The incorrect data in the take-home unfortunately affected tests we sent out in a one week period when the open-source API we used broke its data (more rows, nonsense data, but columns unchanged), and we would like to apologize again for disruption of running the altered rows through the analysis again and interpreting your updated results. Our take-home test is designed to minimize the time burden on candidates and this certainly went against it. This unfortunately was concurrent with Foundry being unusually busy on projects and having slow interviewing loops, 5 weeks is far above both our average and what we aim for. We now have a larger pool of interview staff, have changed our ATS software, and updated our processes to ensure that interviewing with us will not be that slow in the future regardless of constraints. Our in-person interviews do include topics outside of data science algorithms and statistics. Some of these are more closely aligned with software engineering (fairly common for Data Scientist interviews), but some focus on structuring problems and the practical, real-world so-what of data inputs and modelling outputs (fairly uncommon). We think that is an important part of what it takes to succeed in Foundry and our vision of practical AI, but it is certainly not the only way for AI/ML teams to work. We are reorganising our interview system to expose candidates to strategic problems in our phone interview rounds before the on-site.