The process you see is always traceable.
Sovarion scans many assets, but customers should only see a card when the expected move is meaningful from the forecast start price. Each published call has a source chain: live market data, catalyst context, market-understanding impact, Brain thesis and archive proof after expiry. The AI writes the thesis; it does not get to skip the gate.
One score. Five inputs. Each one traceable.
Catalyst and context pressure
Measures whether something real is pushing the asset: live news, filings, macro events, sector rotation, commodity moves, policy decisions or dependency-chain pressure from the market-understanding layer.
Setup strength
Checks whether the chart structure can actually carry the move: trend start, higher lows or lower highs, range expansion, volume support and distance to the next obvious liquidity level.
Directional evidence
Decides long or short from the balance of evidence. We compare asset momentum, peer moves, sector breadth, index pressure and whether the move is continuing or exhausting.
Volatility and liquidity context
Filters moves that are too small, too stale or too hard to execute. A published day-trade card must target a meaningful move from the forecast start price, not a cosmetic change in the day percentage.
Brain conviction
The Brain receives only qualified candidates. It must return direction, target price, variable intraday ETA, evidence, counter-evidence and an expiry window before the card becomes visible.
From raw market movement to a scored thesis only when the move is worth showing.
Latencies are design targets, measured on simulated workloads. p50 from production shadow-run lands on /status once closed Alpha begins.
Four inputs decide whether the Brain should spend attention.
Live price acceleration
The scanner looks for a move beginning now: quote change, candle pressure, day-range expansion and persistence against the asset's normal intraday behaviour.
Volume and range expansion
A price move without participation is usually noise. We weight relative volume, clean range expansion and whether buyers or sellers keep control after the first impulse.
News and event pressure
Direct and indirect news is mapped to assets, sectors and dependency chains. A headline does not need to name an asset to matter if it touches that asset's drivers.
Cross-asset and sector lead
Sector peers, indices, commodities, currencies and market regimes can move first. The system checks whether the asset is following a larger chain or moving alone.
We tell you why we might be wrong.
Every published forecast ships with an explicit counter-evidence section. Not a disclaimer, not a hedge — a concrete reason the thesis could fail, tied to a measurable downgrade trigger.
This is part of the forecast contract: no counter-evidence → no customer-grade thesis. If the system cannot argue against its own call, the call should not be treated as finished.
“Semiconductor breadth is no longer one-way; if sector relative strength breaks intraday, the card should downgrade instead of chasing.”
A good setup in the wrong regime is a bad trade.
Sovarion classifies the current regime continuously: risk-on, risk-off, rotation, compression and stress. The regime does not replace the signal; it changes how much trust we place in the same setup under different market conditions.
The market-understanding layer sits beside that regime view. It maps events through sectors, commodities, policy, supply chains, currencies and asset-specific profiles so the Brain can see why an asset may react even when the headline does not name it.
What the engine doesn't do.
We don't publish tiny moves.
A 0.1% forecast is noise for the customer we are building for. Day-trade opportunities must clear the expected-move gate from the forecast start price.
We don't force fixed horizons.
The ETA is part of the analysis. A valid day-trade call can mature in minutes or hours, but it must remain intraday and carry its own expiry.
We don't let AI skip the gate.
The Brain can explain, rank and forecast a qualified candidate. It should not invent a card when scanner quality, live data, or actionability is missing.
We don't pretend black swans are known.
Some shocks are not forecastable before they exist. The system can react faster and map the impact chain, but it should not claim impossible foresight.
We don't hide misses.
Resolved calls move into the archive with start price, target price, final price and hit-rate accounting. The uncomfortable sample stays visible.
We don't give advice.
Sovarion is research software. Execution, sizing, brokerage choice and whether to trade are user decisions.
Fast · Transparent · Signal-focused
Watch the market, not a watchlist.
The current universe is around 230 liquid assets and the architecture is built to expand by exchange, category and provider without redesigning the product.
Only publish moves worth acting on.
A 0.05% prediction is noise for a day trader. Sovarion gates by actionability, expected move, confidence, live market state and forecast horizon.
Every call faces the archive.
Cards do not disappear after the trade window. The archive stores the original forecast, target, starting price, final price and the resulting hit-rate sample.