How it works

How the data gets from the internet to a forecast and onto this page. Everything heavy runs on one laptop in Nepal. This site is static files: your browser downloads a small Parquet snapshot and runs SQL on it, so every chart is a query you can repeat in the SQL explorer.

1. The whole system

• Laptop does everything heavy. It sits in Nepal because some government sites block foreign IPs.
• The website has no server. The browser downloads the Parquet snapshot and runs SQL on it with DuckDB-WASM. Every chart is one of those queries.

2. What runs when

A failing step is logged and skipped; the other steps still run. The exact cron lines are in the repository's docs/flow.md.

3. One ingest step, in detail

ok is written only after the table write commits, so a crash mid-run is simply redone next time.

4. How a forecast is generated

Model details and why each was chosen: the Models and Method & limits pages, and docs/models/ in the repository.