Build vs buy

You could build this. We did. Here is the honest bill.

If your research agent reads filings, you are already paying for a pipeline like this — in tokens, in scrapers, and in your own weekends. This page itemizes the bill so you can check our work. No fear-mongering; every engineer on this page is treated like one.

01 · The plumbing

Seven markets, seven scrapers you now own

Source	What it carries	What fights back
US — EDGAR	8-Ks, 13Ds, tender offers, merger proxies, Form 15s, Form 4s — 28 form types	Rate limits that 403 you during the 4:51pm filing rush, exactly when it matters
Japan — TDnet	Every disclosure, summarized in English	A 20-page daily list cap, and everything underneath is a Japanese PDF
UK — RNS + Takeover Panel	Special-sits RNS, Rule 9, Part 26A plans, live bid states with Rule 2.6 deadlines	Two separate sources that disagree about what a “state change” is
Germany — EQS	MAR ad hoc, directors’ dealings, voting rights, takeover news	The friendly one. Enjoy it; it is not representative
Sweden — MFN + FI	Regulatory announcements plus the insider register	Insider buys arrive as CSV from a regulator with its own encoding opinions
Switzerland — SIX + Takeover Board	Official notices, buyback lines, live offers	The notices endpoint hides in a versioned config bundle that moves without notice
Australia — ASX	Substantial holders, takeovers, director dealings, buybacks, DOCA recaps	Fronted by a rotating token you re-harvest when it changes. It changes

Call it two to five days per source if nothing fights back. Something always fights back. A realistic first build is four to eight weeks — and then it starts breaking.

02 · The token bill

The LLM meter runs whether you build or buy

Filings analyzed per month, across seven markets	~25,000
Average filing, as model input	8k–30k tokens
Cheapest capable model (Haiku-class, $1/M in, $5/M out)	≈ $450 / month
A model you would actually trust for nuance (Sonnet-class)	≈ $1,300 / month

The arithmetic is checkable: ~25,000 filings at an average ~15k input tokens is ~375M tokens a month before a single retry, re-analysis, or translation pass. Analysis with a budget model costs more than our entire subscription — and that is the meter alone, running on plumbing you still had to build.

03 · The upkeep

Last month, on our side of the fence

EDGAR started rate-limiting during the after-close filing rush. Our circuit breaker opened, retried, and recovered every filing. Your cron job would have silently missed the most interesting hour of the day.
A market data endpoint moved inside a versioned frontend bundle. Finding the new path took a debugging session, not a config change.
A frontend API token rotated without notice. Ours re-harvests automatically; a hand-rolled scraper returns empty lists and no error.

For us these are maintenance commits. For a one-person pipeline they are Saturdays — and the failure mode is never a crash, it is silence that looks like a quiet news day.

04 · The bottom line

Build

$450–$1,300 /mo in tokens

plus four to eight weeks of build, plus the scrapers, plus every Saturday one of them breaks quietly.

Buy

$399 /mo flat

analyzed minutes after publication, seven markets, one cursor, and the breakage is our problem. 7-day trial.

Start the trial

or hand it straight to your agent: npx -y @situationatlas/mcp

Not ready for a card? Sign up and mint a free sandbox key — every endpoint, filings delayed 7 days. Enough to check our work end to end.