Imagine putting a math problem in front of someone and asking them to solve it. They correctly identify it as a system of linear equations. They volunteer that they would solve it using x algorithm which has a time complexity of y.
Then you ask them to actually solve it, and they can't even make the first movement towards doing so. They mentioned LU decomposition, but they can't even do Gaussian elimination on paper. They don't know what elementary row operations are. They can't obtain an augmented matrix or put it into (reduced) row echelon form. They don't know anything about linear independence or the rank of a matrix. You put an inconsistent system in front of them and they keep banging away at it, determined to find a solution...etc.
That's what it's like interviewing one of these senior engineers. It's surreal - they confidently pattern match the problem using limited heuristics, and they toss away low hanging fruit to demonstrate knowledge. But when you ask them to do something practical and specific, they either refuse and zoom out into abstract-land again, or they hopelessly fail.