Walkthroughs, recipe deep-dives, and the occasional opinion on what SEC filing intelligence looks like when the agent is the user.
A 25-NSE means a stock is leaving an exchange — but most of the time nothing bad happened. The one character in the filing that tells you whether it's a routine bond maturity or a company in trouble.
"MCP vs web search" is two different questions wearing one phrase. Web-search MCP servers vs built-in search is a preference call. Domain MCP tools vs searching the web at all is a measured one: 10–21× fewer tokens, and correctness flips.
Every "MCP token efficiency" number you've read measures protocol overhead — schemas, tool definitions, the input-side tax. Real, and worth dieting. We benchmarked the other side: domain MCP tools instead of web search. 10–21× fewer tokens, 10–21× cheaper, and correctness flipped.
Luminar's Oct 31 8-K said cash wouldn't survive Q1 and a covenant breach loomed. The Chapter 11 came 45 days later. The full filing trail, with receipts.
An AI session split e.l.f.'s 25% revenue growth into $293.5M acquired vs $29.5M organic, every number cited to an SEC accession. The workflow, timed and costed.
A fake company, a fabricated 8-K, a consensus ask, a stale 13F, and a fake-bullish insider tape — five traps run against the live toolkit, outputs unedited.
What a 10-K, 8-K, 13D, Form 4, S-1, and 13F actually are — who files them, how often, and what you'd read each one for. A working map of the EDGAR forms that move.
When a 13D vs a 13G gets filed, the activist-intent test that decides which, and the shortened 2024 filing deadlines every investor-relations and event-driven desk now works to.
A complete reference to all 8-K item codes — what each one discloses, which ones carry real signal, and how to filter the ~30 triggers down to the events that matter.
edgar.tools the platform stands on edgartools the open-source library. Here's the lineage in three numbers — and what it means for an analyst, a quant, or an AI team building on SEC data.