20 agents.
One live probability engine.
AGILERA is the ensemble layer underneath the stack: 20 specialised agents, one calibrated probability vector, and a deployment story that now matches the product quality.
From raw event
to calibrated action.
The engine is not just a set of models. It is a validation discipline, a calibration discipline, and an interface discipline. That is what makes the probabilities credible.
Triple-barrier labelling gives every forecast an upper target, lower stop, and time-bounded expiry so outcomes are measurable.
Technical, statistical, ML, RL, and alternative-data agents execute in parallel, each carrying its own Hedge-MWU adaptive weight.
Hedge, Bayesian, VW, and stack blending compete, then Platt Scaling calibrates the final probability vector.
Purged cross-validation with anchored and rolling splits keeps the whole engine honest against look-ahead bias.
AGILERA emits predictions in both native format and DDLM-69 ledger format. Admin, auth, and investor narrative now align around the Oracle surface.