Sugra API: the question that could not be bought

The question has been the same since 2017: what is happening with this company’s stock, what is being written about it, and is what is being written true.

The first two parts were purchasable. Somebody sold the price. Somebody else sold the headlines. The third part could not be bought at any price, and nobody sold it, because there was nothing to sell: checking a claim needs a source that could not have produced it. A news service cannot confirm itself.

In 2021 the question was reborn. Not because I got smarter, but because tools appeared that answered it instantly and confidently and were usually wrong. That changed the problem. Before, the difficulty was getting an answer. Now the difficulty was that an answer arrived every time, in the right grammar, with a number in it, and nothing about it indicated whether it had come from anywhere.

In 2023 I built the first prototype. Not to get a price. To give a model a path to something published instead of to its own memory.

The third part is where the work is

It took me a while to see that the three parts of my question are not three sizes of the same task.

Price is an observation. Somebody recorded a trade and the record exists. Getting it is plumbing, and plumbing is a solved problem sold by many companies.

Coverage is also an observation, of a different kind. Somebody published a sentence. That the sentence exists is a fact. That the sentence is correct is not.

Verification is not an observation at all. It is a comparison against something that could not have been produced by the thing you are checking. That distinction is the entire platform, and it is why this is not a data reseller with a nice envelope.

Here is what that looks like when it works. Oil rises twenty-seven percent over two weeks and every news service explains it with tension somewhere. That sentence is compatible with two different worlds: a market pricing cargo that genuinely stopped moving, and a market pricing a fear that has not reached the water. Both produce the same headline, and reading more headlines will never separate them.

Counting ships will. Tanker tonnage through the strait fell roughly fiftyfold across that same window and stayed down. The story was real, and now you know it rather than believe it. Had the transit held near its normal band, you would have known something else entirely: that you were looking at a risk premium which unwinds on a headline, because nothing physical has to be repaired for it to disappear. Same news, opposite trade, and only the third source tells you which one you are in.

Why one question turned into seven

The original three parts became market data, media, and a way to check one against the other.

The rest arrived because the same question kept being asked about different objects. Not what is happening with a company’s stock, but with a company we were about to sign. With this address that just hit our edge. With this site, before I send people out there this morning. With this paper somebody just cited at me.

Every one of those has the same structure. Something happened, somebody says why, and the saying and the being are different things. And every one of them fails the same way without a source: you get a fluent answer that is indistinguishable in form from a correct one.

That is the shape worth carrying into the rest of this series. A crashed service is obvious. A stale price is obvious once you look at the clock. But a fabricated citation looks exactly like a real one, a revised forecast reads exactly like a current one, and a name that merely shares a word with a sanctions entry looks exactly like the company you asked about. The error is invisible in the shape of the answer, so no amount of rereading the answer finds it.

I did not design seven directions. I kept hitting that and kept building the missing check, and seven is where it landed.

How it works, briefly

One host, https://sugra.ai. One header, x-api-key. Every response carries the data and a meta block with the source and the observation time, and the whole argument of this series is that the second part is not decoration.

Plans differ by volume only. Every direction is on every plan, including the free one, because gating a direction would mean deciding for you which question you are allowed to ask.

Two things you will notice reading the posts. Numbers arrive with their own caveats attached, including the places our own method is weak. And where we compute something instead of passing it through, the formula is in the response, because a score you cannot take apart is worth less than a rough one you can.

What I will not claim

Coverage is uneven, by direction and by country, and it is thinner outside the largest economies. Some sources publish with a lag no engineering removes. Screening is a signal and never a determination. Nothing here is investment advice, and no number in this platform tells you what to do.

We also do not pretend to have produced the underlying record. Sovereign registries, central banks, statistical agencies, exchanges, regulators and academic bodies do the publishing. What we built is the reaching, the normalising, and the discipline of keeping provenance attached so you can go back and argue with the original.

Start anywhere

The posts stand alone, so read whichever matches the question in front of you rather than working through in order. If you want the argument at its sharpest, Finance and Macro carry it best. If you want the reason any of this exists, it is in AI is confidently wrong about live data. And What to use when is the map, if you would rather pick by question than by product.

A key takes a minute at app.sugra.ai, and the free plan is enough to find out whether I am right.

curl -H "x-api-key: $SUGRA_API_KEY" \
  "https://sugra.ai/api/v1/fred/series/CPIAUCSL?limit=1"

If that comes back with data and a meta block, everything else in this series is the same idea applied to six more domains.