HORIZON is calibrated AI for trading. A 30-factor regime-aware ensemble that publishes probability forecasts with honest confidence bands — not "the market will go up." Probability-stated, horizon-stated, Brier-tracked, error-published. The first AI trading platform that ships with the calibration receipts a fiduciary needs to defend a position.
Momentum, low-volatility, quality, value, carry, term, default, liquidity, regime, breadth, sentiment, vol-of-vol, correlation, dispersion — and more. Each factor carries an independent Brier history. Ensemble weights are recalibrated quarterly with documented drift, not silently re-fit on the latest data.
Three regime classifiers operate above the factor stack: GFC-2008 (deleveraging), COVID-2020 (volatility-spike), Dotcom-2000 (mean-reversion-failure). Forecasts are conditioned on regime probability, not delivered as point estimates. The published probability includes regime-conditional bands.
HORIZON refuses to forecast over horizons where it has no calibrated edge. If the question is "where is the S&P next year," the published answer is "we don't have edge at that horizon — here's where we do." The discipline is to know the boundary of competence, and stay inside it.
The trading-research market is full of newsletters, signals services, and quant-by-name funds that publish point estimates with implicit certainty. Their predictions don't get scored, their errors don't get published, and their methodology doesn't have a hash. HORIZON inverts that: every forecast is a Brier prediction, every horizon is stated, every error is published in the next quarterly cycle. The product isn't "we're right more often" — the product is "you can verify our calibration, and trust the forecast accordingly."
HORIZON is the trading-forecast composition that the rest of the Aequara stack makes possible. NEXUS routes the workload; AXISTRUTH provides the math; LUMEN delivers the human-legible surface; CIPHER holds pre-publication research where Trustee custodians pressure-test the forecasts before they go public. The same anti-issuer-pay firewall, the same Brier discipline, the same methodology-hash stability.