Blogclose
What to use when: the answer is only worth what you can say about where it came from
Seven posts, one point. What each direction taught in a line, how to pick by the question rather than the product, what this costs and where it will annoy you, stated plainly enough to decide with.
Seven posts, one point. If you take nothing else: the answer is only worth what you can say about where it came from.
I did not start from that principle, I arrived at it. Every direction in this series exists because I needed a number, got a confident one, and had no way to check it. The fix was never a better model. It was a path to something published, and a habit of keeping the source next to the conclusion.
What each one taught, in one line
Finance. A ten percent fall is the same number whether it is panic or profit-taking. What separates them is who holds the name and what they paid.
Macro. A stale price is obviously stale. A stale forecast looks exactly like a fresh one, and cites a real institution while it does it.
Entity. You are not screening a company, you are screening a string, until you have an identifier.
NetAtlas. Registration, announcement, authorisation and inference are four different claims about one address. When they disagree, that is information, not an error.
News. A price says what happened, an official series says in what regime, neither says why, and physical movement is what decides whether the offered why is real.
Earth. Both ends of a route can be clear while the middle is severe. Checking the endpoints returns a confident green.
Research. An identifier either resolves or it does not. Everything else is a story about a paper.
Pick by the question, not by the product
| The question in front of you | Start here |
|---|---|
| What did the market do, and who is exposed | Finance |
| What did the official record say, and has it been revised | Macro |
| Can we onboard this company | Entity |
| Who is this address and should we trust the route | NetAtlas |
| What is being said, how loudly, who noticed | News |
| Can we run the plan today | Earth |
| Does this paper exist and does it say that | Research |
Composition is yours. We return each leg with its own source and its own clock, and we deliberately do not sell you a single number that melts them together, because there is no honest arithmetic between a monthly index, an hourly theme share and a daily transit count. If your product needs one number, you write that function and you own the formula. That is not us avoiding work. It is the only version where you can defend the output.
Why this costs money, stated plainly
I would rather you decide with the trade-offs in front of you than be persuaded.
What you would otherwise assemble yourself. A quote feed, a filings parser, a sanctions list ingester, a routing table, a media stream, a weather source, a scholarly index. Each with its own auth, schema, rate limit and failure mode, and every one of them a permanent job rather than a project, because the day one silently stops being correct is the day it matters. None of that is your product.
What we actually built rather than resold. Fundamentals reconstructed from the filing record, 503 reported concepts for one company, each arriving with its taxonomy definition. Cross-company ranking across thousands of filers on any concept. Derived quantities with the equation in the response. Screening against sovereign registers with the entry id and publication date attached. An atlas that tells you which evidence layer an answer rests on.
What it costs. Free is fifty calls a day and every direction is in it. Dev is twenty-five dollars for five thousand a day, Pro fifty-nine for fifty thousand, and beyond that we talk. Plans differ by volume only. I will not gate a direction, because that would mean deciding which questions you are allowed to ask.
Where it will annoy you. Coverage is uneven, and it is thinner outside the largest economies. Some sources publish with a lag that no amount of engineering removes. Our limits are daily, so you cannot spend a month of quota in an afternoon, which is a real constraint and occasionally the wrong one for a backfill. And some of what we return is deliberately less confident than a competitor’s equivalent, because we would rather hand you the uncertainty than hide it.
If that last sentence reads as a weakness, this is probably not the right platform for you, and I would rather you knew now.
What I actually want you to take away
Not that we have a lot of endpoints. There are plenty of places to buy data.
The thing worth carrying is the habit. Ask a source. Keep the timestamp. Store the identifier, not the summary. When you compute something, publish the formula. When you do not know, return that instead of a plausible number.
Do that with our API or without it. I built this because I wanted it to exist and could not buy it, and the posts above are the argument, not the brochure.
Where to go now
Try it. A key takes a minute at app.sugra.ai. The free plan is enough to check whether any of this holds.
Read the reference. Full OpenAPI and guides at docs.sugra.ai.
Run something real. The cookbook has working recipes rather than snippets.
Wire an agent to it. The agent-native surface is documented at sugra.systems/api#mcp, and the connector installs from PyPI as sugra-api-mcp. That is the shortest path from reading this to having a model that stops guessing.
Start at Platform intro if you came in from the middle. Thank you for reading this far.