Open data — Taylor, NFP, and the research archive

Saturday JSONL

Every Saturday ~5 AM ET a full snapshot of published notes is written as JSONL. No login. robomacro.ai/fine_tune_data/latest/

• documents.jsonl — one object per document: id, type, date, title, text, source_url, origin, tokens_estimate
• chunks.jsonl — section chunks (~300–800 tokens) with doc_id
• stats.json — run_date, counts, by_newsletter_type

Dated snapshots: https://robomacro.ai/fine_tune_data/snapshots/YYYY-MM-DD/. latest/ always points at the newest Saturday.

How to cite

RoboMacro Research Archive (JSONL), snapshot YYYY-MM-DD, https://robomacro.ai/fine_tune_data/snapshots/YYYY-MM-DD/. Notes remain copyright RoboMacro; the dump is for research, RAG, and fine-tuning with attribution.

Recipes (human pages)

• Hawk/dove scoring
• G10 Taylor Rule — formula in HTML, FRED sources, HP-filter vs CBO gap
• Payrolls Lab first-print scoreboard

ChatGPT, Grok, and Cursor

Official Macro is a free MCP for ChatGPT, Grok, and Cursor. It serves official prints plus the public desks on this site (hawk/dove, Taylor, inflation, calendar, elections, payrolls, oil, daily-note titles, finance jobs). It is not the paid x402 gateway at /mcp.

• MCP: https://mcp.robomacro.com/mcp
• Health: mcp.robomacro.com/health
• Jobs board: /jobs

In ChatGPT: Settings → Apps & Connectors → Developer Mode → add that MCP URL, No Auth. In Grok: grok.com/connectors → Custom → same URL.

Machine APIs

• Taylor: GET /api/v1/taylor/data
• Central banks: GET /api/v1/central-banks/dashboard
• Payrolls cache: /payrolls_forecasting_cache/
• /llms.txt and /llms-full.txt

GitHub-facing write-up of the same recipes lives in docs/methodology/README.md in the RoboMacro repo.