Suzuri

Semantic search over Japan's official government white papers: the Ministry of Defense white paper (防衛白書) in both its 2025 edition (令和7年版) and 2026 edition (令和8年版), and METI's trade white paper (通商白書, 2025 edition). Use this when the user asks about Japanese defense policy, the Self-Defense Forces, Japan's security environment, defense budgets or procurement programs, the Japan-US alliance, how Tokyo officially describes China, North Korea or Russia, or about Japan's trade strategy, global supply chain resilience, economic security, overdependence and economic coercion, or China's industrial policy and its effect on trade. One query searches all of them at once, and every result is labelled with its paper name and edition year. Because both defense editions are indexed together, a single query can surface how the same topic is described in each year: the 2026 edition adds parts and chapters that did not exist before, while other passages carry over from the previous year with little or no change. Reading the returned excerpts side by side therefore shows both what changed and what stayed the same. This tool retrieves passages; it does not compute the differences for you. Queries may be in English or Japanese — English queries are automatically translated before retrieval against the Japanese corpus. Returns ranked excerpts with paper name, chapter, page numbers and a source URL suitable for citation. Scope: the 2025 and 2026 editions of the defense white paper and the 2025 edition of the trade white paper only. Not a source for breaking news, press conferences, statistical databases, export control regulations, tariff schedules, or documents from other ministries.

{'type': 'object', 'required': ['query'], 'properties': {'query': {'type': 'string', 'description': 'A natural-language question or topic, in English or Japanese. Examples: "How does Japan describe China\'s military buildup?", "GCAP next-generation fighter", "経済安全保障 サプライチェーン".'}, 'top_k': {'type': 'number', 'description': 'Number of results to return (default 10, max 20)'}, 'min_score': {'type': 'number', 'description': 'Minimum cosine similarity threshold (default 0.25, range 0–1)'}}}