Unifying naming conventions and fixing broken event schemas.

Reconstructing missing outcome data from raw logs.

Training initial models and establishing performance baselines.

Deploying inference and feeding signals to CS playbooks.