The company is considering launching new recommendation-system features and wants to judge their value before any live deployment.

Using only historical data, how would you evaluate whether releasing this recommendation feature is a good or bad idea? Detail the analyses, metrics, and assumptions you would use (do not answer "run an A/B test").

Think offline replay, counterfactual evaluation, uplift or propensity modeling, simulation, historical hold-out metrics.

Meta Data Scientist interview question: Evaluate Recommendation Feature with Historical Data Analysis. {"blocks": [{"key": "8be2490e", "text": "Scenario", "type": "header-two", "depth": 0, "inlineStyleRanges": [], "entityRanges": [], "data": {...