[FLUX-YOU]

>Y is valid, and when Y is a problem

Here is my issue with this reasoning: Y could turn into a problem in production and cause real impact of the business before you know it's a problem. You're writing hundreds of Y's over the course of months as the project goes on so some amount of first-guessing is good to have. You don't want to play wait-and-see with all of them. If it is a product where demand can spike very rapidly then it is more important to take the time to find a best solution first. One day, you thought Y would never reach this magnitude but it just did because something went viral and the internet is essentially DDOSing you (happens all the time with sites posted here going down). And some companies may take to this approach on the first step and don't want to wait until things fail to realize they used the wrong data structure for this demand. Because often, it only takes a single thing to break a product or produce the wrong output or fail gracefully.

Now naturally if your interviewer wraps this problem up in vagueness, confusion, and gotchas then you aren't going to do well. A sane interview should be more than willing to answer questions about magnitude and that tells you that Y is now a problem and you need to think of a better solution than your first one even though it is fine for most use cases.

If it seems like they were trying to confuse you or just throw their ego around, just write this business off. They are only interested in playing slots. They pull the lever until eventually someone walks in and just happens to know the answer and also wants to work there.

[rawnlq]

I think might have been Google who said this, but you should design for 100x of your expected load and when you get within 10x you redesign and rewrite it.

It will make sure you don't try to design for a billion users when you have a thousand but still gives you decent room to handle spikes and more time to write a more lasting architecture.