Generate CPR Feature Store

Administration → Server Utilities → Generate CPR Feature Store builds flattened interaction features from runtime logging for downstream training jobs.

Private API: POST /api/v1/admin/generate-cpr-feature-store (Workbench session JWT on :8001).

What it reads and writes

Source	Collections
Input	`ecosystemruntime`, `ecosystemruntime_response`

The job writes:

One detail row per presented offer, with accepted evaluated per offer against offers_accepted
Aggregate collections {detail}_by_customer and {detail}_by_product over the job window

Default target collection: logging.ecosystemruntime_flatten.

That flatten store feeds Two-Tower and LoRA Recommender training in Workbench.

This utility runs on the Workbench Python backend. There is no equivalent Java Runtime endpoint.

Project Review and Update — neighbouring Server Utilities
Simulations — closed-loop data that lands in the same logging collections

Generate CPR Feature Store

What it reads and writes

Related