secwatch / observer

About this site

I built secwatch.observer because I wanted to read 8-K filings the way I read the news, not the way the SEC's website asks me to. EDGAR is the source of truth, and the source-of-truth UI shows it: a 1990s table of accession numbers, no summaries, no event taxonomy, no real-time feed unless you scrape it yourself.

So I scraped it myself, plugged the body and any 99.x exhibits through a fast LLM with a strict-JSON prompt, and pushed the results to a browser over Server-Sent Events. Every filing gets a one-line headline, three bullets, an event-type tag (earnings, M&A, leadership, debt, dividend, litigation, cyber, regulatory), and a confidence rating. The source document is one click away on EDGAR — the summary is a reading aid, not a replacement.

What you see

What you don't see

Who's behind it

One person, in Python, on a single Hetzner box in Helsinki. No team, no funding, no commercial product (yet). If you want to ask a question, file an issue, or suggest a missing event taxonomy term, the source is open: github.com/rotoole1230/edgar-streaming.

For an under-the-hood look at the pipeline, see methodology.