Triple-barrier labelling: upper target, lower stop, vertical expiry. Every prediction has a measurable, time-bounded outcome.
Technical, statistical, ML/AI, and alternative-data agents execute in parallel. Each maintains its own weight updated after every resolved outcome.
Hedge/Bayesian/VW aggregation. Platt Scaling post-processing ensures p=0.70 genuinely means 70% of comparable events resolve correctly.
Purged cross-validation with anchored and rolling splits eliminates look-ahead bias. Out-of-sample regime exposure on every fold.
Every prediction is emitted in DDL-69 ledger format (p_upper/p_lower/p_none). Agents update weights after outcome resolves.
AGILERA emits predictions in both native format and DDL-69 ledger format. Auth and billing flow through Staz Media, LLC.