Start here. This is the direct spoken answer to practice first.
Overview
The interview value is safe application behavior, not predicting deliberately obscure coercion puzzles.
I use strict equality by default because it compares values without converting between types. Objects compare by identity, so two separate objects with the same fields are not equal. I normalize external strings, numbers, and booleans at the boundary instead of relying on implicit coercion throughout the application. I also treat nullish values and NaN deliberately.