Multi-Horizon Signal Intelligence — Crypto (live)
Every asset in the crypto universe is scored across four horizons simultaneously—12-hour, 1-day, 7-day, and 30-day—so a desk can see where a tactical view and a strategic view agree and where they contradict each other. Each horizon returns a discrete action of BUY, WATCH, or AVOID, carried alongside a directional probability, an expected return net of costs, a model confidence level, and the regime the call was formed in. Because crypto trades continuously, signals incorporate perpetual funding rates, open interest, liquidation pressure, on-chain flow, and relative strength against Bitcoin rather than price and volume alone. Nothing is returned as a bare score: every call arrives with reason codes explaining what qualified it, and a signal that clears the 7-day horizon while failing the 30-day is presented as exactly that—a tactical opportunity inside an unsupportive strategic backdrop.
Multi-Horizon Signal Intelligence — Commodities (in validation and staged rollout)
Commodities are scored on the same multi-horizon logic, adapted to markets that behave nothing like crypto. Horizons are aligned to session structure rather than continuous trading, and the shortest tactical view is expressed against the active session rather than a rolling clock. The signals themselves draw on the drivers that actually move energy, metals, grains, and softs—curve shape and the shift between contango and backwardation; inventory and storage cycles; seasonal demand patterns; freight and physical spreads; dollar strength; and supply disruption risk—instead of transplanting crypto momentum logic onto a different asset class. Each call returns the same decision object: BUY, WATCH, or AVOID with probability, expected return, confidence, regime, and reason codes, so a commodities view sits in the same workflow and the same audit record as a digital asset view without either being distorted to fit the other.
Multi-Horizon Regime Detection — Crypto (live)
Before any crypto forecast is produced, the platform establishes what kind of market is actually running — and how confident it is in that reading. Several independent methods vote on the question, and the output is a probability distribution across market states rather than a single label: bull, bear, sideways, high-volatility bear, low-volatility accumulation, Bitcoin-led, and altcoin-led. Regime is assessed per horizon, because a market that is constructive over 30 days can be hostile over 12 hours, and the platform is explicit when those two readings diverge. That distribution then governs everything downstream—which specialist models get weight, how tightly risk gates are set, and how much exposure is permitted. When regime confidence is low, the system tightens rather than guesses, because an uncertain read on market state is information, not an inconvenience to be smoothed over.
Multi-Horizon Regime Detection — Commodities (in validation and staged rollout)
Commodity markets have their own state space, and the platform models it directly rather than reusing crypto labels. States are built around the conditions that determine whether a commodity trade behaves as expected: curve regime and the transition between contango and backwardation; supply-shock versus demand-led moves; carry-favorable and carry-punitive environments; seasonal phases specific to each complex; and high-dispersion stress conditions where cross-commodity correlations break down. As with crypto, detection runs per horizon and returns probabilities with an explicit confidence score, so a seasonal tailwind visible over 30 days is not confused with a supportive read over the next session. Regime also travels across asset classes: a dollar or rates regime that reshapes commodity behavior is the same macro condition propagating into digital assets, and the platform is built to carry that context in both directions rather than analyzing the two universes in isolation.
Multi-Horizon Validation Suite — Crypto (live)
Every crypto model is tested through walk-forward analysis with leakage safeguards; combinatorial purged cross-validation with purging and embargo to handle overlapping multi-horizon targets; Monte Carlo and multi-asset simulation; and regime-switching scenarios that measure performance in each market state separately rather than reporting one flattering blended number. Validation is run per horizon because a system that is strong at 7 days and merely calibration-grade at 12 hours should be described that way rather than averaged into a single headline. Critically, backtests validate the deployable policy—the guarded, risk-gated strategy a client would actually run—not raw classifier accuracy, since a model can be directionally right and still lose money after gates, costs, and sizing. The output is a validation package with model cards, data lineage, and threshold logs, built to be read by a risk committee.
Multi-Horizon Validation Suite — Commodities (in validation and staged rollout)
Commodities demand validation discipline that crypto does not. Continuous price histories have to be constructed across contract rolls without letting roll artifacts leak in as false signals; session gaps, holidays, and settlement conventions have to be respected rather than resampled away; and strong seasonality means naive cross-validation can quietly train on the same calendar phase it later claims to predict. The suite therefore applies purging and embargo periods calibrated to each complex’s seasonal cycle, tests across multi-year windows that include genuine supply shocks rather than only benign conditions, and reports results by regime and by horizon so that performance in a backwardated market is never averaged together with performance in contango. As with crypto, success is judged on the guarded strategy after costs, and all figures are labeled internal directional accuracy until independent third-party validation is complete.
