We don't tell you what to trade.
We show you the structure behind the move.
We built a directional prediction model. We tested it under a protocol written before we looked at the results. It failed, so we stopped selling the thing it was supposed to power. This page is the experiment, what we publish instead, and what to ask anyone who sells directional calls.
The Promise You Came Here For
You searched for something that tells you what to trade, so let us be straight about what that usually means: a service that tells you which stock, and ideally which way, so the judgement is handled and you only have to act.
That is a reasonable thing to want. Watching markets is exhausting, information arrives faster than anyone can read it, and the appeal of a system that compresses all of it into "trade this now" is obvious. We are not going to be condescending about the search — we built toward exactly that product for months.
We are just not going to offer it to you, and the rest of this page is why.
Why We Stopped
Our architecture originally assumed three components had to work: detect unusual activity, estimate its direction, and time the exit. We tested the direction step properly.
Properly means pre-registered. Before touching the analysis, we wrote down what would count as success: an AUC above 0.60, balanced accuracy above 0.57, stability across weeks, and beating the naive baseline of "the opening gap continues". Fixing the bar first is what stops you from moving it afterwards.
The model failed every criterion:
Two details matter more than the table. One week came in at 0.450 — below a coin flip, which is the signature of noise rather than a weak effect. And on the subset where the model was most confident, precisely where a directional-call product would concentrate its claims, the directional bets still lost money net of costs. No subset survived costs.
We did not soften this into a weaker version of the same promise. We abandoned it, and rebuilt the product around what was left.
This is also what finance theory would have predicted. Volatility clusters and persists — that is why the size of moves tends to cluster. Direction is the one thing every participant is competing to price correctly, so it is arbitraged away first. Finding that direction was unpredictable and magnitude was not is the ordinary result, not a surprising one.
What We Measure Instead
We measure structure: Estimated Gamma Exposure, insider Form 4 activity, unusual volume against each ticker’s own baseline, short interest & FTD, analyst rating history and the market conditions behind every trade in your journal. Market events are one module: each carries its type, price at detection and the tone of the source text. No direction, no entry, no target.
Why is that useful without a direction? Because knowing that something material just happened, in which name, how unusual it is, and having the context to evaluate it — a median of 22.6 minutes after the source publishes (measured Sep 22 – Oct 2, 2026), without watching six feeds — is most of the work. The judgement it leaves you is the part that was always yours, and the part no vendor can honestly take.
Four Questions for Anyone Who Sells Directional Calls
If you keep shopping, and you reasonably might, these are the questions we would ask.
We think these questions are more useful than any ranking, partly because we have failed some of them ourselves and changed the product as a result. We would rather you asked us them than took this page on faith.
If You Want to Be Told What to Trade
Then we are not it, and we would rather say so on the page you landed on than after you have paid.
If instead you want to see the structure — positioning, insider activity, volume, short interest and events in context — and make the call yourself, the free plan requires no card. It is capped and truncated, and it arrives a median of 23.8 minutes after the source publishes (measured over the 30 days to Oct 2, 2026) — but it is real, and a week of it will tell you more than this page can.
Frequently Asked Questions
Do you tell me what to trade?
No. TradeAI News is an all-in-one market intelligence platform — Estimated Gamma Exposure, insider activity, unusual volume, short interest & FTD, analyst ratings, market events and a trade journal. We do not publish a direction, an entry, a target or a stop, because we tested directional prediction under a pre-registered protocol and it failed every criterion we had set in advance.
What exactly failed in the test?
We asked whether the direction of a post-news move is predictable from the data available at detection time. The success criteria were fixed before the analysis: AUC above 0.60, balanced accuracy above 0.57, stability week over week, and beating the naive baseline of "the opening gap continues". The model reached AUC 0.573 and balanced accuracy 0.528, barely level with the naive baseline. Week to week it swung from 0.713 down to 0.450 — below a coin flip, which is the signature of noise rather than a weak effect. And on the subset where the model was most confident, the directional calls still lost money net of costs.
Why is magnitude easier to predict than direction?
It is what finance theory would predict. Volatility clusters and persists, so the size of a move carries some forecastable structure. Direction is precisely the thing every participant is competing to price correctly, so it gets arbitraged away first. Finding that magnitude showed some measurable structure and direction did not is the ordinary result, not a surprising one.
So what do I actually receive?
A notification containing the ticker, the catalyst type from classified event types, the price at detection, and the tone of the source text — whether the news reads favourably or unfavourably for the company. Delivered to the dashboard, and by REST API on Elite.
Do you publish an accuracy figure?
No. We have not validated one we would defend in public, so we do not quote one. Events are shown with their type and context, not a ranking.
Is the news tone a direction in disguise?
No. It describes whether the source text reads favourably or unfavourably for the company — a property of the language, not a forecast of the price. Favourable news frequently precedes a fall, and the reverse, because the expectation was already priced. We do not model the link between the two.
Who is this for, then?
Traders whose bottleneck is finding out that something happened, in which name, and how unusual it is — without watching six feeds. If your bottleneck is deciding what to do once you know, this will not solve it, and we would rather say so here than after you have paid.
See the events. Make your own call.
Free plan, no card. Capped and truncated, but real.
Not financial advice. Trading involves risk.
Last updated October 14, 2026