> I'd have assumed...
I would expect a good candidate/colleague to ask instead of assuming things.
> And then forget to... > Shit happens...
If someone can't manage to complete a simple assignment without forgetting half the steps, I don't want them pushing code to production.
And a service with one http endpoint having reasonable uptime seems very far from magic.
Literally, the only point of the exercise is to get that email address. It says that much in the description of the task, alongside raising a GitHub issue if the server is down. It was hosted on Azure, and had our uptime tool checking it anyway. If it was down, there'd be a badge on the repo homepage saying it was down. It happened once, and I invited the person to the interview regardless regardless.
I say this much when we send over the technical test. It's like the weigh-in for a boxing match. If they want to showboat, it's on them, but all I care about is whether they can follow a simple instruction with minimal code in the time we said it'd take. I couldn't give a shit what fancy tricks you do in your code, because it's code you've written without time constraints in your own time. The whole reason for chatting about past experience is to see what you do in a REAL development situation.
If you can't connect to the server, or it returns a fatal HTTP error, a reasonable candidate would reach out via other means (assuming they aren't deliberately hidden or unavailable).
We were also keen to stress that this should take far less than an hour of their time. We didn't want to see dependency injection, metaprogramming, or anything overboard that showcased a developers knowledge - this is purely just to see if they can deliver code that matches a requirement in the time provided.
In the case where we had downtime (once in 18 months), I just forwarded their email to HR and we spent some extra time going over code in the sit-down interview.
Shows understanding, shows willingness to reach out and solve problems, shows ability to delegate responsibility.
Even if you specifically told me that I must actually hit your server, I'd still test on my own server, and then possibly forget to change the endpoint before submitting (no way to unit test that). Shit happens, and you lose someone on a technicality.
Also, what if your server stops working? Now you're wasting people's time.
Don't rely on magic; it eventually bites you in the ass.