> genocide thing

I can only think of one social network when I hear that word. Are there others?

The funny thing with social networks is they all have both of these cohorts, so long as you go down the right rabbit-holes and engage with the right content creators

Rachel worked at Facebook for some time a while ago. She's been consistently mentioning the cat pictures to refer to it in her blog.

I suppose this way of referring to it is also meant to belittle it.

I assume the "the whole genocide thing and enabling fascism" part is more a dig than something to narrow down.

Yeah, might be Twitter, but might as well be Facebook. Though I'm leaning towards Twitter

Maybe she didn’t really want to tell you which one it is :)

Rachel Kroll worked at Facebook. Her work history is not explicitly listed on her blog, but you can find it:

https://news.ycombinator.com/item?id=13400687

https://medium.com/wogrammer/rachel-kroll-7944eeb8c692

https://www.usenix.org/conference/srecon16/speaker-or-organi...

Maybe she left when it transitioned from cat pictures to fascism.

I would say you were less of a douchebag for taking those gigs once upon a time than now days.

This is really important - because "not supported" can range from "eh, we don't have to help you if it explodes" (which they often DO help with, but it's not official) to "this is literally against the license, the law, and the basics of human decency to even attempt".

>or maybe it’s their executives who are afraid?

I'd say yes. It seems like an unforced error to them to broadcast publicly how bad things might be, but that's because they want software to be an efficient factory that turns computer cycles into money.

You do need to be careful that you don't end up with a "build it and they will come" mindset when creating that Golden Path, though - it needs to have early and continuous input from actual users.

There's a tension between a theoretical Golden Path that leads someplace no-one actually wants to go, and simply paving every possible "desire line". Managing that is one of the trickiest parts of platform engineering.

Step 3 is where I've seen my org completely fall on its face.

Yes, the golden path is "fully supported" yet after a year or two the company executive swoop in and say "why are we spending so much money on the golden path" and slowly, but surely, the support for is whittled away into nothing until the golden path is out of date and actively punishing anyone that chose to use it.

For example, one of our golden paths was a UX framework built on top of standard web tech for the time. The team maintaining that framework is no more and it's now very far out of date. Adopting it means you are pulling in Angular circa 2016 and that you'll be dealing with incompatibilities between that and any new web component you want to start using.

You can get screwed over by external (to your org, even more globally) requirements

For example I work for a Research University, mostly our software procurement is - while not always excellent - pretty good stuff. Maybe the supplier isn't as responsive as we'd like, maybe the software is buggier, maybe the documentation is worse, probably not all three.

However, the Government, responding to the usual anti-immigrant sentiment, decided it needs all Universities to check that people who are here on a restricted visa to get a degree attend classes.

The underlying sentiment is clearly racist, but OK in some cases you could imagine that's a real issue, a cheap course with foreign students who are registered but actually never attending because they're out delivering pizza or whatever. For a prestige research University though it doesn't make much sense - maybe you graduated top of your class in Taiwan, your parents pay an eyewatering sum so you can study here for an EE Masters to get that job back home with a team designing CPUs - then instead you skip classes to work as a taxi driver? No, absurd. But the government legally requires we solve this imaginary problem, and the only bidding supplier is garbage. So they're basically requiring us to procure garbage.

Because the supplier knows we're obligated, why should they support anything? Why care if it works, or is documented properly, or integrates with all the things they've told the government it can do? They know they're getting free money because of anti-immigrant sentiment, and they can take advantage of that until the winds change.

Their attendance tracking stuff could be useful. You can legitimately imagine having an early warning, OK, Sarah took a week for her mum's funeral, that's sad, but then she didn't attend any lectures at all for the next three weeks, we should have somebody go check Sarah is OK, or, well, she's clearly not OK, have them figure out what Sarah should do next. Take a year pause? Counselling on our dime? Right now, we aren't required to track Sarah unless she's on a restricted visa and the software is awful (for her, for us, for her teaching staff, everything) so we don't track her. So chances are nobody notices, especially if her class is big, and then in another month's time we find out Sarah was struggling after losing a parent, and we find that out because Police have to break her door open as she's been dead for long enough that neighbours noticed the smell.

She mentioned experiments. Facebook is indeed crazy loose with how they handle experiments on the site's user base.

https://journals.sagepub.com/doi/full/10.1177/17470161155995...

Think "hardware groups with next generation experimental hardware". Facebook runs their own data centers, including designing their own servers. How do you verify that the servers actually perform well? One natural choice was to put them to work as web servers. Given its size, there was a lot of tooling there to be able to measure very precisely what the throughput of a web server was, so things like "How many of these new servers should we order" could be answered, in addition to the "does the rack catch fire" questions.

One example of a source of tension that such standards were trying to deal with here was in a group trying to run web servers on machines with SSDs that were way too small: obviously for the bean counter saving a bunch of money on the SSDs was nice, but for the team trying to make sure the disk can fit all the code and logs on it, it was less nice.

It helps, but it’s important to have a name and a green check mark for each result so people understand WHY the magic doesn’t permit them to use the machine.

Yep, this is good. On one occasion, I saved myself a huge amount of back and forth by writing a dumb "test suite" for a software artifact to be shipped to a client; i.e. make sure it was compressed as a zipfile; it included an INSTALL.md; it was installable via [specific installer program]; it included [specific runtime libraries versions]; etc. etc.

Turns out human software engineers are really bad at fulfilling even well-specified requirements, so you will always save time by automating as much of the conformance testing as possible.