Data Methodology

We use public statistical releases, regulatory filings, published market data, vendor reports, and our own structured compilations. Each article lists its sources. Where sources conflict, we present indexed trends rather than false precision, and we say so.

Python jobs (pandas-based) regenerate each article’s chart and table artifacts on a recurring schedule — typically every few days. A refresh can update headlines, summaries, percent changes, charts, tables, narrative framing, and the freshness timestamp. Every refresh appends to the article’s “what changed” log. Article URLs never change when data refreshes.

Generated artifacts are validated before publication; a refresh that fails validation is discarded and the previous good data stays live. The frontend additionally falls back to the article’s last embedded dataset if an artifact is missing or corrupt. A broken pipeline can delay freshness — it cannot blank a chart.

The starter articles currently published ship with illustrative mock data generated by this same pipeline, clearly noted in each methodology section. Live data connections replace mock generators at launch without changing the article structure or URLs.

Estimates are labeled as estimates. Survey data carries survey bias. Small samples are called small. Our Editorial Policy governs how a weakened thesis gets updated rather than defended.