Analysis that doesn't decide is noise. Quantitative models weigh every factor across crypto, US stocks and indices, and the read takes a position: what matters most — and why, with the confidence stated and the condition that would void it.
Like forecasters, not fortune tellers: the AI measures, it doesn't opine — and the misses stay on the record.
We process your address to send you the bulletin, and you can withdraw consent with one click at the bottom of every email: read the privacy notice.
{{ notaHero }}
The wind is turning on perpetuals: liquidity leaving for three sessions, funding still negative.
On the S&P nothing weighs either way: no forced read, flat stance.
Open a read in full →Anyone can open a trade in three taps. Working out whether it makes sense — which factors really weigh, how much risk sits underneath — is still the hard part. The enemy is not misinformation: it is analysis that will not decide.
“It could go up or it could go down”
Technically impeccable, practically useless: it leaves you exactly the job you wanted to delegate, which is choosing.
The morning-after analysis
Commentary on things that already happened. Interesting to read, useless if you have to decide now.
The black-box tip
“Buy here”, with no why, no cross-check, no history of outcomes. You just trust it — until it stops working.
{{ letturaTitolo }}
{{ letturaSintesi }}
{{ letturaDecade }}
Example reconstructed from historical data, not a live read. Reads are not financial advice.
Three rules, always the same, on every market we cover. If one of them can't be met, the read doesn't go out.
You always know what the model looks at: flows, derivatives, market structure, revisions. Every source is listed under the read.
Every factor has a weight and a direction. When the picture changes, you see which weight moved and by how much.
The read takes a position, states its confidence and says what would void it. Below 45 in weight, it stays flat.
The model doesn't predict the price. It estimates how much each force in play weighs and which way it pushes — then adds them up. If the sum doesn't clear the threshold, no read goes out.
confidence = agreement among factors × data quality × historical reliability of the dominant factor
A number from 0 to 100. It is not a probability of profit: it is how consistent the picture is with itself.
On-chain and exchange flows, derivatives (funding, open interest, skew), order-book microstructure, analyst revisions, macro calendar. Frequency: from one minute to daily.
Every series becomes a deviation from its own regime, not an absolute value: funding at −0.012% weighs differently in a calm market and in a stressed one. Dirty data is discarded, not interpolated.
Each factor gets a weight from 0 to 40 based on how it has held up historically on that market and in that regime. Weights are summed by direction: whatever stays below the threshold is, by definition, noise.
A language model writes the read from the computed weights alone — it invents no factors. The invalidation condition is monitored: when it triggers, the read closes and goes on the record.
The core of Sinottica is quantitative. The models work on numbers — they recognize the market regime, estimate how well each factor has held up, measure how coherent the picture is. Language comes last, and only to report what the numbers have already decided.
A state model classifies the market — calm, strained, stressed — because the same data point weighs differently depending on the weather. It is the step that makes very different series comparable.
Tree-based models trained on that market's history assign every factor a weight from 0 to 40. Validation is walk-forward: train on the past, measure on a future never seen before, never the other way round.
No single model decides. When the models in the ensemble disagree, confidence falls — and if it falls below 45, the read doesn't go out. Disagreement is information, not a problem to be hidden.
A language model turns the weights already computed into plain English, and may cite only the factors present in the calculation. It adds no causes, rounds no numbers, decides no direction.
What the AI does not do. It doesn't predict the price, doesn't generate news, doesn't comment on charts and writes nothing the quantitative models haven't already measured.
Why it is verifiable. Every read carries with it the weights and the sources that produced it: you can check the reasoning line by line, even when it ends up among the misses.
Why it stays honest. The models are retrained on a regular cycle and the results, good or bad, go into the public record unfiltered.
Every market has its own data and its own rules, but the way of reading it is the same: the factors are weighed, the confidence is stated, the condition that would void the read is written down. The areas are also the topics of the group — content and channels.
The historic core: perpetual funding, open interest, options. This is where it all started.
A weekly screen sifts the US market and pulls out the names worth stopping on — with the reason the others were discarded.
S&P 500, Nasdaq, sector indices and the funds that track them: dealer exposure, options expiries, flows. They go through the weekly screen too.
Equity baskets built around a theme and followed over time: the radar beyond the single stock.
The markets that price events: how much money really believes what the papers take for granted.
Above them all sits the macro — rates, data, the calendar. It is not an area of its own: it is the current that moves all five, which is why it enters every read.
The reads reach you where you already are: on Telegram when you are out, on the site when you sit down to think them over. Same flow, two windows — no software to learn, no dashboard to man.
Enter PlanciaThe morning bulletin and the reads when the picture changes, with the condition that would void them. On your phone, without opening anything.
The operating picture with the factor weights, the levels and the full reasoning. And the record, to reread the original call next to its outcome.
Every closed read enters the record with the confidence it had on the day it went out. We take nothing down: a model that hides its errors isn't a model, it's a salesman.
These are 86 Bitcoin options trades actually closed on Deribit between 16 November 2025 and 10 July 2026, taken from the broker statement: 63 in profit, 23 at a loss, profit factor 1.18. It is the typical profile of selling volatility — you win often and small, you lose rarely and large — which is why the number of winning trades, on its own, would say very little.
The record of published reads, on the other hand, has only been open since May 2026 and still has too few closed outcomes to draw an honest percentage from. When it has enough, the percentage will appear here — not before.
What a line of the record looks like
Example rows, to show the format. A read is a hit if, within the stated horizon, the market moves in the direction indicated beyond the noise threshold. Broker data before commissions; the percentage return is not computable because the account capital is not recorded. Past performance does not indicate future results.
Record in Plancia →Where the reads reach you: the operating picture, alerts when a weight moves, the record.
The tools: supporting artifacts and software, tracking, educational content.
The 07:00 crypto bulletin: public, free, warmer in tone and just as rigorous.
The 08:00 New York time bulletin on US stocks, indices and futures: it lands before the opening bell, with the pre-market already showing its hand.
You start with the free bulletin. If the method convinces you, move up a tier — and move back down whenever you like.
The daily briefing and the Telegram group.
Every read, the moment the picture changes.
For those who want to get their hands on the model.
Reads are not financial advice. They are the picture you decide on.
“You can outsource thinking, but not understanding.”
— Andrej Karpathy, AI Ascent 2026
A teacher shows you how to understand the markets. Sinottica does the daily work of reading them for you. Understanding comes from teachers; the everyday thinking you delegate to us. They are two different trades, which is why we do not compete: we work together.
Are you a teacher? Let us work togetherThe rest are in the full FAQ; if yours is missing, write to us from Contact.
No. We give reads: a reasoned direction, with the weights that support it, the confidence and the condition that voids it. We don't say how much to buy, at what price to enter or where to put the stop — those are your decisions.
Analysis lists the scenarios and leaves you there. A read weighs those scenarios and says which one prevails, by how much and until when. The price of that choice is that sometimes we are wrong in a verifiable way: that is what the record is for.
When the picture changes, not on a fixed schedule. On average three to eight open reads a week across the markets we cover, plus the morning bulletin. In flat weeks few reads come out: that is the correct behavior, not a fault.
That the picture is highly consistent: almost every factor pushes the same way and the data is clean. It is not a probability of profit, nor a historical success rate. A read at 72 can be wrong, and the record tells you how often that has happened.
No, but you do need to manage risk yourself. The reads are written in plain English and every technical term is explained the first time it appears. If you are looking for someone to make your trades for you, Sinottica is not for you.
Yes, from your profile, without writing to anyone. The subscription stays active until the end of the period you paid for, and the public record remains accessible afterwards.
Activate the Free plan: you get the daily briefing and an invitation to the Telegram group for your market. No third-party promotions, no invitation to trade.