[vanusa]

The reason for this is that resume review is really ineffective as a screen.

No is saying they should rely solely on resume review as a screen.

What is under question here is the perceived wisdom (on Google's side) of screening for Director positions using badly programmed chatbots (or recruiters, whatever) instead of having them be asked the same questions by an actual human.

IMO, a better approach now are basic on line code screens designed to take 15m for a qualified candidate.

Perhaps, but companies like to screw that up as well -- having people code into a Word doc, for example (try it sometime), or assigning problems that are either brain teasers or essentially willingness-to-cram filters rather than what they should be -- a simple, straightforward test of minimal programming ability (to determine whether further time investment is merited).

[lokar]

I've used (I think) hackerrank, and just done really basic stuff. Like freshman CS basics, starting with working code and extending it (eg a class that calculates the average of its contents, and you add min, max and median). Surprisingly it screens out about half.

[vanusa]

Finding the median with or without sorting first?

If it's the latter (essentially a test of whether they've memorized quickselect) then that sounds like a horrible filter, actually. Unless it's for the position of "Senior Quicksort Engineer".

[lokar]

They are free to use the standard library, and use whatever language they like (eg python). It's not about how to sort

[vanusa]

If some asks me to "calculate" something I assume they mean from scratch (without using a built-in). That's probably what a fair number of these people did.

Also, my sense is that Python is something of an outlier in having a mean and median functions in it standard library. I could be wrong, but AFAIK Go and JS do not, for example -- so people using those languages would surely bomb (at least on the median calculation part -- again, assuming they interpreted your question as "calculate from scratch").

I don't mean to be pedantic or split hairs. The point I'm trying to make is that even simple-seeming problems can have gotchas to them, depending on the context.

Interviewers could do better by either thinking just a bit more about the problems they select, or just communicating better. But many do not, unfortunately -- I have the sense they just pull problems out of the air, and see what sticks. Meanwhile counting the high number of fails as a success signal.

[lokar]

The site has a detailed spec and test input/output, you can build and run the tests

The standard library is to sort, not take the median

[vanusa]

Well it does seem to be a cute filter for Python programming ability in that they definitely need to have taken a tour of the modern standard library to come up with

  min(x)
  max(x)
  statistics.mean(x)
  statistics.median(x)

But still - no 'for' loops? No if-then-else? You might be giving some of these candidates a pass.

[vanusa]

"No [one] is saying ..."