Any system on the market this morning can produce four fluent paragraphs about a semiconductor company. It will read well. It will have a thesis, a risk section, and a tidy closing sentence. Whether any of it is true is a separate question, and increasingly the only one that matters.
Fluency has stopped being evidence of anything. That is a genuinely new condition for financial writing, and it has an uncomfortable implication: a reader can no longer judge an analysis by how competent it sounds. They need to be able to check it. Which means the interesting engineering has moved from the prose to the machinery underneath it.
The Rule That Shapes Everything Else
The most consequential design decision in this kind of product is also the most boring one to describe. If a claim cannot be traced to a specific source document, it does not get published. Not softened, not hedged, not framed as a possibility. Dropped.
That sounds like a content policy. It is really an architectural one, because it forces the pipeline to carry a reference all the way from ingestion to the rendered sentence. The filing, the transcript line, the exchange notice, whatever the source is, has to survive every transformation step with its identifier attached. Systems that summarise first and attribute later cannot do this, which is why so few of them try.
Iteno is built on that constraint. It watches a subscriber’s tickers continuously and, when something material happens to one of them, publishes a short analysis in which every claim carries the document it came from. The reader is expected to click through and disagree. That is the intended use, not a fallback.
Disclose the Delay as Loudly as the Number
Here is a smaller rule with a similar spirit. Every price shown is displayed with its delay status in the same type size as the price itself.
Anyone who has built a market data surface knows why this is unusual. Delay disclosures are traditionally set in six point grey type at the bottom of a page, technically present and practically invisible. Enlarging that text costs nothing and buys nothing commercially. It simply removes the possibility that a reader mistakes a fifteen minute quote for a live one, which is the sort of misunderstanding that produces a bad decision and then a very reasonable complaint.
Product teams tend to treat this category of honesty as friction. It is closer to insurance.
Why the Archive Is the Real Product
A single sourced analysis is a nice artifact. A few thousand of them, dated and left up, is something harder to fake.
Every published piece stays published, which means the reasoning offered in March can be read against what actually happened by September. This is uncomfortable for the publisher and useful for everyone else. It also produces the only credential that means anything in this category, because a written record cannot be retroactively improved and a marketing claim can.
Notably, the company does not claim to have invented monitored watchlist analysis. Plenty of firms have done versions of it, some for decades. The claim is narrower and easier to test: that you can check the work.
There is a practical consequence for anyone evaluating one of these services. Ask for the archive before you ask for the accuracy rate. A vendor quoting a hit rate is asking you to trust an arithmetic you cannot inspect. A vendor handing you two years of dated, sourced writeups is handing you the raw material to compute it yourself, and to notice what they chose not to cover.
Priced Like Software, Scoped Like a Publication
The commercial model follows the same logic. A free tier, Core at $249 a year or $29 a month, Pro at $708 a year or $79 a month for traders acting intraday, and a Desk plan at $199 a month for small funds. You subscribe, you read, you cancel. There is no advisory relationship being sold underneath.
That boundary is stated rather than implied: this is a publication, not personalised investment advice, and it does not place trades on anyone’s behalf. In a category where the temptation is to blur that line for a better conversion rate, saying it plainly is a product decision with revenue attached.
The bar for a machine that writes about markets used to be whether it sounded like an analyst. It should be whether you can catch it being wrong.
