Varun Gupta
Sep 7, 2018 · 1 min read

Great post, Artem! Being able to regression/back-test at the point of committing a diff dramatically improves deploying both new code and new models with confidence.

One question: you note that “Offline, we can’t hit any production endpoints so we stub in production raw data responses from the RiskLog for any service calls made by the risk library.”

Will this ever change? One can imagine the system expanding its purpose beyond “Risk Verification” to something like “Risk Exploration,” wherein we can arbitrarily replace raw data responses from the service calls with whatever we would like to explore.

    Varun Gupta

    Written by