subcategory,merchant,description,account_id,payment_method,person,project,tags,is_fixed,source,import_batch_id,external_id,dedupe_key,memo,transaction_datetime,counterparty,balance_after,branch,raw_summary,raw_payload,created_at,updated_at)
-                           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
-                        (
-                            tx_id,
-                            tx.transaction_date,
-                            "",
-                            tx.type,
-                            tx.amount,
-                            "KRW",
-                            category,
-                            subcategory,
-                            merchant,
-                            description,
-                            linked_account_id,
-                            tx.payment_kind,
-                            person,
-                            "",
-                            normalize_tags(["seoulpay_xlsx", tx.channel, tx.payment_kind, tx.transaction_kind]),
-                            int(is_fixed),
-                            "seoulpay_xlsx",
-                            batch_id,
-                            external_id,
-                            dedupe_key,
-                            tx.payment_kind,
-                            tx.transaction_datetime,
-                            tx.merchant,
-                            None,
-                            tx.channel,
-                            tx.transaction_kind,
-                            tx.raw_payload(),
-                            ts,
-                            ts,
-                        ),
-                    )
-                    inserted += 1
-                    inserted_ids.append(tx_id)
-                except Exception as e:
-                    if "UNIQUE constraint failed" in str(e) and skip_duplicates:
-                        skipped_duplicates += 1
-                        continue
-                    raise
-            conn.execute("UPDATE import_batches SET row_count=? WHERE id=?", (inserted, batch_id))
-    except ValueError as e:
-        return fail("validation_error", str(e))
-    except Exception as e:
-        return fail("db_error", str(e))
-
-    return ok(
-        import_batch_id=batch_id,
-        account_id=linked_account_id,
-        source="seoulpay_xlsx",
-        parsed_row_count=len(transactions),
-        inserted_count=inserted,
-        skipped_duplicate_count=skipped_duplicates,
-        inserted_transaction_ids=inserted_ids[:20],
-        validation=validation,
-        metadata=metadata,
-    )
-
-
-def preview_toss_bank_pdf_import(file_path: str, preview_limit: int = 20, db_path: str | None = None) -> dict[str, Any]:
-    try:
-        parsed = parse_toss_bank_pdf(file_path)
-    except FileNotFoundError as e:
-        return fail("not_found", str(e))
-    except ValueError as e:
-        return fail("pdf_parse_error", str(e))
-    except Exception as e:
-        return fail("pdf_parse_error", str(e))
-
-    transactions = parsed["transactions"]
-    source_hash = parsed["source_hash"]
-    duplicate_count = 0
-    with db(db_path) as conn:
-        metadata = parsed["metadata"]
-        account_number_masked = metadata.get("account_number_masked") or ""
-        linked_account_id = ""
-        if account_number_masked:
-            row = conn.execute(
-                "SELECT id FROM accounts WHERE institution=? AND account_number_masked=? AND active=1 ORDER BY created_at LIMIT 1",
-                ("토스뱅크", account_number_masked),
-            ).fetchone()
-            linked_account_id = row["id"] if row else ""
-        for tx in transactions:
-            merchant = f"토스뱅크 {tx.action}" + (f" {tx.counterparty}" if tx.counterparty else "")
-            dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, merchant, linked_account_id, tx.external_id(source_hash))
-            found = conn.execute("SELECT 1 FROM transactions WHERE dedupe_key=?", (dedupe_key,)).fetchone()
-            duplicate_count += 1 if found else 0
-
-    preview = [toss_bank_transaction_preview(tx, source_hash) for tx in transactions[: max(1, min(int(preview_limit), 100))]]
-    by_action: dict[str, dict[str, int]] = {}
-    for tx in transactions:
-        item = by_action.setdefault(tx.action, {"count": 0, "amount": 0})
-        item["count"] += 1
-        item["amount"] += int(tx.amount)
-    return ok(
-        source="toss_bank_pdf",
-        file_path=parsed["file_path"],
-        source_hash=source_hash,
-        page_count=parsed["page_count"],
-        metadata=parsed["metadata"],
-        row_count=len(transactions),
-        duplicate_count=duplicate_count,
-        validation=parsed["validation"],
-        by_action=by_action,
-        preview=preview,
-    )
-
-
-def import_toss_bank_pdf(file_path: str, account_id: str = "", skip_duplicates: bool = True, db_path: str | None = None) -> dict[str, Any]:
-    try:
-        parsed = parse_toss_bank_pdf(file_path)
-    except FileNotFoundError as e:
-        return fail("not_found", str(e))
-    except ValueError as e:
-        return fail("pdf_parse_error", str(e))
-    except Exception as e:
-        return fail("pdf_parse_error", str(e))
-
-    transactions = parsed["transactions"]
-    validation = parsed["validation"]
-    if not validation.get("ok"):
-        return fail("pdf_parse_error", "Toss Bank PDF parsing produced no valid transactions", validation=validation)
-
-    source_hash = parsed["source_hash"]
-    metadata = parsed["metadata"]
-    ts = now_iso()
-    batch_id = new_id("imp")
-    inserted = 0
-    skipped_duplicates = 0
-    inserted_ids: list[str] = []
-
-    try:
-        with db(db_path) as conn:
-            linked_account_id = _find_or_create_toss_bank_account(conn, metadata, account_id)
-            conn.execute(
-                """INSERT INTO import_batches(id,source,filename,imported_at,row_count,status,memo)
-                   VALUES(?,?,?,?,?,?,?)""",
-                (
-                    batch_id,
-                    "toss_bank_pdf",
-                    parsed["file_path"],
-                    ts,
-                    0,
-                    "active",
-                    json.dumps(
-                        {
-                            "source_hash": source_hash,
-                            "page_count": parsed["page_count"],
-                            "metadata": metadata,
-                            "validation": validation,
-                            "parsed_row_count": len(transactions),
-                        },
-                        ensure_ascii=False,
-                    ),
-                ),
-            )
-            for tx in transactions:
-                tx_id = new_id("tx")
-                external_id = tx.external_id(source_hash)
-                merchant = f"토스뱅크 {tx.action}" + (f" {tx.counterparty}" if tx.counterparty else "")
-                category, subcategory, person, is_fixed = _classify_toss_bank_transaction(tx, metadata.get("holder_name") or "")
-                if category == "이체":
-                    tx_type = "transfer"
-                else:
-                    tx_type = tx.type
-                dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, merchant, linked_account_id, external_id)
-                description = f"토스뱅크 PDF import; 구분={tx.action}; 상대={tx.counterparty}; 잔액={tx.balance_after}"
-                payload = tx.raw_payload()
-                try:
-                    conn.execute(
-                        """INSERT INTO transactions(id,transaction_date,posted_date,type,amount,currency,category,subcategory,merchant,description,account_id,payment_method,person,project,tags,is_fixed,source,import_batch_id,external_id,dedupe_key,memo,transaction_datetime,counterparty,balance_after,branch,raw_summary,raw_payload,created_at,updated_at)
-                           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
-                        (
-                            tx_id,
-                            tx.transaction_date,
-                            "",
-                            tx_type,
-                            tx.amount,
-                            "KRW",
-                            category,
-                            subcategory,
-                            merchant,
-                            description,
-                            linked_account_id,
-                            tx.action,
-                            person,
-                            "",
-                            normalize_tags(["toss_bank_pdf", tx.action] + ([tx.counterparty] if tx.counterparty else [])),
-                            int(is_fixed),
-                            "toss_bank_pdf",
-                            batch_id,
-                            external_id,
-                            dedupe_key,
-                            tx.action,
-                            tx.transaction_datetime,
-                            tx.counterparty,
-                            tx.balance_after,
-                            "토스뱅크",
-                            tx.raw_summary,
-                            payload,
-                            ts,
-                            ts,
-                        ),
-                    )
-                    inserted += 1
-                    inserted_ids.append(tx_id)
-                except Exception as e:
-                    if "UNIQUE constraint failed" in str(e) and skip_duplicates:
-                        skipped_duplicates += 1
-                        continue
-                    raise
-            conn.execute("UPDATE import_batches SET row_count=? WHERE id=?", (inserted, batch_id))
-            if transactions:
-                conn.execute(
-                    "UPDATE accounts SET last_balance=?, last_synced_at=?, updated_at=? WHERE id=?",
-                    (transactions[-1].balance_after, ts, ts, linked_account_id),
-                )
-    except ValueError as e:
-        return fail("validation_error", str(e))
-    except Exception as e:
-        return fail("db_error", str(e))
-
-    return ok(
-        import_batch_id=batch_id,
-        account_id=linked_account_id,
-        source="toss_bank_pdf",
-        parsed_row_count=len(transactions),
-        inserted_count=inserted,
-        skipped_duplicate_count=skipped_duplicates,
-        inserted_transaction_ids=inserted_ids[:20],
-        validation=validation,
-        metadata=metadata,
-    )
-
-
-def _find_or_create_toss_securities_account(conn: Any, metadata: dict[str, Any], account_id: str = "") -> str:
-    if account_id:
-        row = conn.execute("SELECT id FROM accounts WHERE id=?", (account_id,)).fetchone()
-        if row:
-            return account_id
-        raise ValueError(f"account_id not found: {account_id}")
-    account_number_masked = metadata.get("account_number_masked") or ""
-    row = None
-    if account_number_masked:
-        row = conn.execute(
-            "SELECT id FROM accounts WHERE institution=? AND account_number_masked=? AND active=1 ORDER BY created_at LIMIT 1",
-            ("토스증권", account_number_masked),
-        ).fetchone()
-    if row:
-        return row["id"]
-    ts = now_iso()
-    new_account_id = new_id("acct")
-    display_name = f"토스증권 {metadata.get('account_name') or '계좌'} {account_number_masked}".strip()
-    conn.execute(
-        """INSERT INTO accounts(id,name,type,institution,currency,active,memo,account_number_masked,holder_name,last_balance,last_synced_at,created_at,updated_at)
-           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?)""",
-        (
-            new_account_id,
-            display_name,
-            "investment",
-            "토스증권",
-            "KRW",
-            1,
-            "토스증권 거래내역 PDF import로 자동 생성",
-            account_number_masked,
-            metadata.get("holder_name") or "",
-            None,
-            ts,
-            ts,
-            ts,
-        ),
-    )
-    return new_account_id
-
-
-def _classify_toss_transaction(tx: Any, owner_name: str = "") -> tuple[str, str, str, bool]:
-    action = tx.action
-    counterparty = (tx.name or "").strip()
-    owner_name = (owner_name or "").strip()
-    if "배당" in action:
-        return "수입", "배당", counterparty, False
-    if "이자" in action:
-        return "수입", "이자", counterparty, False
-    if "환급" in action:
-        return "수입", "환급", counterparty, False
-    if "이체" in action:
-        # 토스증권 입출금에서 상대가 계좌 소유자와 다르면 본인 투자 이동으로 확정하지 않는다.
-        # 예: 이체출금(국민은행) 조상문 -> 외부/타인 송금 가능성이 있어 확인필요.
-        if counterparty and owner_name and counterparty != owner_name:
-            return "이체", "확인필요", counterparty, False
-        return "이체", "투자이동", counterparty, False
-    return "이체", "투자이동", counterparty, False
-
-
-def preview_toss_securities_pdf_import(file_path: str, preview_limit: int = 20, db_path: str | None = None) -> dict[str, Any]:
-    try:
-        parsed = parse_toss_securities_pdf(file_path)
-    except FileNotFoundError as e:
-        return fail("not_found", str(e))
-    except ValueError as e:
-        return fail("pdf_parse_error", str(e))
-    except Exception as e:
-        return fail("pdf_parse_error", str(e))
-    transactions = parsed["transactions"]
-    source_hash = parsed["source_hash"]
-    duplicate_count = 0
-    with db(db_path) as conn:
-        metadata = parsed["metadata"]
-        account_number_masked = metadata.get("account_number_masked") or ""
-        linked_account_id = ""
-        if account_number_masked:
-            row = conn.execute(
-                "SELECT id FROM accounts WHERE institution=? AND account_number_masked=? AND active=1 ORDER BY created_at LIMIT 1",
-                ("토스증권", account_number_masked),
-            ).fetchone()
-            linked_account_id = row["id"] if row else ""
-        for tx in transactions:
-            merchant = f"토스증권 {tx.action}" + (f" {tx.name}" if tx.name else "")
-            dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, merchant, linked_account_id, tx.external_id(source_hash))
-            found = conn.execute("SELECT 1 FROM transactions WHERE dedupe_key=?", (dedupe_key,)).fetchone()
-            duplicate_count += 1 if found else 0
-    preview = [toss_transaction_preview(tx, source_hash) for tx in transactions[: max(1, min(int(preview_limit), 100))]]
-    by_action: dict[str, dict[str, int]] = {}
-    for tx in transactions:
-        item = by_action.setdefault(tx.action, {"count": 0, "amount": 0})
-        item["count"] += 1
-        item["amount"] += int(tx.amount)
-    return ok(
-        source="toss_securities_pdf",
-        file_path=parsed["file_path"],
-        source_hash=source_hash,
-        page_count=parsed["page_count"],
-        metadata=parsed["metadata"],
-        row_count=len(transactions),
-        duplicate_count=duplicate_count,
-        validation=parsed["validation"],
-        by_action=by_action,
-        preview=preview,
-    )
-
-
-def import_toss_securities_pdf(file_path: str, account_id: str = "", skip_duplicates: bool = True, db_path: str | None = None) -> dict[str, Any]:
-    try:
-        parsed = parse_toss_securities_pdf(file_path)
-    except FileNotFoundError as e:
-        return fail("not_found", str(e))
-    except ValueError as e:
-        return fail("pdf_parse_error", str(e))
-    except Exception as e:
-        return fail("pdf_parse_error", str(e))
-    transactions = parsed["transactions"]
-    validation = parsed["validation"]
-    if not validation.get("ok"):
-        return fail("pdf_parse_error", "Toss Securities PDF parsing produced no valid transactions", validation=validation)
-    source_hash = parsed["source_hash"]
-    metadata = parsed["metadata"]
-    ts = now_iso()
-    batch_id = new_id("imp")
-    inserted = 0
-    skipped_duplicates = 0
-    inserted_ids: list[str] = []
-    try:
-        with db(db_path) as conn:
-            linked_account_id = _find_or_create_toss_securities_account(conn, metadata, account_id)
-            conn.execute(
-                """INSERT INTO import_batches(id,source,filename,imported_at,row_count,status,memo)
-                   VALUES(?,?,?,?,?,?,?)""",
-                (
-                    batch_id,
-                    "toss_securities_pdf",
-                    parsed["file_path"],
-                    ts,
-                    0,
-                    "active",
-                    json.dumps(
-                        {
-                            "source_hash": source_hash,
-                            "page_count": parsed["page_count"],
-                            "metadata": metadata,
-                            "validation": validation,
-                            "parsed_row_count": len(transactions),
-                        },
-                        ensure_ascii=False,
-                    ),
-                ),
-            )
-            for tx in transactions:
-                tx_id = new_id("tx")
-                external_id = tx.external_id(source_hash)
-                merchant = f"토스증권 {tx.action}" + (f" {tx.name}" if tx.name else "")
-                category, subcategory, person, is_fixed = _classify_toss_transaction(tx, metadata.get("holder_name") or "")
-                dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, merchant, linked_account_id, external_id)
-                description = f"토스증권 PDF import; 구분={tx.action}; 종목/상대={tx.name}; 수량={tx.quantity}; 단가={tx.unit_price}; 수수료={tx.fee}; 세금={tx.tax}; 잔액={tx.balance_after}"
-                try:
-                    conn.execute(
-                        """INSERT INTO transactions(id,transaction_date,posted_date,type,amount,currency,category,subcategory,merchant,description,account_id,payment_method,person,project,tags,is_fixed,source,import_batch_id,external_id,dedupe_key,memo,transaction_datetime,counterparty,balance_after,branch,raw_summary,raw_payload,created_at,updated_at)
-                           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
-                        (
-                            tx_id,
-                            tx.transaction_date,
-                            "",
-                            tx.type,
-                            tx.amount,
-                            "KRW",
-                            category,
-                            subcategory,
-                            merchant,
-                            description,
-                            linked_account_id,
-                            tx.action,
-                            person,
-                            "",
-                            normalize_tags(["toss_securities_pdf", tx.section, tx.action, tx.code] if tx.code else ["toss_securities_pdf", tx.section, tx.action]),
-                            int(is_fixed),
-                            "toss_securities_pdf",
-                            batch_id,
-                            external_id,
-                            dedupe_key,
-                            tx.code,
-                            tx.transaction_date,
-                            tx.name,
-                            tx.balance_after,
-                            "토스증권",
-                            tx.action,
-                            tx.raw_payload(),
-                            ts,
-                            ts,
-                        ),
-                    )
-                    inserted += 1
-                    inserted_ids.append(tx_id)
-                except Exception as e:
-                    if "UNIQUE constraint failed" in str(e) and skip_duplicates:
-                        skipped_duplicates += 1
-                        continue
-                    raise
-            conn.execute("UPDATE import_batches SET row_count=? WHERE id=?", (inserted, batch_id))
-            if transactions:
-                conn.execute(
-                    "UPDATE accounts SET last_balance=?, last_synced_at=?, updated_at=? WHERE id=?",
-                    (transactions[-1].balance_after, ts, ts, linked_account_id),
-                )
-    except ValueError as e:
-        return fail("validation_error", str(e))
-    except Exception as e:
-        return fail("db_error", str(e))
-    return ok(
-        import_batch_id=batch_id,
-        account_id=linked_account_id,
-        source="toss_securities_pdf",
-        parsed_row_count=len(transactions),
-        inserted_count=inserted,
-        skipped_duplicate_count=skipped_duplicates,
-        inserted_transaction_ids=inserted_ids[:20],
-        validation=validation,
-        metadata=metadata,
-    )
-
-
-def list_ledger_import_batches(source: str = "", include_rolled_back: bool = True, limit: int = 50, db_path: str | None = None) -> dict[str, Any]:
-    clauses=[]; params=[]
-    if source:
-        clauses.append("source=?"); params.append(source)
-    if not include_rolled_back:
-        clauses.append("status='active'")
-    where = " WHERE " + " AND ".join(clauses) if clauses else ""
-    params.append(max(1, min(int(limit), 500)))
-    with db(db_path) as conn:
-        rows=conn.execute(f"SELECT * FROM import_batches{where} ORDER BY imported_at DESC LIMIT ?", params).fetchall()
-    return ok(import_batches=[row_to_dict(r) for r in rows], count=len(rows))
-
-
-def rollback_ledger_import_batch(import_batch_id: str, hard_delete: bool = False, db_path: str | None = None) -> dict[str, Any]:
-    with db(db_path) as conn:
-        batch = conn.execute("SELECT * FROM import_batches WHERE id=?", (import_batch_id,)).fetchone()
-        if not batch:
-            return fail("not_found", "import batch not found")
-        if hard_delete:
-            cur = conn.execute("DELETE FROM transactions WHERE import_batch_id=?", (import_batch_id,))
-        else:
-            cur = conn.execute("UPDATE transactions SET status='deleted', updated_at=? WHERE import_batch_id=? AND status='active'", (now_iso(), import_batch_id))
-        conn.execute("UPDATE import_batches SET status='rolled_back' WHERE id=?", (import_batch_id,))
-    return ok(import_batch_id=import_batch_id, affected_transactions=cur.rowcount, hard_delete=hard_delete)
+from .imports_service import (
+    import_kb_bank_pdf,
+    import_seoulpay_xlsx,
+    import_toss_bank_pdf,
+    import_toss_securities_pdf,
+    list_ledger_import_batches,
+    preview_kb_bank_pdf_import,
+    preview_ledger_csv_import,
+    preview_seoulpay_xlsx_import,
+    preview_toss_bank_pdf_import,
+    preview_toss_securities_pdf_import,
+    rollback_ledger_import_batch,
+)
+from .cross_transfer_service import (
+    find_cancel_original_candidates,
+    find_self_transfer_candidates,
+    link_cancel_to_original,
+    link_self_transfer_pair,
+    list_cross_transfer_review_queue,
+    preview_cross_account_transfers,
+    run_cross_account_transfer_match,
+)
diff --git a/jarca_ledger_mcp/application/imports_service.py b/jarca_ledger_mcp/application/imports_service.py
new file mode 100644
index 0000000..76460bc
--- /dev/null
+++ b/jarca_ledger_mcp/application/imports_service.py
@@ -0,0 +1,922 @@
+from __future__ import annotations
+
+from .shared import *
+from .shared import _append_memo_tag, _memo_has_tag, _normalize_counterparty_for_match, _party_match_score
+
+def preview_ledger_csv_import(file_path: str, source: str = "generic", db_path: str | None = None) -> dict[str, Any]:
+    path=Path(file_path).expanduser()
+    if not path.exists():
+        return fail("not_found", f"CSV file not found: {path}")
+    with path.open("r", encoding="utf-8-sig", newline="") as f:
+        rows=list(csv.DictReader(f))
+    preview=[]
+    for r in rows[:20]:
+        preview.append(r)
+    return ok(source=source, filename=str(path), row_count=len(rows), preview=preview, required_columns=["transaction_date/date", "amount", "type optional", "category optional", "merchant/description optional"])
+
+
+def preview_kb_bank_pdf_import(file_path: str, password: str, preview_limit: int = 20, db_path: str | None = None) -> dict[str, Any]:
+    try:
+        parsed = parse_kb_bank_pdf(file_path, password)
+    except FileNotFoundError as e:
+        return fail("not_found", str(e))
+    except ValueError as e:
+        return fail("pdf_password_error", str(e))
+    except Exception as e:
+        return fail("pdf_parse_error", str(e))
+    transactions = parsed["transactions"]
+    validation = parsed["validation"]
+    source_hash = parsed["source_hash"]
+    duplicate_count = 0
+    with db(db_path) as conn:
+        for tx in transactions:
+            dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, tx.counterparty, "", tx.external_id(source_hash))
+            found = conn.execute("SELECT 1 FROM transactions WHERE dedupe_key=?", (dedupe_key,)).fetchone()
+            duplicate_count += 1 if found else 0
+    preview = [transaction_preview(tx, source_hash) for tx in transactions[: max(1, min(int(preview_limit), 100))]]
+    return ok(
+        source="kb_bank_pdf",
+        file_path=parsed["file_path"],
+        source_hash=source_hash,
+        page_count=parsed["page_count"],
+        metadata=parsed["metadata"],
+        row_count=len(transactions),
+        duplicate_count=duplicate_count,
+        validation=validation,
+        preview=preview,
+    )
+
+
+def _find_or_create_kb_account(conn: Any, metadata: dict[str, Any], account_id: str = "") -> str:
+    if account_id:
+        row = conn.execute("SELECT id FROM accounts WHERE id=?", (account_id,)).fetchone()
+        if row:
+            return account_id
+        raise ValueError(f"account_id not found: {account_id}")
+    account_number_masked = metadata.get("account_number_masked") or ""
+    row = None
+    if account_number_masked:
+        row = conn.execute(
+            "SELECT id FROM accounts WHERE institution=? AND account_number_masked=? AND active=1 ORDER BY created_at LIMIT 1",
+            ("KB국민은행", account_number_masked),
+        ).fetchone()
+    if row:
+        return row["id"]
+    ts = now_iso()
+    new_account_id = new_id("acct")
+    account_name = metadata.get("account_name") or "KB국민은행 계좌"
+    display_name = f"KB국민은행 {account_name} {account_number_masked}".strip()
+    conn.execute(
+        """INSERT INTO accounts(id,name,type,institution,currency,active,memo,account_number_masked,holder_name,last_balance,last_synced_at,created_at,updated_at)
+           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?)""",
+        (
+            new_account_id,
+            display_name,
+            "bank",
+            "KB국민은행",
+            "KRW",
+            1,
+            "국민은행 거래내역 PDF import로 자동 생성",
+            account_number_masked,
+            metadata.get("holder_name") or "",
+            metadata.get("final_balance"),
+            ts,
+            ts,
+            ts,
+        ),
+    )
+    return new_account_id
+
+
+def import_kb_bank_pdf(file_path: str, password: str, account_id: str = "", skip_duplicates: bool = True, db_path: str | None = None, skip_validation: bool = False) -> dict[str, Any]:
+    try:
+        parsed = parse_kb_bank_pdf(file_path, password)
+    except FileNotFoundError as e:
+        return fail("not_found", str(e))
+    except ValueError as e:
+        return fail("pdf_password_error", str(e))
+    except Exception as e:
+        return fail("pdf_parse_error", str(e))
+    transactions = parsed["transactions"]
+    validation = parsed["validation"]
+    if not validation.get("ok") and not skip_validation:
+        return fail("balance_validation_failed", "KB Bank PDF balance flow validation failed", validation=validation)
+    source_hash = parsed["source_hash"]
+    metadata = parsed["metadata"]
+    ts = now_iso()
+    batch_id = new_id("imp")
+    inserted = 0
+    skipped_duplicates = 0
+    inserted_ids: list[str] = []
+    try:
+        with db(db_path) as conn:
+            linked_account_id = _find_or_create_kb_account(conn, metadata, account_id)
+            conn.execute(
+                """INSERT INTO import_batches(id,source,filename,imported_at,row_count,status,memo)
+                   VALUES(?,?,?,?,?,?,?)""",
+                (
+                    batch_id,
+                    "kb_bank_pdf",
+                    parsed["file_path"],
+                    ts,
+                    0,
+                    "active",
+                    json.dumps(
+                        {
+                            "source_hash": source_hash,
+                            "page_count": parsed["page_count"],
+                            "metadata": metadata,
+                            "validation": validation,
+                            "parsed_row_count": len(transactions),
+                        },
+                        ensure_ascii=False,
+                    ),
+                ),
+            )
+            for tx in transactions:
+                tx_id = new_id("tx")
+                external_id = tx.external_id(source_hash)
+                dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, tx.counterparty, linked_account_id, external_id)
+                payload = tx.raw_payload()
+                description = f"국민은행 PDF import; 적요={tx.raw_summary}; 거래점={tx.branch}; 잔액={tx.balance_after}"
+                try:
+                    conn.execute(
+                        """INSERT INTO transactions(id,transaction_date,posted_date,type,amount,currency,category,subcategory,merchant,description,account_id,payment_method,person,project,tags,is_fixed,source,import_batch_id,external_id,dedupe_key,memo,transaction_datetime,counterparty,balance_after,branch,raw_summary,raw_payload,created_at,updated_at)
+                           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
+                        (
+                            tx_id,
+                            tx.transaction_date,
+                            "",
+                            tx.type,
+                            tx.amount,
+                            "KRW",
+                            "",
+                            "",
+                            tx.counterparty,
+                            description,
+                            linked_account_id,
+                            tx.raw_summary,
+                            "",
+                            "",
+                            normalize_tags(["kb_bank_pdf", tx.raw_summary] if tx.raw_summary else ["kb_bank_pdf"]),
+                            0,
+                            "kb_bank_pdf",
+                            batch_id,
+                            external_id,
+                            dedupe_key,
+                            tx.remittance_memo,
+                            tx.transaction_datetime,
+                            tx.counterparty,
+                            tx.balance_after,
+                            tx.branch,
+                            tx.raw_summary,
+                            payload,
+                            ts,
+                            ts,
+                        ),
+                    )
+                    inserted += 1
+                    inserted_ids.append(tx_id)
+                except Exception as e:
+                    if "UNIQUE constraint failed" in str(e) and skip_duplicates:
+                        skipped_duplicates += 1
+                        continue
+                    raise
+            conn.execute("UPDATE import_batches SET row_count=? WHERE id=?", (inserted, batch_id))
+            if transactions:
+                conn.execute(
+                    "UPDATE accounts SET last_balance=?, last_synced_at=?, updated_at=? WHERE id=?",
+                    (transactions[-1].balance_after, ts, ts, linked_account_id),
+                )
+    except ValueError as e:
+        return fail("validation_error", str(e))
+    except Exception as e:
+        return fail("db_error", str(e))
+    return ok(
+        import_batch_id=batch_id,
+        account_id=linked_account_id,
+        source="kb_bank_pdf",
+        parsed_row_count=len(transactions),
+        inserted_count=inserted,
+        skipped_duplicate_count=skipped_duplicates,
+        inserted_transaction_ids=inserted_ids[:20],
+        validation=validation,
+        metadata=metadata,
+    )
+
+
+def _find_or_create_seoulpay_account(conn: Any, metadata: dict[str, Any], account_id: str = "") -> str:
+    if account_id:
+        row = conn.execute("SELECT id FROM accounts WHERE id=?", (account_id,)).fetchone()
+        if row:
+            return account_id
+        raise ValueError(f"account_id not found: {account_id}")
+    holder_name = metadata.get("holder_name") or ""
+    row = conn.execute(
+        "SELECT id FROM accounts WHERE institution=? AND name=? AND active=1 ORDER BY created_at LIMIT 1",
+        ("서울페이", f"서울페이 이용내역 {holder_name}".strip()),
+    ).fetchone()
+    if row:
+        return row["id"]
+    ts = now_iso()
+    new_account_id = new_id("acct")
+    display_name = f"서울페이 이용내역 {holder_name}".strip()
+    conn.execute(
+        """INSERT INTO accounts(id,name,type,institution,currency,active,memo,account_number_masked,holder_name,last_balance,last_synced_at,created_at,updated_at)
+           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?)""",
+        (
+            new_account_id,
+            display_name,
+            "cash",
+            "서울페이",
+            "KRW",
+            1,
+            "서울페이 이용내역 XLSX import로 자동 생성",
+            "",
+            holder_name,
+            None,
+            ts,
+            ts,
+            ts,
+        ),
+    )
+    return new_account_id
+
+
+def _classify_seoulpay_transaction(tx: Any) -> tuple[str, str, str, bool]:
+    merchant = (tx.merchant or "").strip()
+    text = f"{merchant} {tx.transaction_kind} {tx.payment_kind}"
+    if tx.type == "refund":
+        return "수입", "환급", merchant, False
+    if "땡겨요" in text or "배달" in text:
+        return "식비", "배달", merchant, False
+    if any(keyword in text for keyword in ("교습소", "학원", "바이올린", "피아노")):
+        return "교육", "학원", merchant, False
+    if any(keyword in text for keyword in ("김영모", "베이커리", "제과", "카페", "커피")):
+        return "식비", "카페", merchant, False
+    if "온라인" in text:
+        return "쇼핑", "온라인", merchant, False
+    return "쇼핑", "생활용품", merchant, False
+
+
+def _find_or_create_toss_bank_account(conn: Any, metadata: dict[str, Any], account_id: str = "") -> str:
+    if account_id:
+        row = conn.execute("SELECT id FROM accounts WHERE id=?", (account_id,)).fetchone()
+        if row:
+            return account_id
+        raise ValueError(f"account_id not found: {account_id}")
+    account_number_masked = metadata.get("account_number_masked") or ""
+    row = None
+    if account_number_masked:
+        row = conn.execute(
+            "SELECT id FROM accounts WHERE institution=? AND account_number_masked=? AND active=1 ORDER BY created_at LIMIT 1",
+            ("토스뱅크", account_number_masked),
+        ).fetchone()
+    if row:
+        return row["id"]
+    ts = now_iso()
+    new_account_id = new_id("acct")
+    account_name = metadata.get("account_name") or "토스뱅크 계좌"
+    display_name = f"토스뱅크 {account_name} {account_number_masked}".strip()
+    conn.execute(
+        """INSERT INTO accounts(id,name,type,institution,currency,active,memo,account_number_masked,holder_name,last_balance,last_synced_at,created_at,updated_at)
+           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?)""",
+        (
+            new_account_id,
+            display_name,
+            "bank",
+            "토스뱅크",
+            "KRW",
+            1,
+            "토스뱅크 거래내역 PDF import로 자동 생성",
+            account_number_masked,
+            metadata.get("holder_name") or "",
+            metadata.get("final_balance"),
+            ts,
+            ts,
+            ts,
+        ),
+    )
+    return new_account_id
+
+
+def _classify_toss_bank_transaction(tx: Any, owner_name: str = "") -> tuple[str, str, str, bool]:
+    action = (tx.action or "").strip()
+    counterparty = (tx.counterparty or "").strip()
+    owner_name = (owner_name or "").strip()
+    if "이자" in action:
+        return "수입", "이자", counterparty, False
+    if "캐시백" in action:
+        return "수입", "캐시백", counterparty, False
+    if "체크카드결제" in action:
+        return "식비", "카드결제", counterparty, False
+    transfer_like = {
+        "이체",
+        "내계좌간자동이체",
+        "오픈뱅킹",
+        "펌뱅킹출금",
+        "펌뱅킹입금",
+        "적금 개설",
+        "출금",
+        "입금",
+        "ATM출금",
+    }
+    if any(keyword in action for keyword in transfer_like):
+        if counterparty and owner_name and counterparty != owner_name:
+            return "이체", "확인필요", counterparty, False
+        return "이체", "본인계좌이동", counterparty, False
+    if "캐시백" in action:
+        return "수입", "캐시백", counterparty, False
+    return "이체", "확인필요", counterparty, False
+
+def preview_seoulpay_xlsx_import(file_path: str, password: str = "", preview_limit: int = 20, db_path: str | None = None) -> dict[str, Any]:
+    try:
+        parsed = parse_seoulpay_xlsx(file_path, password=password)
+    except FileNotFoundError as e:
+        return fail("not_found", str(e))
+    except ValueError as e:
+        return fail("xlsx_parse_error", str(e))
+    except Exception as e:
+        return fail("xlsx_parse_error", str(e))
+
+    transactions = parsed["transactions"]
+    source_hash = parsed["source_hash"]
+    duplicate_count = 0
+    by_category: dict[str, dict[str, int]] = {}
+    with db(db_path) as conn:
+        linked_account_id = ""
+        metadata = parsed["metadata"]
+        holder_name = metadata.get("holder_name") or ""
+        if holder_name:
+            row = conn.execute(
+                "SELECT id FROM accounts WHERE institution=? AND name=? AND active=1 ORDER BY created_at LIMIT 1",
+                ("서울페이", f"서울페이 이용내역 {holder_name}".strip()),
+            ).fetchone()
+            linked_account_id = row["id"] if row else ""
+        for tx in transactions:
+            category, subcategory, _, _ = _classify_seoulpay_transaction(tx)
+            key = f"{category}/{subcategory}"
+            item = by_category.setdefault(key, {"count": 0, "amount": 0})
+            item["count"] += 1
+            item["amount"] += int(tx.amount)
+            merchant = f"서울페이 {tx.merchant}".strip()
+            dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, merchant, linked_account_id, tx.external_id(source_hash))
+            found = conn.execute("SELECT 1 FROM transactions WHERE dedupe_key=?", (dedupe_key,)).fetchone()
+            duplicate_count += 1 if found else 0
+
+    preview = [seoulpay_transaction_preview(tx, source_hash) for tx in transactions[: max(1, min(int(preview_limit), 100))]]
+    return ok(
+        source="seoulpay_xlsx",
+        file_path=parsed["file_path"],
+        source_hash=source_hash,
+        sheet_name=parsed["sheet_name"],
+        metadata=parsed["metadata"],
+        row_count=len(transactions),
+        duplicate_count=duplicate_count,
+        validation=parsed["validation"],
+        by_category=by_category,
+        preview=preview,
+    )
+
+
+def import_seoulpay_xlsx(file_path: str, password: str = "", account_id: str = "", skip_duplicates: bool = True, db_path: str | None = None) -> dict[str, Any]:
+    try:
+        parsed = parse_seoulpay_xlsx(file_path, password=password)
+    except FileNotFoundError as e:
+        return fail("not_found", str(e))
+    except ValueError as e:
+        return fail("xlsx_parse_error", str(e))
+    except Exception as e:
+        return fail("xlsx_parse_error", str(e))
+
+    transactions = parsed["transactions"]
+    validation = parsed["validation"]
+    if not validation.get("ok"):
+        return fail("xlsx_parse_error", "SeoulPay XLSX parsing produced invalid transactions", validation=validation)
+
+    source_hash = parsed["source_hash"]
+    metadata = parsed["metadata"]
+    ts = now_iso()
+    batch_id = new_id("imp")
+    inserted = 0
+    skipped_duplicates = 0
+    inserted_ids: list[str] = []
+
+    try:
+        with db(db_path) as conn:
+            linked_account_id = _find_or_create_seoulpay_account(conn, metadata, account_id)
+            conn.execute(
+                """INSERT INTO import_batches(id,source,filename,imported_at,row_count,status,memo)
+                   VALUES(?,?,?,?,?,?,?)""",
+                (
+                    batch_id,
+                    "seoulpay_xlsx",
+                    parsed["file_path"],
+                    ts,
+                    0,
+                    "active",
+                    json.dumps(
+                        {
+                            "source_hash": source_hash,
+                            "sheet_name": parsed["sheet_name"],
+                            "metadata": metadata,
+                            "validation": validation,
+                            "parsed_row_count": len(transactions),
+                        },
+                        ensure_ascii=False,
+                    ),
+                ),
+            )
+            for tx in transactions:
+                tx_id = new_id("tx")
+                external_id = tx.external_id(source_hash)
+                merchant = f"서울페이 {tx.merchant}".strip()
+                category, subcategory, person, is_fixed = _classify_seoulpay_transaction(tx)
+                dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, merchant, linked_account_id, external_id)
+                description = f"서울페이 XLSX import; 거래유형={tx.transaction_kind}; 결제구분={tx.payment_kind}; 거래번호={tx.transaction_no}; 가맹점ID={tx.merchant_id}"
+                try:
+                    conn.execute(
+                        """INSERT INTO transactions(id,transaction_date,posted_date,type,amount,currency,category,subcategory,merchant,description,account_id,payment_method,person,project,tags,is_fixed,source,import_batch_id,external_id,dedupe_key,memo,transaction_datetime,counterparty,balance_after,branch,raw_summary,raw_payload,created_at,updated_at)
+                           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
+                        (
+                            tx_id,
+                            tx.transaction_date,
+                            "",
+                            tx.type,
+                            tx.amount,
+                            "KRW",
+                            category,
+                            subcategory,
+                            merchant,
+                            description,
+                            linked_account_id,
+                            tx.payment_kind,
+                            person,
+                            "",
+                            normalize_tags(["seoulpay_xlsx", tx.channel, tx.payment_kind, tx.transaction_kind]),
+                            int(is_fixed),
+                            "seoulpay_xlsx",
+                            batch_id,
+                            external_id,
+                            dedupe_key,
+                            tx.payment_kind,
+                            tx.transaction_datetime,
+                            tx.merchant,
+                            None,
+                            tx.channel,
+                            tx.transaction_kind,
+                            tx.raw_payload(),
+                            ts,
+                            ts,
+                        ),
+                    )
+                    inserted += 1
+                    inserted_ids.append(tx_id)
+                except Exception as e:
+                    if "UNIQUE constraint failed" in str(e) and skip_duplicates:
+                        skipped_duplicates += 1
+                        continue
+                    raise
+            conn.execute("UPDATE import_batches SET row_count=? WHERE id=?", (inserted, batch_id))
+    except ValueError as e:
+        return fail("validation_error", str(e))
+    except Exception as e:
+        return fail("db_error", str(e))
+
+    return ok(
+        import_batch_id=batch_id,
+        account_id=linked_account_id,
+        source="seoulpay_xlsx",
+        parsed_row_count=len(transactions),
+        inserted_count=inserted,
+        skipped_duplicate_count=skipped_duplicates,
+        inserted_transaction_ids=inserted_ids[:20],
+        validation=validation,
+        metadata=metadata,
+    )
+
+
+def preview_toss_bank_pdf_import(file_path: str, preview_limit: int = 20, db_path: str | None = None) -> dict[str, Any]:
+    try:
+        parsed = parse_toss_bank_pdf(file_path)
+    except FileNotFoundError as e:
+        return fail("not_found", str(e))
+    except ValueError as e:
+        return fail("pdf_parse_error", str(e))
+    except Exception as e:
+        return fail("pdf_parse_error", str(e))
+
+    transactions = parsed["transactions"]
+    source_hash = parsed["source_hash"]
+    duplicate_count = 0
+    with db(db_path) as conn:
+        metadata = parsed["metadata"]
+        account_number_masked = metadata.get("account_number_masked") or ""
+        linked_account_id = ""
+        if account_number_masked:
+            row = conn.execute(
+                "SELECT id FROM accounts WHERE institution=? AND account_number_masked=? AND active=1 ORDER BY created_at LIMIT 1",
+                ("토스뱅크", account_number_masked),
+            ).fetchone()
+            linked_account_id = row["id"] if row else ""
+        for tx in transactions:
+            merchant = f"토스뱅크 {tx.action}" + (f" {tx.counterparty}" if tx.counterparty else "")
+            dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, merchant, linked_account_id, tx.external_id(source_hash))
+            found = conn.execute("SELECT 1 FROM transactions WHERE dedupe_key=?", (dedupe_key,)).fetchone()
+            duplicate_count += 1 if found else 0
+
+    preview = [toss_bank_transaction_preview(tx, source_hash) for tx in transactions[: max(1, min(int(preview_limit), 100))]]
+    by_action: dict[str, dict[str, int]] = {}
+    for tx in transactions:
+        item = by_action.setdefault(tx.action, {"count": 0, "amount": 0})
+        item["count"] += 1
+        item["amount"] += int(tx.amount)
+    return ok(
+        source="toss_bank_pdf",
+        file_path=parsed["file_path"],
+        source_hash=source_hash,
+        page_count=parsed["page_count"],
+        metadata=parsed["metadata"],
+        row_count=len(transactions),
+        duplicate_count=duplicate_count,
+        validation=parsed["validation"],
+        by_action=by_action,
+        preview=preview,
+    )
+
+
+def import_toss_bank_pdf(file_path: str, account_id: str = "", skip_duplicates: bool = True, db_path: str | None = None) -> dict[str, Any]:
+    try:
+        parsed = parse_toss_bank_pdf(file_path)
+    except FileNotFoundError as e:
+        return fail("not_found", str(e))
+    except ValueError as e:
+        return fail("pdf_parse_error", str(e))
+    except Exception as e:
+        return fail("pdf_parse_error", str(e))
+
+    transactions = parsed["transactions"]
+    validation = parsed["validation"]
+    if not validation.get("ok"):
+        return fail("pdf_parse_error", "Toss Bank PDF parsing produced no valid transactions", validation=validation)
+
+    source_hash = parsed["source_hash"]
+    metadata = parsed["metadata"]
+    ts = now_iso()
+    batch_id = new_id("imp")
+    inserted = 0
+    skipped_duplicates = 0
+    inserted_ids: list[str] = []
+
+    try:
+        with db(db_path) as conn:
+            linked_account_id = _find_or_create_toss_bank_account(conn, metadata, account_id)
+            conn.execute(
+                """INSERT INTO import_batches(id,source,filename,imported_at,row_count,status,memo)
+                   VALUES(?,?,?,?,?,?,?)""",
+                (
+                    batch_id,
+                    "toss_bank_pdf",
+                    parsed["file_path"],
+                    ts,
+                    0,
+                    "active",
+                    json.dumps(
+                        {
+                            "source_hash": source_hash,
+                            "page_count": parsed["page_count"],
+                            "metadata": metadata,
+                            "validation": validation,
+                            "parsed_row_count": len(transactions),
+                        },
+                        ensure_ascii=False,
+                    ),
+                ),
+            )
+            for tx in transactions:
+                tx_id = new_id("tx")
+                external_id = tx.external_id(source_hash)
+                merchant = f"토스뱅크 {tx.action}" + (f" {tx.counterparty}" if tx.counterparty else "")
+                category, subcategory, person, is_fixed = _classify_toss_bank_transaction(tx, metadata.get("holder_name") or "")
+                if category == "이체":
+                    tx_type = "transfer"
+                else:
+                    tx_type = tx.type
+                dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, merchant, linked_account_id, external_id)
+                description = f"토스뱅크 PDF import; 구분={tx.action}; 상대={tx.counterparty}; 잔액={tx.balance_after}"
+                payload = tx.raw_payload()
+                try:
+                    conn.execute(
+                        """INSERT INTO transactions(id,transaction_date,posted_date,type,amount,currency,category,subcategory,merchant,description,account_id,payment_method,person,project,tags,is_fixed,source,import_batch_id,external_id,dedupe_key,memo,transaction_datetime,counterparty,balance_after,branch,raw_summary,raw_payload,created_at,updated_at)
+                           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
+                        (
+                            tx_id,
+                            tx.transaction_date,
+                            "",
+                            tx_type,
+                            tx.amount,
+                            "KRW",
+                            category,
+                            subcategory,
+                            merchant,
+                            description,
+                            linked_account_id,
+                            tx.action,
+                            person,
+                            "",
+                            normalize_tags(["toss_bank_pdf", tx.action] + ([tx.counterparty] if tx.counterparty else [])),
+                            int(is_fixed),
+                            "toss_bank_pdf",
+                            batch_id,
+                            external_id,
+                            dedupe_key,
+                            tx.action,
+                            tx.transaction_datetime,
+                            tx.counterparty,
+                            tx.balance_after,
+                            "토스뱅크",
+                            tx.raw_summary,
+                            payload,
+                            ts,
+                            ts,
+                        ),
+                    )
+                    inserted += 1
+                    inserted_ids.append(tx_id)
+                except Exception as e:
+                    if "UNIQUE constraint failed" in str(e) and skip_duplicates:
+                        skipped_duplicates += 1
+                        continue
+                    raise
+            conn.execute("UPDATE import_batches SET row_count=? WHERE id=?", (inserted, batch_id))
+            if transactions:
+                conn.execute(
+                    "UPDATE accounts SET last_balance=?, last_synced_at=?, updated_at=? WHERE id=?",
+                    (transactions[-1].balance_after, ts, ts, linked_account_id),
+                )
+    except ValueError as e:
+        return fail("validation_error", str(e))
+    except Exception as e:
+        return fail("db_error", str(e))
+
+    return ok(
+        import_batch_id=batch_id,
+        account_id=linked_account_id,
+        source="toss_bank_pdf",
+        parsed_row_count=len(transactions),
+        inserted_count=inserted,
+        skipped_duplicate_count=skipped_duplicates,
+        inserted_transaction_ids=inserted_ids[:20],
+        validation=validation,
+        metadata=metadata,
+    )
+
+
+def _find_or_create_toss_securities_account(conn: Any, metadata: dict[str, Any], account_id: str = "") -> str:
+    if account_id:
+        row = conn.execute("SELECT id FROM accounts WHERE id=?", (account_id,)).fetchone()
+        if row:
+            return account_id
+        raise ValueError(f"account_id not found: {account_id}")
+    account_number_masked = metadata.get("account_number_masked") or ""
+    row = None
+    if account_number_masked:
+        row = conn.execute(
+            "SELECT id FROM accounts WHERE institution=? AND account_number_masked=? AND active=1 ORDER BY created_at LIMIT 1",
+            ("토스증권", account_number_masked),
+        ).fetchone()
+    if row:
+        return row["id"]
+    ts = now_iso()
+    new_account_id = new_id("acct")
+    display_name = f"토스증권 {metadata.get('account_name') or '계좌'} {account_number_masked}".strip()
+    conn.execute(
+        """INSERT INTO accounts(id,name,type,institution,currency,active,memo,account_number_masked,holder_name,last_balance,last_synced_at,created_at,updated_at)
+           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?)""",
+        (
+            new_account_id,
+            display_name,
+            "investment",
+            "토스증권",
+            "KRW",
+            1,
+            "토스증권 거래내역 PDF import로 자동 생성",
+            account_number_masked,
+            metadata.get("holder_name") or "",
+            None,
+            ts,
+            ts,
+            ts,
+        ),
+    )
+    return new_account_id
+
+
+def _classify_toss_transaction(tx: Any, owner_name: str = "") -> tuple[str, str, str, bool]:
+    action = tx.action
+    counterparty = (tx.name or "").strip()
+    owner_name = (owner_name or "").strip()
+    if "배당" in action:
+        return "수입", "배당", counterparty, False
+    if "이자" in action:
+        return "수입", "이자", counterparty, False
+    if "환급" in action:
+        return "수입", "환급", counterparty, False
+    if "이체" in action:
+        # 토스증권 입출금에서 상대가 계좌 소유자와 다르면 본인 투자 이동으로 확정하지 않는다.
+        # 예: 이체출금(국민은행) 조상문 -> 외부/타인 송금 가능성이 있어 확인필요.
+        if counterparty and owner_name and counterparty != owner_name:
+            return "이체", "확인필요", counterparty, False
+        return "이체", "투자이동", counterparty, False
+    return "이체", "투자이동", counterparty, False
+
+
+def preview_toss_securities_pdf_import(file_path: str, preview_limit: int = 20, db_path: str | None = None) -> dict[str, Any]:
+    try:
+        parsed = parse_toss_securities_pdf(file_path)
+    except FileNotFoundError as e:
+        return fail("not_found", str(e))
+    except ValueError as e:
+        return fail("pdf_parse_error", str(e))
+    except Exception as e:
+        return fail("pdf_parse_error", str(e))
+    transactions = parsed["transactions"]
+    source_hash = parsed["source_hash"]
+    duplicate_count = 0
+    with db(db_path) as conn:
+        metadata = parsed["metadata"]
+        account_number_masked = metadata.get("account_number_masked") or ""
+        linked_account_id = ""
+        if account_number_masked:
+            row = conn.execute(
+                "SELECT id FROM accounts WHERE institution=? AND account_number_masked=? AND active=1 ORDER BY created_at LIMIT 1",
+                ("토스증권", account_number_masked),
+            ).fetchone()
+            linked_account_id = row["id"] if row else ""
+        for tx in transactions:
+            merchant = f"토스증권 {tx.action}" + (f" {tx.name}" if tx.name else "")
+            dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, merchant, linked_account_id, tx.external_id(source_hash))
+            found = conn.execute("SELECT 1 FROM transactions WHERE dedupe_key=?", (dedupe_key,)).fetchone()
+            duplicate_count += 1 if found else 0
+    preview = [toss_transaction_preview(tx, source_hash) for tx in transactions[: max(1, min(int(preview_limit), 100))]]
+    by_action: dict[str, dict[str, int]] = {}
+    for tx in transactions:
+        item = by_action.setdefault(tx.action, {"count": 0, "amount": 0})
+        item["count"] += 1
+        item["amount"] += int(tx.amount)
+    return ok(
+        source="toss_securities_pdf",
+        file_path=parsed["file_path"],
+        source_hash=source_hash,
+        page_count=parsed["page_count"],
+        metadata=parsed["metadata"],
+        row_count=len(transactions),
+        duplicate_count=duplicate_count,
+        validation=parsed["validation"],
+        by_action=by_action,
+        preview=preview,
+    )
+
+
+def import_toss_securities_pdf(file_path: str, account_id: str = "", skip_duplicates: bool = True, db_path: str | None = None) -> dict[str, Any]:
+    try:
+        parsed = parse_toss_securities_pdf(file_path)
+    except FileNotFoundError as e:
+        return fail("not_found", str(e))
+    except ValueError as e:
+        return fail("pdf_parse_error", str(e))
+    except Exception as e:
+        return fail("pdf_parse_error", str(e))
+    transactions = parsed["transactions"]
+    validation = parsed["validation"]
+    if not validation.get("ok"):
+        return fail("pdf_parse_error", "Toss Securities PDF parsing produced no valid transactions", validation=validation)
+    source_hash = parsed["source_hash"]
+    metadata = parsed["metadata"]
+    ts = now_iso()
+    batch_id = new_id("imp")
+    inserted = 0
+    skipped_duplicates = 0
+    inserted_ids: list[str] = []
+    try:
+        with db(db_path) as conn:
+            linked_account_id = _find_or_create_toss_securities_account(conn, metadata, account_id)
+            conn.execute(
+                """INSERT INTO import_batches(id,source,filename,imported_at,row_count,status,memo)
+                   VALUES(?,?,?,?,?,?,?)""",
+                (
+                    batch_id,
+                    "toss_securities_pdf",
+                    parsed["file_path"],
+                    ts,
+                    0,
+                    "active",
+                    json.dumps(
+                        {
+                            "source_hash": source_hash,
+                            "page_count": parsed["page_count"],
+                            "metadata": metadata,
+                            "validation": validation,
+                            "parsed_row_count": len(transactions),
+                        },
+                        ensure_ascii=False,
+                    ),
+                ),
+            )
+            for tx in transactions:
+                tx_id = new_id("tx")
+                external_id = tx.external_id(source_hash)
+                merchant = f"토스증권 {tx.action}" + (f" {tx.name}" if tx.name else "")
+                category, subcategory, person, is_fixed = _classify_toss_transaction(tx, metadata.get("holder_name") or "")
+                dedupe_key = make_dedupe_key(tx.transaction_date, tx.amount, merchant, linked_account_id, external_id)
+                description = f"토스증권 PDF import; 구분={tx.action}; 종목/상대={tx.name}; 수량={tx.quantity}; 단가={tx.unit_price}; 수수료={tx.fee}; 세금={tx.tax}; 잔액={tx.balance_after}"
+                try:
+                    conn.execute(
+                        """INSERT INTO transactions(id,transaction_date,posted_date,type,amount,currency,category,subcategory,merchant,description,account_id,payment_method,person,project,tags,is_fixed,source,import_batch_id,external_id,dedupe_key,memo,transaction_datetime,counterparty,balance_after,branch,raw_summary,raw_payload,created_at,updated_at)
+                           VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
+                        (
+                            tx_id,
+                            tx.transaction_date,
+                            "",
+                            tx.type,
+                            tx.amount,
+                            "KRW",
+                            category,
+                            subcategory,
+                            merchant,
+                            description,
+                            linked_account_id,
+                            tx.action,
+                            person,
+                            "",
+                            normalize_tags(["toss_securities_pdf", tx.section, tx.action, tx.code] if tx.code else ["toss_securities_pdf", tx.section, tx.action]),
+                            int(is_fixed),
+                            "toss_securities_pdf",
+                            batch_id,
+                            external_id,
+                            dedupe_key,
+                            tx.code,
+                            tx.transaction_date,
+                            tx.name,
+                            tx.balance_after,
+                            "토스증권",
+                            tx.action,
+                            tx.raw_payload(),
+                            ts,
+                            ts,
+                        ),
+                    )
+                    inserted += 1
+                    inserted_ids.append(tx_id)
+                except Exception as e:
+                    if "UNIQUE constraint failed" in str(e) and skip_duplicates:
+                        skipped_duplicates += 1
+                        continue
+                    raise
+            conn.execute("UPDATE import_batches SET row_count=? WHERE id=?", (inserted, batch_id))
+            if transactions:
+                conn.execute(
+                    "UPDATE accounts SET last_balance=?, last_synced_at=?, updated_at=? WHERE id=?",
+                    (transactions[-1].balance_after, ts, ts, linked_account_id),
+                )
+    except ValueError as e:
+        return fail("validation_error", str(e))
+    except Exception as e:
+        return fail("db_error", str(e))
+    return ok(
+        import_batch_id=batch_id,
+        account_id=linked_account_id,
+        source="toss_securities_pdf",
+        parsed_row_count=len(transactions),
+        inserted_count=inserted,
+        skipped_duplicate_count=skipped_duplicates,
+        inserted_transaction_ids=inserted_ids[:20],
+        validation=validation,
+        metadata=metadata,
+    )
+
+
+def list_ledger_import_batches(source: str = "", include_rolled_back: bool = True, limit: int = 50, db_path: str | None = None) -> dict[str, Any]:
+    clauses=[]; params=[]
+    if source:
+        clauses.append("source=?"); params.append(source)
+    if not include_rolled_back:
+        clauses.append("status='active'")
+    where = " WHERE " + " AND ".join(clauses) if clauses else ""
+    params.append(max(1, min(int(limit), 500)))
+    with db(db_path) as conn:
+        rows=conn.execute(f"SELECT * FROM import_batches{where} ORDER BY imported_at DESC LIMIT ?", params).fetchall()
+    return ok(import_batches=[row_to_dict(r) for r in rows], count=len(rows))
+
+
+def rollback_ledger_import_batch(import_batch_id: str, hard_delete: bool = False, db_path: str | None = None) -> dict[str, Any]:
+    with db(db_path) as conn:
+        batch = conn.execute("SELECT * FROM import_batches WHERE id=?", (import_batch_id,)).fetchone()
+        if not batch:
+            return fail("not_found", "import batch not found")
+        if hard_delete:
+            cur = conn.execute("DELETE FROM transactions WHERE import_batch_id=?", (import_batch_id,))
+        else:
+            cur = conn.execute("UPDATE transactions SET status='deleted', updated_at=? WHERE import_batch_id=? AND status='active'", (now_iso(), import_batch_id))
+        conn.execute("UPDATE import_batches SET status='rolled_back' WHERE id=?", (import_batch_id,))
+    return ok(import_batch_id=import_batch_id, affected_transactions=cur.rowcount, hard_delete=hard_delete)
diff --git a/jarca_ledger_mcp/application/reconciliation_service.py b/jarca_ledger_mcp/application/reconciliation_service.py
new file mode 100644
index 0000000..5efe4cd
--- /dev/null
+++ b/jarca_ledger_mcp/application/reconciliation_service.py
@@ -0,0 +1,94 @@
+from __future__ import annotations
+
+from .shared import *
+
+
+def _transaction_signed_amount(tx_type: str, amount: int) -> int:
+    if tx_type in {"income", "refund", "adjustment"}:
+        return int(amount)
+    if tx_type in {"expense", "transfer"}:
+        return -int(amount)
+    return 0
+
+
+def _account_balance_row(conn: sqlite3.Connection, account_id: str, baseline_at: str) -> dict[str, Any]:
+    date_cutoff = normalize_date(baseline_at[:10])
+    rows = conn.execute(
+        """SELECT type, amount, balance_after, transaction_date, transaction_datetime, created_at
+             FROM transactions
+            WHERE status='active' AND account_id=? AND transaction_date<=?
+            ORDER BY transaction_date ASC, transaction_datetime ASC, created_at ASC""",
+        (account_id, date_cutoff),
+    ).fetchall()
+    signed_total = sum(_transaction_signed_amount(row["type"], int(row["amount"] or 0)) for row in rows)
+    latest_balance_row = next((row for row in reversed(rows) if row["balance_after"] is not None), None)
+    if latest_balance_row is not None:
+        calculated_balance = int(latest_balance_row["balance_after"])
+        basis = "latest_transaction_balance_after"
+        basis_at = latest_balance_row["transaction_datetime"] or latest_balance_row["transaction_date"]
+    else:
+        calculated_balance = int(signed_total)
+        basis = "net_transaction_delta_without_opening_balance"
+        basis_at = baseline_at
+    return {
+        "calculated_balance": calculated_balance,
+        "calculation_basis": basis,
+        "calculation_basis_at": basis_at,
+        "transaction_count": len(rows),
+        "signed_transaction_total": int(signed_total),
+    }
+
+
+def get_balance_reconciliation_report(account_id: str = "", baseline_balance: int | None = None, baseline_at: str = "", source_note: str = "", memo: str = "", tolerance: int = BALANCE_RECONCILIATION_TOLERANCE_KRW, db_path: str | None = None) -> dict[str, Any]:
+    """Preview manual/partial-import account balance reconciliation without mutating transactions."""
+    try:
+        baseline_at_norm = normalize_snapshot_at(baseline_at.strip(), default_now=False) if baseline_at and baseline_at.strip() else None
+        tolerance_int = int(tolerance)
+        if tolerance_int < 0:
+            raise ValueError("tolerance must be non-negative")
+        baseline_balance_int = int(baseline_balance) if baseline_balance is not None else None
+    except Exception as e:
+        return fail("validation_error", str(e))
+    with db(db_path) as conn:
+        clauses = ["active=1"]
+        params: list[Any] = []
+        if account_id:
+            clauses.append("id=?")
+            params.append(account_id)
+        else:
+            placeholders = ",".join("?" for _ in RECONCILIATION_TARGET_STATUSES)
+            clauses.append(f"statement_status IN ({placeholders})")
+            params.extend(sorted(RECONCILIATION_TARGET_STATUSES))
+        accounts = conn.execute(f"SELECT * FROM accounts WHERE {' AND '.join(clauses)} ORDER BY name", params).fetchall()
+        items: list[dict[str, Any]] = []
+        for account in accounts:
+            effective_baseline = baseline_balance_int if baseline_balance_int is not None else account["last_balance"]
+            effective_at = baseline_at_norm or account["last_synced_at"] or now_iso()
+            calc = _account_balance_row(conn, account["id"], effective_at)
+            if effective_baseline is None:
+                difference = None
+                ratio = None
+                status = "BASELINE_REQUIRED"
+            else:
+                difference = int(effective_baseline) - int(calc["calculated_balance"])
+                ratio = abs(difference) / abs(int(effective_baseline)) if int(effective_baseline) else None
+                status = "OK" if abs(difference) <= tolerance_int else "REVIEW_NEEDED"
+            items.append({
+                "account_id": account["id"],
+                "account_name": account["name"],
+                "statement_status": account["statement_status"],
+                "is_reconciliation_target": account["statement_status"] in RECONCILIATION_TARGET_STATUSES,
+                "baseline_balance": effective_baseline,
+                "baseline_at": effective_at,
+                "calculated_balance": calc["calculated_balance"],
+                "difference": difference,
+                "tolerance": tolerance_int,
+                "reference_ratio_threshold": BALANCE_RECONCILIATION_REFERENCE_RATIO,
+                "difference_ratio": ratio,
+                "status": status,
+                "recommended_decisions": ["계좌이동", "실지출", "실수입", "투자이동", "분석제외", "확인필요"] if status == "REVIEW_NEEDED" else [],
+                "source_note": source_note,
+                "memo": memo,
+                **calc,
+            })
+    return ok(items=items, count=len(items), tolerance=tolerance_int, baseline_at=baseline_at_norm, routing="preview_only_manual_confirmation_required")
diff --git a/jarca_ledger_mcp/application/shared.py b/jarca_ledger_mcp/application/shared.py
index 0ed17cc..708a6fe 100644
--- a/jarca_ledger_mcp/application/shared.py
+++ b/jarca_ledger_mcp/application/shared.py
@@ -11,11 +11,17 @@ from typing import Any
 from uuid import uuid4
 from zoneinfo import ZoneInfo
 
+from ..infrastructure.parsers import (
+    parse_kb_bank_pdf,
+    parse_seoulpay_xlsx,
+    parse_toss_bank_pdf,
+    parse_toss_securities_pdf,
+    seoulpay_transaction_preview,
+    toss_bank_transaction_preview,
+    toss_transaction_preview,
+    transaction_preview,
+)
 from ..infrastructure.sqlite import db, init_db, resolve_db_path
-from ..kb_bank_pdf import parse_kb_bank_pdf, transaction_preview
-from ..toss_bank_pdf import parse_toss_bank_pdf, transaction_preview as toss_bank_transaction_preview
-from ..toss_securities_pdf import parse_toss_securities_pdf, transaction_preview as toss_transaction_preview
-from ..seoulpay_xlsx import parse_seoulpay_xlsx, transaction_preview as seoulpay_transaction_preview
 
 VALID_TRANSACTION_TYPES = {"income", "expense", "transfer", "refund", "adjustment"}
 VALID_ACCOUNT_TYPES = {"bank", "card", "cash", "investment", "loan", "other"}
diff --git a/jarca_ledger_mcp/application/transaction_report_services.py b/jarca_ledger_mcp/application/transaction_report_services.py
index b609205..d477704 100644
--- a/jarca_ledger_mcp/application/transaction_report_services.py
+++ b/jarca_ledger_mcp/application/transaction_report_services.py
@@ -23,98 +23,7 @@ def list_ledger_transactions(start_date: str = "", end_date: str = "", type: str
         rows = conn.execute(sql, params).fetchall()
     return ok(transactions=[row_to_dict(r) for r in rows], count=len(rows))
 
-
-def _transaction_signed_amount(tx_type: str, amount: int) -> int:
-    if tx_type in {"income", "refund", "adjustment"}:
-        return int(amount)
-    if tx_type in {"expense", "transfer"}:
-        return -int(amount)
-    return 0
-
-
-def _account_balance_row(conn: sqlite3.Connection, account_id: str, baseline_at: str) -> dict[str, Any]:
-    date_cutoff = normalize_date(baseline_at[:10])
-    rows = conn.execute(
-        """SELECT type, amount, balance_after, transaction_date, transaction_datetime, created_at
-             FROM transactions
-            WHERE status='active' AND account_id=? AND transaction_date<=?
-            ORDER BY transaction_date ASC, transaction_datetime ASC, created_at ASC""",
-        (account_id, date_cutoff),
-    ).fetchall()
-    signed_total = sum(_transaction_signed_amount(row["type"], int(row["amount"] or 0)) for row in rows)
-    latest_balance_row = next((row for row in reversed(rows) if row["balance_after"] is not None), None)
-    if latest_balance_row is not None:
-        calculated_balance = int(latest_balance_row["balance_after"])
-        basis = "latest_transaction_balance_after"
-        basis_at = latest_balance_row["transaction_datetime"] or latest_balance_row["transaction_date"]
-    else:
-        calculated_balance = int(signed_total)
-        basis = "net_transaction_delta_without_opening_balance"
-        basis_at = baseline_at
-    return {
-        "calculated_balance": calculated_balance,
-        "calculation_basis": basis,
-        "calculation_basis_at": basis_at,
-        "transaction_count": len(rows),
-        "signed_transaction_total": int(signed_total),
-    }
-
-
-def get_balance_reconciliation_report(account_id: str = "", baseline_balance: int | None = None, baseline_at: str = "", source_note: str = "", memo: str = "", tolerance: int = BALANCE_RECONCILIATION_TOLERANCE_KRW, db_path: str | None = None) -> dict[str, Any]:
-    """Preview manual/partial-import account balance reconciliation without mutating transactions."""
-    try:
-        baseline_at_norm = normalize_snapshot_at(baseline_at.strip(), default_now=False) if baseline_at and baseline_at.strip() else None
-        tolerance_int = int(tolerance)
-        if tolerance_int < 0:
-            raise ValueError("tolerance must be non-negative")
-        baseline_balance_int = int(baseline_balance) if baseline_balance is not None else None
-    except Exception as e:
-        return fail("validation_error", str(e))
-    with db(db_path) as conn:
-        clauses = ["active=1"]
-        params: list[Any] = []
-        if account_id:
-            clauses.append("id=?")
-            params.append(account_id)
-        else:
-            placeholders = ",".join("?" for _ in RECONCILIATION_TARGET_STATUSES)
-            clauses.append(f"statement_status IN ({placeholders})")
-            params.extend(sorted(RECONCILIATION_TARGET_STATUSES))
-        accounts = conn.execute(f"SELECT * FROM accounts WHERE {' AND '.join(clauses)} ORDER BY name", params).fetchall()
-        items: list[dict[str, Any]] = []
-        for account in accounts:
-            effective_baseline = baseline_balance_int if baseline_balance_int is not None else account["last_balance"]
-            effective_at = baseline_at_norm or account["last_synced_at"] or now_iso()
-            calc = _account_balance_row(conn, account["id"], effective_at)
-            if effective_baseline is None:
-                difference = None
-                ratio = None
-                status = "BASELINE_REQUIRED"
-            else:
-                difference = int(effective_baseline) - int(calc["calculated_balance"])
-                ratio = abs(difference) / abs(int(effective_baseline)) if int(effective_baseline) else None
-                status = "OK" if abs(difference) <= tolerance_int else "REVIEW_NEEDED"
-            items.append({
-                "account_id": account["id"],
-                "account_name": account["name"],
-                "statement_status": account["statement_status"],
-                "is_reconciliation_target": account["statement_status"] in RECONCILIATION_TARGET_STATUSES,
-                "baseline_balance": effective_baseline,
-                "baseline_at": effective_at,
-                "calculated_balance": calc["calculated_balance"],
-                "difference": difference,
-                "tolerance": tolerance_int,
-                "reference_ratio_threshold": BALANCE_RECONCILIATION_REFERENCE_RATIO,
-                "difference_ratio": ratio,
-                "status": status,
-                "recommended_decisions": ["계좌이동", "실지출", "실수입", "투자이동", "분석제외", "확인필요"] if status == "REVIEW_NEEDED" else [],
-                "source_note": source_note,
-                "memo": memo,
-                **calc,
-            })
-    return ok(items=items, count=len(items), tolerance=tolerance_int, baseline_at=baseline_at_norm, routing="preview_only_manual_confirmation_required")
-
-
+from .reconciliation_service import get_balance_reconciliation_report
 
 def list_ledger_budgets(month: str = "", db_path: str | None = None) -> dict[str, Any]:
     clauses=[]; params=[]
diff --git a/jarca_ledger_mcp/application/uncategorized_workbench_service.py b/jarca_ledger_mcp/application/uncategorized_workbench_service.py
new file mode 100644
index 0000000..ec3891a
--- /dev/null
+++ b/jarca_ledger_mcp/application/uncategorized_workbench_service.py
@@ -0,0 +1,246 @@
+from __future__ import annotations
+
+from .shared import *
+
+def _rule_matches(rule: dict[str, Any], tx: dict[str, Any]) -> bool:
+    if rule.get("transaction_type") and rule.get("transaction_type") != tx.get("type", ""):
+        return False
+    amount = int(tx.get("amount") or 0)
+    if rule.get("amount_min") is not None and amount < int(rule["amount_min"]):
+        return False
+    if rule.get("amount_max") is not None and amount > int(rule["amount_max"]):
+        return False
+    if rule.get("day_of_month") is not None:
+        tx_date = tx.get("transaction_datetime") or tx.get("transaction_date") or ""
+        try:
+            if datetime.fromisoformat(str(tx_date)[:10]).day != int(rule["day_of_month"]):
+                return False
+        except ValueError:
+            return False
+    value = str(tx.get(rule["match_field"], "") or "")
+    pattern = str(rule.get("pattern", ""))
+    match_type = rule.get("match_type") or "contains"
+    if match_type == "contains":
+        return pattern.lower() in value.lower()
+    if match_type == "equals":
+        return value == pattern
+    if match_type == "starts_with":
+        return value.startswith(pattern)
+    if match_type == "ends_with":
+        return value.endswith(pattern)
+    if match_type == "regex":
+        return re.search(pattern, value) is not None
+    return False
+
+def list_uncategorized_transactions(limit: int = 100, min_amount: int = 0, db_path: str | None = None) -> dict[str, Any]:
+    with db(db_path) as conn:
+        rows=conn.execute("""SELECT * FROM transactions WHERE status='active' AND COALESCE(category,'')='' AND amount>=? ORDER BY amount DESC, transaction_datetime DESC LIMIT ?""", (int(min_amount), max(1,min(int(limit),1000)))).fetchall()
+    return ok(transactions=[row_to_dict(r) for r in rows], count=len(rows))
+
+
+def get_uncategorized_merchant_summary(limit: int = 50, db_path: str | None = None) -> dict[str, Any]:
+    with db(db_path) as conn:
+        rows=conn.execute("""SELECT merchant, raw_summary, type, COUNT(*) count, SUM(amount) total_amount, MAX(amount) max_amount
+                             FROM transactions WHERE status='active' AND COALESCE(category,'')=''
+                             GROUP BY merchant, raw_summary, type ORDER BY total_amount DESC LIMIT ?""", (max(1,min(int(limit),500)),)).fetchall()
+    return ok(items=[row_to_dict(r) for r in rows], count=len(rows))
+
+
+def _recommend_category_for_transaction(tx: dict[str, Any], rules: list[dict[str, Any]], history: dict[tuple[str, str], list[dict[str, Any]]]) -> dict[str, Any]:
+    """Recommend a category for one transaction using deterministic, explainable signals."""
+    for rule in rules:
+        try:
+            if _rule_matches(rule, tx):
+                return {
+                    "category": rule.get("category", ""),
+                    "subcategory": rule.get("subcategory", ""),
+                    "confidence": round(float(rule.get("confidence") or 1.0), 2),
+                    "source": "rule",
+                    "reason": f"활성 규칙 `{rule.get('pattern', '')}` 매칭",
+                    "rule_id": rule.get("id", ""),
+                }
+        except re.error:
+            continue
+
+    key = (tx.get("merchant", "") or "", tx.get("type", "") or "")
+    candidates = history.get(key, [])
+    if candidates:
+        best = candidates[0]
+        count = int(best.get("count") or 0)
+        confidence = min(0.9, 0.55 + min(count, 5) * 0.07)
+        return {
+            "category": best.get("category", ""),
+            "subcategory": best.get("subcategory", ""),
+            "confidence": round(confidence, 2),
+            "source": "history",
+            "reason": f"같은 사용처의 기존 분류 {count}건",
+            "rule_id": "",
+        }
+
+    return {"category": "", "subcategory": "", "confidence": 0.0, "source": "none", "reason": "추천 근거 없음", "rule_id": ""}
+
+
+def get_uncategorized_workbench(
+    limit: int = 100,
+    min_amount: int = 0,
+    sort_by: str = "amount_desc",
+    db_path: str | None = None,
+) -> dict[str, Any]:
+    row_limit = max(1, min(int(limit), 1000))
+    minimum_amount = int(min_amount)
+    order_by_map = {
+        "amount_desc": "amount DESC, transaction_datetime DESC, transaction_date DESC, created_at DESC",
+        "date_desc": "transaction_datetime DESC, transaction_date DESC, created_at DESC, amount DESC",
+    }
+    order_by = order_by_map.get(sort_by, order_by_map["amount_desc"])
+    with db(db_path) as conn:
+        tx_rows = [row_to_dict(r) for r in conn.execute(
+            f"""SELECT id, transaction_date, transaction_datetime, type, amount, currency, merchant, raw_summary,
+                      counterparty, account_id, payment_method, person, is_fixed, source, memo, created_at
+                 FROM transactions
+                WHERE status='active' AND COALESCE(category,'')='' AND amount>=?
+                ORDER BY {order_by}
+                LIMIT ?""",
+            (minimum_amount, row_limit),
+        ).fetchall()]
+        total_row = conn.execute(
+            """SELECT COUNT(*) AS count, COALESCE(SUM(amount),0) AS total_amount,
+                      COUNT(DISTINCT COALESCE(NULLIF(merchant,''), raw_summary, counterparty, '')) AS merchant_group_count
+                 FROM transactions
+                WHERE status='active' AND COALESCE(category,'')='' AND amount>=?""",
+            (minimum_amount,),
+        ).fetchone()
+        history_rows = [row_to_dict(r) for r in conn.execute(
+            """SELECT merchant, type, category, subcategory, COUNT(*) AS count,
+                      SUM(amount) AS total_amount, MAX(transaction_date) AS latest_date
+                 FROM transactions
+                WHERE status='active' AND COALESCE(category,'')<>'' AND COALESCE(merchant,'')<>''
+                GROUP BY merchant, type, category, subcategory
+                ORDER BY merchant, type, count DESC, total_amount DESC"""
+        ).fetchall()]
+        rules = [row_to_dict(r) for r in conn.execute(
+            "SELECT * FROM category_rules WHERE active=1 ORDER BY priority, id"
+        ).fetchall()]
+
+    history: dict[tuple[str, str], list[dict[str, Any]]] = {}
+    for row in history_rows:
+        history.setdefault((row.get("merchant", "") or "", row.get("type", "") or ""), []).append(row)
+
+    txs: list[dict[str, Any]] = []
+    groups: dict[tuple[str, str, str], dict[str, Any]] = {}
+    for tx in tx_rows:
+        rec = _recommend_category_for_transaction(tx, rules, history)
+        public_tx = {
+            **tx,
+            "recommendation": rec,
+            "recommended_category": rec.get("category", ""),
+            "recommended_subcategory": rec.get("subcategory", ""),
+            "recommendation_confidence": rec.get("confidence", 0.0),
+            "recommendation_reason": rec.get("reason", ""),
+            "recommendation_source": rec.get("source", "none"),
+        }
+        txs.append(public_tx)
+        gkey = (tx.get("merchant", "") or "", tx.get("raw_summary", "") or "", tx.get("type", "") or "")
+        group = groups.setdefault(gkey, {
+            "merchant": gkey[0],
+            "raw_summary": gkey[1],
+            "type": gkey[2],
+            "count": 0,
+            "total_amount": 0,
+            "max_amount": 0,
+            "latest_date": "",
+            "sample_transaction_id": tx.get("id", ""),
+            "recommendation": rec,
+        })
+        group["count"] += 1
+        group["total_amount"] += int(tx.get("amount") or 0)
+        group["max_amount"] = max(int(group["max_amount"] or 0), int(tx.get("amount") or 0))
+        date_value = tx.get("transaction_datetime") or tx.get("transaction_date") or ""
+        if date_value > (group.get("latest_date") or ""):
+            group["latest_date"] = date_value
+        if not group["recommendation"].get("category") and rec.get("category"):
+            group["recommendation"] = rec
+
+    group_items = sorted(groups.values(), key=lambda g: (int(g["count"]), int(g["total_amount"])), reverse=True)
+    recommendable_count = sum(1 for tx in txs if tx.get("recommended_category"))
+    repeated_group_count = sum(1 for g in group_items if int(g.get("count") or 0) >= 2)
+    total = row_to_dict(total_row) if total_row else {"count": 0, "total_amount": 0, "merchant_group_count": 0}
+    return ok(
+        summary={
+            "uncategorized_count": int(total.get("count") or 0),
+            "uncategorized_total_amount": int(total.get("total_amount") or 0),
+            "merchant_group_count": int(total.get("merchant_group_count") or 0),
+            "recommendable_count": recommendable_count,
+            "repeated_group_count": repeated_group_count,
+        },
+        groups=group_items,
+        transactions=txs,
+        count=len(txs),
+    )
+
+
+def apply_uncategorized_recommendation(transaction_id: str, db_path: str | None = None) -> dict[str, Any]:
+    if not transaction_id:
+        return fail("validation_error", "transaction_id is required")
+    with db(db_path) as conn:
+        row = conn.execute("SELECT * FROM transactions WHERE id=? AND status='active'", (transaction_id,)).fetchone()
+        if not row:
+            return fail("not_found", "transaction not found")
+        tx = row_to_dict(row)
+        if tx.get("category"):
+            return fail("validation_error", "transaction is already categorized")
+        rules = [row_to_dict(r) for r in conn.execute("SELECT * FROM category_rules WHERE active=1 ORDER BY priority, id").fetchall()]
+        history_rows = [row_to_dict(r) for r in conn.execute(
+            """SELECT merchant, type, category, subcategory, COUNT(*) AS count, SUM(amount) AS total_amount
+                 FROM transactions
+                WHERE status='active' AND COALESCE(category,'')<>'' AND COALESCE(merchant,'')<>''
+                GROUP BY merchant, type, category, subcategory
+                ORDER BY count DESC, total_amount DESC"""
+        ).fetchall()]
+        history: dict[tuple[str, str], list[dict[str, Any]]] = {}
+        for h in history_rows:
+            history.setdefault((h.get("merchant", "") or "", h.get("type", "") or ""), []).append(h)
+        rec = _recommend_category_for_transaction(tx, rules, history)
+        if not rec.get("category"):
+            return fail("no_recommendation", "추천 분류가 없습니다")
+        cur = conn.execute(
+            """UPDATE transactions
+                  SET category=?, subcategory=?, updated_at=?
+                WHERE id=? AND status='active' AND COALESCE(category,'')=''""",
+            (rec.get("category", ""), rec.get("subcategory", ""), now_iso(), transaction_id),
+        )
+    return ok(applied_count=cur.rowcount, transaction_id=transaction_id, recommendation=rec)
+
+
+def apply_uncategorized_group_recommendation(merchant: str, raw_summary: str = "", transaction_type: str = "", db_path: str | None = None) -> dict[str, Any]:
+    with db(db_path) as conn:
+        sample = conn.execute(
+            """SELECT * FROM transactions
+                WHERE status='active' AND COALESCE(category,'')='' AND merchant=? AND raw_summary=? AND type=?
+                ORDER BY amount DESC, transaction_datetime DESC, transaction_date DESC LIMIT 1""",
+            (merchant, raw_summary, transaction_type),
+        ).fetchone()
+        if not sample:
+            return fail("not_found", "uncategorized group not found")
+        tx = row_to_dict(sample)
+        rules = [row_to_dict(r) for r in conn.execute("SELECT * FROM category_rules WHERE active=1 ORDER BY priority, id").fetchall()]
+        history_rows = [row_to_dict(r) for r in conn.execute(
+            """SELECT merchant, type, category, subcategory, COUNT(*) AS count, SUM(amount) AS total_amount
+                 FROM transactions
+                WHERE status='active' AND COALESCE(category,'')<>'' AND COALESCE(merchant,'')<>''
+                GROUP BY merchant, type, category, subcategory
+                ORDER BY count DESC, total_amount DESC"""
+        ).fetchall()]
+        history: dict[tuple[str, str], list[dict[str, Any]]] = {}
+        for h in history_rows:
+            history.setdefault((h.get("merchant", "") or "", h.get("type", "") or ""), []).append(h)
+        rec = _recommend_category_for_transaction(tx, rules, history)
+        if not rec.get("category"):
+            return fail("no_recommendation", "추천 분류가 없습니다")
+        cur = conn.execute(
+            """UPDATE transactions
+                  SET category=?, subcategory=?, updated_at=?
+                WHERE status='active' AND COALESCE(category,'')='' AND merchant=? AND raw_summary=? AND type=?""",
+            (rec.get("category", ""), rec.get("subcategory", ""), now_iso(), merchant, raw_summary, transaction_type),
+        )
+    return ok(applied_count=cur.rowcount, merchant=merchant, raw_summary=raw_summary, type=transaction_type, recommendation=rec)
diff --git a/jarca_ledger_mcp/infrastructure/parsers/__init__.py b/jarca_ledger_mcp/infrastructure/parsers/__init__.py
new file mode 100644
index 0000000..3120591
--- /dev/null
+++ b/jarca_ledger_mcp/infrastructure/parsers/__init__.py
@@ -0,0 +1,4 @@
+from .kb_bank_pdf import parse_kb_bank_pdf, transaction_preview
+from .toss_bank_pdf import parse_toss_bank_pdf, transaction_preview as toss_bank_transaction_preview
+from .toss_securities_pdf import parse_toss_securities_pdf, transaction_preview as toss_transaction_preview
+from .seoulpay_xlsx import parse_seoulpay_xlsx, transaction_preview as seoulpay_transaction_preview
diff --git a/jarca_ledger_mcp/infrastructure/parsers/kb_bank_pdf.py b/jarca_ledger_mcp/infrastructure/parsers/kb_bank_pdf.py
new file mode 100644
index 0000000..c6e9bce
--- /dev/null
+++ b/jarca_ledger_mcp/infrastructure/parsers/kb_bank_pdf.py
@@ -0,0 +1 @@
+from ...kb_bank_pdf import parse_kb_bank_pdf, transaction_preview
diff --git a/jarca_ledger_mcp/infrastructure/parsers/seoulpay_xlsx.py b/jarca_ledger_mcp/infrastructure/parsers/seoulpay_xlsx.py
new file mode 100644
index 0000000..997662a
--- /dev/null
+++ b/jarca_ledger_mcp/infrastructure/parsers/seoulpay_xlsx.py
@@ -0,0 +1 @@
+from ...seoulpay_xlsx import parse_seoulpay_xlsx, transaction_preview
diff --git a/jarca_ledger_mcp/infrastructure/parsers/toss_bank_pdf.py b/jarca_ledger_mcp/infrastructure/parsers/toss_bank_pdf.py
new file mode 100644
index 0000000..f7fc844
--- /dev/null
+++ b/jarca_ledger_mcp/infrastructure/parsers/toss_bank_pdf.py
@@ -0,0 +1 @@
+from ...toss_bank_pdf import parse_toss_bank_pdf, transaction_preview
diff --git a/jarca_ledger_mcp/infrastructure/parsers/toss_securities_pdf.py b/jarca_ledger_mcp/infrastructure/parsers/toss_securities_pdf.py
new file mode 100644
index 0000000..8db351f
--- /dev/null
+++ b/jarca_ledger_mcp/infrastructure/parsers/toss_securities_pdf.py
@@ -0,0 +1 @@
+from ...toss_securities_pdf import parse_toss_securities_pdf, transaction_preview
diff --git a/jarca_ledger_mcp/service.py b/jarca_ledger_mcp/service.py
index 90b085f..f9e1d6c 100644
--- a/jarca_ledger_mcp/service.py
+++ b/jarca_ledger_mcp/service.py
@@ -5,20 +5,30 @@ from types import ModuleType
 
 from .application import accounts_write_services as _accounts_write_services
 from .application import asset_account_services as _asset_account_services
+from .application import assets_service as _assets_service
 from .application import budgets_write_services as _budgets_write_services
 from .application import category_rules_write_services as _category_rules_write_services
 from .application import categorization_services as _categorization_services
+from .application import cross_transfer_service as _cross_transfer_service
+from .application import imports_service as _imports_service
 from .application import import_transfer_services as _import_transfer_services
+from .application import reconciliation_service as _reconciliation_service
 from .application import shared as _shared
 from .application import system_services as _system_services
 from .application import transactions_write_services as _transactions_write_services
 from .application import transaction_report_services as _transaction_report_services
+from .application import uncategorized_workbench_service as _uncategorized_workbench_service
 from .application.asset_account_services import *
+from .application.assets_service import *
 from .application.categorization_services import *
+from .application.cross_transfer_service import *
+from .application.imports_service import *
 from .application.import_transfer_services import *
+from .application.reconciliation_service import *
 from .application.shared import *
 from .application.system_services import *
 from .application.transaction_report_services import *
+from .application.uncategorized_workbench_service import *
 
 # Keep compatibility for tests/monkeypatches that reach into service internals.
 _current_kst_datetime = _shared._current_kst_datetime
@@ -28,11 +38,16 @@ _PROXY_MODULES = (
     _accounts_write_services,
     _system_services,
     _asset_account_services,
+    _assets_service,
     _budgets_write_services,
     _category_rules_write_services,
     _transaction_report_services,
+    _reconciliation_service,
+    _imports_service,
     _import_transfer_services,
+    _cross_transfer_service,
     _categorization_services,
+    _uncategorized_workbench_service,
     _transactions_write_services,
 )