@kansei-link/bantou 0.2.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Synapse Arrows PTE. LTD.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -0,0 +1,110 @@
+# @kansei-link/cockpit
+
+MCP (Model Context Protocol) server for Japanese tax accountant workflow automation.
+
+税理士業務の自動化を実現する MCP サーバー。仕訳分類・消費税判定・インボイス制度チェック・CSV取込を一気通貫で処理します。
+
+## Installation
+
+```bash
+npm install @kansei-link/cockpit
+```
+
+## Quick Start
+
+Add to your `claude_desktop_config.json`:
+
+```json
+{
+  "mcpServers": {
+    "cockpit": {
+      "command": "npx",
+      "args": ["@kansei-link/cockpit"],
+      "env": {
+        "COCKPIT_DATA_DIR": "/path/to/data"
+      }
+    }
+  }
+}
+```
+
+## Key Features
+
+- **Two-stage classifier** -- Stage 1: keyword matching across 19 categories with 500+ keywords; Stage 2: Claude AI fallback for ambiguous entries
+- **TaxRuleEngine** -- Consumption tax with 8%/10% reduced rate logic, overseas SaaS and ad platform detection, withholding tax calculation, asset tier classification, and invoice system (インボイス制度) checker with transitional period handling
+- **CSV adapters** -- Auto-detect and parse Yayoi (弥生), freee, and generic CSV formats
+- **Persistent memory** -- Linksee Memory integration that learns from corrections and recalls learned patterns
+- **Confidence routing** -- Automatic routing to `auto_register`, `auto_register_with_log`, or `human_review` based on confidence scores and business rules
+- **200 tests** -- Comprehensive test suite including end-to-end integration tests
+
+## Architecture
+
+The classification pipeline processes each transaction through sequential stages:
+
+```
+CSV Input
+  |
+  v
+[CSV Adapter] -- Auto-detect format (Yayoi / freee / generic)
+  |
+  v
+[Normalizer] -- Text normalization, full-width/half-width conversion
+  |
+  v
+[Stage 1: Keyword Classifier] -- 19 categories, 500+ keyword rules
+  |  (high confidence)         |  (low confidence)
+  v                            v
+[TaxRuleEngine]        [Stage 2: Claude AI Classifier]
+  |                            |
+  v                            v
+[TaxRuleEngine]         [TaxRuleEngine]
+  |                            |
+  v                            v
+[Confidence Router] -- auto_register / auto_register_with_log / human_review
+  |
+  v
+[Memory Store] -- Persist corrections, recall patterns (Linksee Memory)
+```
+
+### TaxRuleEngine Details
+
+- **Consumption tax**: Standard 10%, reduced rate 8% (food/beverages, newspapers)
+- **Overseas detection**: SaaS platforms (AWS, Azure, GCP, etc.), advertising (Google Ads, Meta Ads, etc.) -- reverse charge applicable
+- **Withholding tax**: Professional fees, royalties, and other specified payments
+- **Asset classification**: Expense / depreciable asset / lump-sum depreciable asset / small-value asset based on acquisition cost tiers
+- **Invoice checker**: Registration number validation, transitional credit periods (2023-10 through 2029-09)
+
+## Configuration
+
+### Environment Variables
+
+| Variable | Description | Default |
+|---|---|---|
+| `COCKPIT_DATA_DIR` | Path to data directory containing keyword dictionaries and tax rules | `./data` |
+
+### Data Files
+
+The `data/` directory contains:
+
+- `keyword-dict/jp-tax-baseline-v1.json` -- Keyword dictionary for Stage 1 classification (19 categories, 500+ rules)
+- `tax-rules/jp-tax-rules-v1.json` -- Consumption tax rates, overseas SaaS/ad platforms, withholding rules, asset tiers, invoice system config
+- `exclusion-rules/jp-tax-baseline-v1.json` -- Patterns to exclude non-classifiable transactions (ATM, salary, transfers, etc.)
+
+## Development
+
+```bash
+# Run in development mode
+npm run dev
+
+# Run tests
+npm test
+
+# Build
+npm run build
+```
+
+## License
+
+MIT -- see [LICENSE](./LICENSE) for details.
+
+Copyright (c) 2026 Synapse Arrows PTE. LTD.

package/data/exclusion-rules/README.md

@@ -0,0 +1,104 @@
+# Cockpit MCP — Exclusion Rules
+
+## What this is
+
+Cockpit MCP's **Stage 0 filter** runs BEFORE the keyword/AI classifier. It identifies transactions that should NOT be auto-journalized — they get escalated to human review instead.
+
+This is the **practitioner-style 7-rule exclusion** pattern. From the reference note:
+
+> 自動仕訳で一番怖いのは「本来仕訳すべきでないものを仕訳してしまう」ことです。
+> 「何を自動化して、何を人間が見るか」の線引きを明確にしているからこそ、安全に60社を回せています。
+
+The 7 rules:
+
+1. **内容不明デビット** — "デビット+数字" だけで店名なし → 人間が利用明細確認
+2. **借入金返済** — 元本+利息 内訳必要 → freee 借入金返済 wizard 推奨
+3. **社会保険料・税金** — 法定福利費 / 預り金 / 租税公課 判断必要
+4. **給与支払い** — 給与計算結果と連動必要 → freee 人事労務 wizard
+5. **投資・資産運用** — 投資有価証券 / 雑所得 / 譲渡益 判断、 税理士判断
+6. **ATM出金・残高調整** — 口座振替別処理、 cash_balance_adjustment workflow へ
+7. **公共料金** — 事業 vs 個人按分必要 → 手動比率設定
+
+## How matching works
+
+```
+Transaction (= memo, amount, date, partner_name)
+       ↓
+Stage 0 (Exclusion Rules)
+       ├─ Rule 1 match? → human_review (rule_id, reason)
+       ├─ Rule 2 match? → human_review
+       ├─ ...
+       └─ No match → proceed to Stage 1
+                        ↓
+                  Stage 1 (Keyword Dict)
+                        ├─ Match? → return classification
+                        └─ No match → Stage 2
+                                        ↓
+                                  Stage 2 (Claude API fallback)
+                                  → return with confidence (high/medium/low)
+                                  → low confidence → human review
+```
+
+## Action types
+
+- **human_review**: Transaction goes to "確認待ち" queue. Tax accountant reviews + manually classifies.
+- **alternative_workflow**: Transaction is routed to a different workflow (e.g., ATM withdrawal → cash_balance_adjustment, NOT auto-journalized but auto-handled).
+- **skip_silently**: Transaction is ignored entirely (= for noise filtering, rare use).
+
+## Per-firm overrides
+
+Each Cockpit user (= 税理士事務所) can define overrides:
+
+```json
+{
+  "firm_id": "tanaka-tax-office",
+  "rule_overrides": [
+    {
+      "rule_id": "investment",
+      "additional_keywords": ["弊社専用ファンド名"],
+      "exclude_keywords": ["NISA"]
+    },
+    {
+      "rule_id": "salary_payment",
+      "disabled": true,
+      "reason": "うちの事務所は freee HR 連動完成済"
+    }
+  ]
+}
+```
+
+→ Stored in Cockpit Web Dashboard, applied per-instance via Cockpit MCP.
+
+## Validation
+
+```bash
+# Schema validation (TODO: add to CI)
+node scripts/validate-exclusion-rules.js data/exclusion-rules/*.json
+```
+
+## Smoke test scenarios
+
+`packages/cockpit-mcp/tests/exclusion-rules.test.ts` covers:
+
+1. "デビット 12345" → unknown_debit, human_review
+2. "公庫 利息支払い" → loan_repayment
+3. "源泉所得税納付" → social_insurance_tax
+4. "給与 山田太郎" + freee HR list 含む → salary_payment
+5. "野村證券 譲渡代金" → investment
+6. "セブン銀行ATM出金" → atm_withdrawal → cash_balance_adjustment workflow
+7. "東京水道局" → utilities (公共料金)
+8. "Suica チャージ" → no match → proceed to Stage 1 (= 旅費交通費 にいく)
+
+## Partnership note
+
+These 7 rules are the canonical practitioner playbook. Year 1 closed beta:
+- Tax practitioner reviews + signs off (= 「これが私の 7 rule で間違いない」)
+- Co-author credit in `maintainer` field
+- His extensions (= 8th, 9th rule) merged via PR
+
+## See also
+
+- Schema: [`../exclusion-rules-schema.json`](../exclusion-rules-schema.json)
+- Architecture: [`../../docs/architecture.md`](../../docs/architecture.md)
+- keyword-dict (Stage 1): [`../keyword-dict/README.md`](../keyword-dict/README.md)
+- Strategy: [synapse-arrows-playbook Doc 10](https://github.com/michielinksee/synapse-arrows-playbook/blob/main/04-tooling-gaps/10-kansei-link-cockpit-strategy.md)

package/data/exclusion-rules/jp-tax-baseline-v1.json

@@ -0,0 +1,185 @@
+{
+  "$schema": "../exclusion-rules-schema.json",
+  "version": "1.0.0",
+  "locale": "ja-JP",
+  "tax_jurisdiction": "JP",
+  "last_updated": "2026-05-09",
+  "maintainer": "Synapse Arrows PTE. LTD.",
+  "license": "MIT",
+  "description": "Cockpit MCP の Stage 0 exclusion filter。 自動仕訳すべきでない transaction を弾いて 人間 review に escalate する。 実務ベースの 7 種除外ルール。",
+  "notes": [
+    "execution: classify_transaction の前に check_exclusion を必ず call",
+    "match → return { excluded: true, rule_id, reason, action: 'human_review' }",
+    "no match → classify_transaction (Stage 1+2) へ進む",
+    "rule priority: 上から順 (= 同 transaction が複数 rule match なら 上のが採用)",
+    "extension: 各事務所が override / 追加可能 (= per-firm exclusion overrides 機構)"
+  ],
+  "rules": [
+    {
+      "id": "unknown_debit",
+      "name_ja": "内容不明デビット",
+      "name_en": "Unknown debit (no merchant name)",
+      "description": "「デビット+数字」だけで店名がない transaction。 詳細不明のため自動仕訳せず人間 review。",
+      "priority": 1,
+      "match": {
+        "type": "regex",
+        "field": "memo",
+        "pattern": "^デビット[\\s\\d\\-]*$",
+        "flags": "i"
+      },
+      "alternative_patterns": [
+        "デビット利用",
+        "DEBIT",
+        "VISA DEBIT 振替"
+      ],
+      "action": {
+        "type": "human_review",
+        "reason_template": "店名不明のデビット取引。 通帳 or 利用明細から店名特定後 人間判断 required。",
+        "suggested_next_step": "freee 上で取引詳細を開き、 利用明細から merchant 確定 → 適切な category へ分類"
+      }
+    },
+    {
+      "id": "loan_repayment",
+      "name_ja": "借入金返済",
+      "name_en": "Loan repayment",
+      "description": "公庫・銀行借入 / ローン返済。 元本+利息 内訳 が必要なため自動仕訳せず。",
+      "priority": 2,
+      "match": {
+        "type": "any_keyword",
+        "field": "memo",
+        "keywords": [
+          "公庫", "日本政策金融公庫", "JFC",
+          "融資", "借入", "借入金返済", "返済",
+          "ローン", "Loan", "LOAN",
+          "信金", "信用金庫",
+          "商工中金",
+          "プロパー融資"
+        ]
+      },
+      "action": {
+        "type": "human_review",
+        "reason_template": "借入金返済または融資関連取引。 元本 / 利息 内訳必要、 freee の 借入金返済 wizard 推奨。",
+        "suggested_next_step": "freee → 取引 → 借入金返済 wizard で元本・利息分離"
+      }
+    },
+    {
+      "id": "social_insurance_tax",
+      "name_ja": "社会保険料・税金",
+      "name_en": "Social insurance & taxes",
+      "description": "年金 / 健保 / 源泉所得税 / 住民税 / 法人税 等の納付。 仕訳科目が複雑なため自動 NG。",
+      "priority": 3,
+      "match": {
+        "type": "any_keyword",
+        "field": "memo",
+        "keywords": [
+          "年金", "厚生年金", "国民年金",
+          "健保", "健康保険", "協会けんぽ", "けんぽ",
+          "源泉", "源泉税", "源泉所得税",
+          "住民税", "市民税", "県民税", "都民税",
+          "法人税", "事業税", "消費税納付",
+          "労災", "雇用保険",
+          "国税", "地方税",
+          "印紙税"
+        ]
+      },
+      "action": {
+        "type": "human_review",
+        "reason_template": "社保・税金納付。 法定福利費 / 預り金 / 租税公課 等の科目選択判断必要。",
+        "suggested_next_step": "freee → 取引 → 社会保険料 / 税金 wizard で適切科目選択"
+      }
+    },
+    {
+      "id": "salary_payment",
+      "name_ja": "給与支払い",
+      "name_en": "Salary payment",
+      "description": "従業員への給与振込。 freee 人事労務 API から従業員名 list を取得して match。 自動仕訳禁止 (= 給与計算と連動の必要)。",
+      "priority": 4,
+      "match": {
+        "type": "any_keyword_or_pattern",
+        "field": "memo",
+        "keywords": [
+          "給与", "給料", "ボーナス", "賞与"
+        ],
+        "patterns": [
+          {
+            "type": "transfer_to_employee",
+            "description": "振込 + freee 人事労務 API の従業員 list 内の名前 match",
+            "requires_external_data": "freee_hr_employee_list"
+          }
+        ]
+      },
+      "action": {
+        "type": "human_review",
+        "reason_template": "給与振込検出。 給与計算結果と整合確認後、 freee 給与仕訳 wizard で起票。",
+        "suggested_next_step": "freee 人事労務 → 給与計算 → 仕訳作成 で自動連動"
+      }
+    },
+    {
+      "id": "investment",
+      "name_ja": "投資・資産運用",
+      "name_en": "Investment / securities",
+      "description": "証券口座への入出金、 投資信託 等。 投資勘定 / 雑所得 / 譲渡益 等の判断必要。",
+      "priority": 5,
+      "match": {
+        "type": "any_keyword",
+        "field": "memo",
+        "keywords": [
+          "証券", "野村", "大和", "SMBC日興", "みずほ証券", "三菱UFJモルガン",
+          "SBI証券", "楽天証券", "マネックス", "auカブコム", "松井証券",
+          "投資信託", "ファンド", "FUND",
+          "株式", "株",
+          "FX", "暗号資産", "Bitflyer", "Coincheck", "GMOコイン",
+          "REIT", "ETF",
+          "NISA", "iDeCo", "つみたて"
+        ]
+      },
+      "action": {
+        "type": "human_review",
+        "reason_template": "投資・資産運用関連取引。 仕訳科目 (= 投資有価証券 / 雑所得 / 譲渡益等) は税理士判断。",
+        "suggested_next_step": "個別 review、 必要なら年度末まで仮勘定で保持"
+      }
+    },
+    {
+      "id": "atm_withdrawal",
+      "name_ja": "ATM出金・残高調整",
+      "name_en": "ATM withdrawal / balance adjustment",
+      "description": "ATM 出金は口座振替で別処理。 仕訳起票せず、 別途 balance adjustment workflow へ。",
+      "priority": 6,
+      "match": {
+        "type": "any_keyword",
+        "field": "memo",
+        "keywords": [
+          "ATM", "ATM出金", "現金引出", "引出", "引き出し",
+          "ゆうちょATM", "コンビニATM",
+          "セブン銀行ATM", "イオン銀行ATM"
+        ]
+      },
+      "action": {
+        "type": "alternative_workflow",
+        "alternative_workflow_id": "cash_balance_adjustment",
+        "reason_template": "ATM 出金 / 現金引出。 口座振替として別処理、 現金勘定残高調整に流す。",
+        "suggested_next_step": "freee → 振替伝票 → 普通預金 → 現金"
+      }
+    },
+    {
+      "id": "utilities",
+      "name_ja": "公共料金",
+      "name_en": "Public utilities (no matching account category)",
+      "description": "個人宅の電気・ガス・水道。 事業所契約と区別困難なため、 自動 NG。 (事業所契約の utilities は keyword-dict utilities category で別途処理)",
+      "priority": 7,
+      "match": {
+        "type": "any_keyword",
+        "field": "memo",
+        "keywords": [
+          "東京水道局", "東京都水道", "水道局", "上下水道",
+          "NHK受信料", "NHK"
+        ]
+      },
+      "action": {
+        "type": "human_review",
+        "reason_template": "公共料金検出。 事業利用 vs 個人利用の按分判断必要。",
+        "suggested_next_step": "事業利用比率 (= 例: 自宅事務所なら 30%) を手動設定後仕訳"
+      }
+    }
+  ]
+}

package/data/exclusion-rules-schema.json

@@ -0,0 +1,109 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "$id": "https://github.com/michielinksee/kansei-link-cockpit/blob/main/data/exclusion-rules-schema.json",
+  "title": "Cockpit MCP Exclusion Rules",
+  "description": "Schema for Stage 0 exclusion filters used by Cockpit MCP. These run BEFORE classify_transaction. Matching transactions are escalated to human review (= NOT auto-journalized).",
+  "type": "object",
+  "required": ["version", "locale", "tax_jurisdiction", "rules"],
+  "properties": {
+    "version": { "type": "string", "pattern": "^\\d+\\.\\d+\\.\\d+$" },
+    "locale": { "type": "string", "pattern": "^[a-z]{2}-[A-Z]{2}$" },
+    "tax_jurisdiction": { "type": "string", "pattern": "^[A-Z]{2}$" },
+    "last_updated": { "type": "string", "format": "date" },
+    "maintainer": { "type": "string" },
+    "license": { "type": "string" },
+    "description": { "type": "string" },
+    "notes": {
+      "type": "array",
+      "items": { "type": "string" }
+    },
+    "rules": {
+      "type": "array",
+      "minItems": 1,
+      "items": {
+        "type": "object",
+        "required": ["id", "name_ja", "description", "priority", "match", "action"],
+        "properties": {
+          "id": {
+            "type": "string",
+            "pattern": "^[a-z_]+$"
+          },
+          "name_ja": { "type": "string" },
+          "name_en": { "type": "string" },
+          "description": { "type": "string" },
+          "priority": {
+            "type": "integer",
+            "description": "Lower = higher priority. Rules evaluated top to bottom; first match wins."
+          },
+          "match": {
+            "type": "object",
+            "required": ["type", "field"],
+            "properties": {
+              "type": {
+                "type": "string",
+                "enum": ["regex", "any_keyword", "any_keyword_or_pattern", "exact"]
+              },
+              "field": {
+                "type": "string",
+                "enum": ["memo", "amount", "partner_name", "date"],
+                "description": "Which transaction field to match against (after normalization)"
+              },
+              "pattern": {
+                "type": "string",
+                "description": "Regex pattern (when type=regex)"
+              },
+              "flags": {
+                "type": "string",
+                "description": "Regex flags like 'i', 'g' (when type=regex)"
+              },
+              "keywords": {
+                "type": "array",
+                "items": { "type": "string" },
+                "description": "Substring list (when type=any_keyword or any_keyword_or_pattern)"
+              },
+              "patterns": {
+                "type": "array",
+                "items": {
+                  "type": "object",
+                  "properties": {
+                    "type": { "type": "string" },
+                    "description": { "type": "string" },
+                    "requires_external_data": { "type": "string" }
+                  }
+                },
+                "description": "Custom pattern detectors (e.g., transfer_to_employee requires freee_hr_employee_list)"
+              }
+            }
+          },
+          "alternative_patterns": {
+            "type": "array",
+            "items": { "type": "string" },
+            "description": "Additional substring fallbacks beyond the primary match"
+          },
+          "action": {
+            "type": "object",
+            "required": ["type", "reason_template"],
+            "properties": {
+              "type": {
+                "type": "string",
+                "enum": ["human_review", "alternative_workflow", "skip_silently"]
+              },
+              "reason_template": {
+                "type": "string",
+                "description": "Human-readable reason shown to the tax accountant"
+              },
+              "suggested_next_step": {
+                "type": "string",
+                "description": "Concrete recommendation for the human reviewer"
+              },
+              "alternative_workflow_id": {
+                "type": "string",
+                "description": "If type=alternative_workflow, ID of the workflow to invoke (e.g., cash_balance_adjustment)"
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+}

package/data/keyword-dict/README.md

@@ -0,0 +1,91 @@
+# Cockpit MCP — Keyword Dictionary
+
+## What this is
+
+Cockpit MCP's **Stage 1 classifier** uses keyword dictionaries to map raw transaction 摘要 (= memo strings) to 勘定科目 (= account categories) + 税区分 (= tax codes).
+
+This is the **practitioner-style 14 categories × 100 keywords** approach, packaged as portable JSON for free distribution. Each tax jurisdiction has its own dictionary file:
+
+- `jp-tax-baseline-v1.json` — Japanese tax accounting baseline (14 categories)
+- (future) `sg-tax-baseline-v1.json` — Singapore GST + ACRA-aligned categories
+- (future) `us-tax-baseline-v1.json` — US sales tax + GAAP categories
+
+## How matching works
+
+1. **Normalize** 摘要 string: 全角 → 半角、 大文字 → 小文字、 trim whitespace
+2. **Iterate categories** (top to bottom) and for each, check if any keyword is a substring of the normalized 摘要
+3. **First match wins**: if "Suica" matches in `travel`, classifier returns `travel`. No further categories checked.
+4. **Apply amount thresholds** (if defined): e.g., 会議費 only matches ≤¥10K; if amount > ¥10K, auto-redirect to 交際費 via `amount_overflow_category`
+5. **Apply special patterns** (if defined): e.g., `transfer_professional` requires the keyword AND a 振込 verb in the 摘要
+6. If no category matches → fallback to **Stage 2** (Claude API classifier with confidence high/medium/low)
+
+## Versioning
+
+- `v1.0.0` = initial baseline (= 14 categories × ~50 keywords average)
+- `v1.x` = expand to 100 keywords/category via dogfood iterations
+- `v2.x` = community-tuned (= Cockpit user 集合知 で keyword 改善)
+- `v3.x` = AI-discovered keywords (= 過去 anonymous data から AI が新 keyword 発見)
+
+## How to extend
+
+### Add new keyword to existing category
+
+1. Open the relevant `*-tax-baseline-v*.json`
+2. Find the category by `id`
+3. Append your keyword to the `keywords` array
+4. Test: run smoke test against your test fixtures
+5. Open PR (= reviewer must verify keyword is unambiguous + likely to match real transactions)
+
+### Add new category
+
+1. Define new entry in `categories` array with all required fields (= `id`, `name_ja`, `freee_account_code`, `default_tax_code`, `keywords`)
+2. Choose `freee_account_code` from freee's standard chart of accounts
+3. Determine `default_tax_code` based on Japanese 消費税 rules
+4. Decide priority (= where in the categories array it goes — earlier = higher priority)
+5. Add at least 10 initial keywords to bootstrap
+6. Run smoke test
+7. Open PR (= reviewer verifies category is genuinely needed and not a duplicate)
+
+### Add jurisdiction-specific dictionary
+
+1. Copy `jp-tax-baseline-v1.json` → `<country-code>-tax-baseline-v1.json`
+2. Update `locale` and `tax_jurisdiction` fields
+3. Replace categories with the target country's chart of accounts
+4. Provide a community maintainer (= regional accountant) to validate
+5. Open PR
+
+## Partnership note
+
+This dictionary is designed to be **co-authored with tax practitioners** starting Year 1 closed beta:
+
+- Practitioner's 6-month-built 14 categories × 100 keywords → contributed as anonymous PR
+- Co-author credit in `maintainer` field
+- Quarterly merge of practitioner improvements + community improvements
+- See `synapse-arrows-playbook/04-tooling-gaps/10-kansei-link-cockpit-strategy.md` §11 for partnership terms
+
+## Validation
+
+```bash
+# Run schema validation (TODO: add to CI)
+node scripts/validate-keyword-dict.js data/keyword-dict/*.json
+```
+
+## Smoke test scenarios
+
+`packages/cockpit-mcp/tests/keyword-dict.test.ts` covers:
+
+1. Basic substring match (= "Suica" → travel)
+2. Multi-keyword match (= 上 category 優先)
+3. Amount threshold (= "スターバックス @ ¥15,000" → entertainment, NOT meeting_meal)
+4. Special pattern: salary employee detection (= 給与 + 従業員名 list match)
+5. Special pattern: 振込 + 士業名 → professional_fee + 取引先抽出
+6. Special pattern: 振込 + カナ人名 → outsourcing + 発生日前月末調整
+7. No-match → Stage 2 fallback signal
+8. Normalize: 全角 / カタカナ / 半角混在の 摘要
+
+## See also
+
+- Schema: [`../keyword-dict-schema.json`](../keyword-dict-schema.json)
+- Architecture: [`../../docs/architecture.md`](../../docs/architecture.md)
+- 7 exclusion rules: [`../exclusion-rules/README.md`](../exclusion-rules/README.md) (= Stage 0 = before classifier runs)
+- Strategy: [synapse-arrows-playbook Doc 10](https://github.com/michielinksee/synapse-arrows-playbook/blob/main/04-tooling-gaps/10-kansei-link-cockpit-strategy.md)