The "know" has two levels (for which I lack good terminologies):

• abstraction, theory ("black box"), model, attitude, perception, etc. and
• data (input), information, fact (instance of abstraction), etc.

People had had very good and pretty correct and accurate astronomical data for a long while even during the ancient times but their theories and perceptions about the universe were still "incorrect" (incorrect by today's standard).

In programming if you get some factual, syntactic or logical things wrong, your program may not even run. But if you got your model or design wrong, it might take a while before you'd figure out. If the goal or business criteria for a program are fuzzy, you might hardly know whether the program itself is correct or not.

Attitude and perception are hard to change. Maybe that's partly why major theories and models propagate slowly (a macro stability, so to speak), albeit data and facts kept pouring in (a micro race). Galois theory in math, quantum mechanics in physics, input-output analysis in economics, for instance, they all took decades to be generally accepted. Once accepted, the (micro) race to applying them began.

Sometimes I feel a CS student learnt so much about algorithms and stuff, but not enough about "analysis" or problem modeling, which is partly art, partly science. A manifestation would be like, someone could write a fairly efficient query but fail to design an efficient and correct database structure that effectively solves the real life problem.

It is like being right at the micro level but wrong at the macro level.