This isn't a great article IMO (I say as someone who is a big fan of qualitative user research, mixed methods, etc).

Per the first example, many types of attitudinal data can be quantified and conflating attitudinal data with qualitative data is itself a fallacy. It's possible that there were quantitative attitudinal signals that could have been captured or created as inputs to a more accurate model.

Per the second example, this is more a question of data validity than the metric itself. If the metric could be validated through better design and gamification prevented then it would likely still be a helpful indicator. Granted this is a very hard problem.