Partial-outcome extension of the delayed-outcome correction layer

Develop a partial-outcome variant of the delayed-outcome online forecast-combination architecture that can use legitimately scorable portions of a forecast horizon before the horizon is fully observed, while preserving the architecture’s downside-control objective.

Background

The proposed correction layer deliberately discards partially observed forecast horizons and updates its experts and Hedge gate only when outcomes have fully matured. The authors note that this conservative protocol can reduce accuracy when a horizon can legitimately be evaluated before completion, as illustrated by their comparison with PETSA. They identify a partial-outcome variant as compatible with the existing architecture but leave its development unresolved.