The uncomfortable part: we refuse to publish numbers we can't defend.
Most retail tools launch with a hand-picked equity curve and work backwards from there. Sovarion records every forecast, evaluates it after the window closes, and only turns the result into a public headline once the verified sample is large enough.
Real outcomes only, pulled from the forecast archive.
Short-term hit rate, per-asset accuracy, weekly trends, and confidence calibration are visible to signed-in users. We do not publish hardcoded headline numbers here.
Rules we don't break when the numbers go live.
Hit rate asks: did the forecast resolve in-direction within its horizon, before hitting the stop? We don't report P&L because P&L depends on your sizing, your execution, your slippage — things we can't measure from the forecast alone.
If a forecast neither hits its target nor its stop within horizon, it counts as break-even and is excluded from the hit-rate numerator. This is more conservative than tools that count time-expired forecasts as 'directional wins'.
We report hit-rate per regime (risk-on, risk-off, rotation, compression, stress). A system that's 70% in risk-on and 45% in stress is not the same as one that's 55% flat — transparency per regime exposes the skew.
Peak-to-trough on the simulated HIGH portfolio. The uncomfortable months stay in. Moving averages, rolling windows, or cherry-picked start-dates are never used to make drawdown look smaller.
Any horizon with n < 200 is flagged explicitly. We'd rather tell you 'too early to judge' than publish a noisy number as though it were signal.
How we get from zero to numbers.
Scanner flags on
Backend feature-flags flip live in production for closed Alpha (BRAIN_COLD_SCORE_ENABLED, BRAIN_NEWS_CLASSIFIER_ENABLED, BRAIN_SEC_POLLER_ENABLED). The engine starts producing real opportunity scores across 232 assets.
Shadow run begins
Every HIGH and WATCH forecast gets persisted to forecast_outcomes with its composite score, catalyst metadata, regime tag, and horizon. We do not publish any numbers yet — we observe.
Resolution window
Forecasts resolve as price hits target, stop, or the horizon expires. Each resolution is timestamped and tagged with the regime it happened in. No retro-editing allowed.
First numbers published
/v1/track-record endpoint goes public. Hit-rate, avg R:R, max drawdown, median hold, and WATCH→HIGH conversion are rendered here per horizon. Sample size declared, per-regime breakdown shown, disclaimer stays in.
Monthly recalibration
Every month we re-report: hit-rate by horizon, by regime, by theme. Methodology changes land in /changelog with dates so you can see exactly what moved a number.
Want to be in the shadow run that generates these numbers?
We’re opening closed Alpha with 20 serious retail traders. You get full Pro-tier access during the 30-day shadow run at no cost, in exchange for honest feedback. Your trades don’t flow through us — we just observe whether our calls resolve.
Get early access →We don't execute, custody or advise. Sovarion is research software; past performance, once published, will never constitute a recommendation. Public numbers unlock only from verified forecast resolutions with disclosed sample size.