Jp Grants

jp-grants - Japan Subsidies / Grants (補助金・助成金)

Operate with a "freshness first" mindset: deadlines and rules change frequently. Prefer official sources (e.g. *.go.jp, *.lg.jp, jgrants-portal.go.jp). Always provide URLs and, when possible, short supporting quotes. When available, also check past award/adoption results (採択結果/採択者一覧/交付決定) from official public sources only to judge fit.

Scope

• This skill is for reusable research workflow, not for maintaining a custom script per subsidy program.
• Prefer general discovery/extraction patterns that work across ministries, prefectures, and municipalities.
• Add program-specific handling only when it reflects a common official pattern that is likely to recur.

Non-goals

• Do not build or maintain one-off scrapers for each individual subsidy program.
• Do not treat consultant sites, rewritten summaries, or private databases as authoritative sources.
• Do not infer individual accepted cases from summary statistics when the official source does not publish them.

Operating rules

• Avoid hallucinating program details. If a field is unknown, say so and provide the official URL to verify.
• For past award/adoption examples, use only official public sources. Do not rely on consultant summaries, media rewrites, or private databases.
• Distinguish clearly between:
  • Subsidies/grants (補助金)
  • Employment-related support / allowances (助成金, often MHLW)
  • Loans/financing (融資) and tax incentives (税制) (not grants)
• Do not provide legal/tax advice; provide procedural guidance and point to official documents.

Minimal intake (ask only if needed)

If the user question is underspecified, ask for only the 3-6 items that most affect eligibility:

1. Applicant type: individual / sole proprietor / SME corporation / NPO / school / other
2. Location: prefecture + city/ward (local programs are common)
3. Purpose: equipment (capex) / IT & DX / hiring & wages / R&D / export / energy saving / disaster recovery / other
4. Company size: employees, capital, sales (rough ranges are fine)
5. Timeline: when they want to apply / start the project
6. Industry constraints: manufacturing / retail / services / construction / healthcare / etc.

If the user refuses details, proceed with generic national programs and give a checklist to narrow down.

Workflow: answer a user question

1. Restate constraints + assumptions (explicitly).
2. Collect candidate programs from official sources.
3. Choose the branch that matches the user need:
   • current program details -> find current official call pages and extract rules
   • award-result statistics -> find official result announcements and extract counts/breakdowns
   • individual accepted cases -> find official 採択者一覧, 交付決定一覧, or case-study pages/PDFs
4. Extract structured details for the chosen branch.
5. Shortlist 3-7 best fits; explain fit and disqualifiers using both current rules and past examples when available.
6. Provide next steps: what to read, what documents to prepare, and where to apply.

Source strategy

• Primary: official government and local government sites
  • *.go.jp (METI, MHLW, SME Agency, etc.)
  • *.lg.jp (prefecture/municipality)
  • jgrants-portal.go.jp (J-Grants)
• Also look for official result pages/PDFs using terms such as 採択結果, 採択者一覧, 採択事例, 交付決定.
• For award-result checks, prioritize chusho.meti.go.jp, meti.go.jp, and *.lg.jp before J-Grants.
• Secondary: public support organizations (confirm against primary)
  • J-Net21, chambers of commerce, SME support centers
• Avoid: consultant blogs as an authority (may be outdated).
• For past award/adoption checks, do not use secondary or executing secretariat sites unless they are themselves the official public publisher for that result document.
• Note: jgrants-portal.go.jp/robots.txt currently disallows almost all crawling (Disallow: /, Allow: /index.html), so search/discovery may be incomplete there.

For a curated list of sources and query templates, read jp-grants/references/sources.md.

Scripts (optional but recommended)

These scripts use Firecrawl for web search + structured extraction, plus a direct official fallback for award-result discovery.

• Find candidate pages: jp-grants/scripts/find_candidates.py
• Find + extract official award-result pages in one step: jp-grants/scripts/extract_official_award_results.py
• Extract official award-result fields: jp-grants/scripts/extract_award_results.py
• Extract individual official award cases: jp-grants/scripts/extract_case_examples.py
• Extract program fields: jp-grants/scripts/extract_programs.py

find_candidates.py flags:

• --include-local: include local government sites (site:lg.jp)
• --include-executing-sites: include common executing secretariat sites (discovery only)
• --include-award-results: also search for past award/adoption result pages and PDFs

Practical decision tree

• Need current eligibility/deadlines/required docs -> use find_candidates.py, then extract_programs.py
• Need official result counts/breakdowns -> use extract_official_award_results.py
• Need individual accepted organizations/projects -> use official public 採択者一覧 / 交付決定一覧 / case-study pages, then extract_case_examples.py
• If official sources publish only statistics, state that individual cases are not publicly available in the checked source

Output pattern (what to return to the user)

Provide a short list of candidates with consistent fields:

• Program name + administering body
• What it funds (eligible costs)
• Amount/rate (if specified)
• Key eligibility constraints
• Application window / deadline (with "as-of" date)
• Official URL (required)
• Past award/adoption examples from official public sources (if found) and what they imply about fit
• Why it fits / why it might not

Keep the list short. If there are many, offer a second pass after the user answers 1-2 narrowing questions.