Timed ML screen: concept questions, then classic models from scratch

A timed online screen for ML-leaning engineers, about ten items in a little over an hour:

1. Multiple-choice ML concepts: compute recall from a confusion matrix, compare bagging and boosting, and say what L1 and L2 regularization each do.
2. Given the weights of a small two-layer network and an input, compute its output by hand.
3. A short string problem: the longest substring made of one repeated character.
4. Implement logistic regression and naive Bayes from scratch (fit and predict) without ML libraries.

The questions are not hard on their own; the pressure is getting all of them right in the time.