We’ve not run any A/B tests since GOV.UK changed to require you to opt in to analytics cookies, due to worries about whether we even can A/B test people without their consent. This has been a bit of a bother, so I made our A/B testing framework only bucket you if you’d opted in to cookies. Which then promptly broke our smoke tests, because they weren’t consenting to cookies, whoops.
My documentation spree continued this week, though it was mostly process improvements than actual writing:
I reviewed all the docs which were assigned to our 2ndline support team, and reassigned almost all of them to other teams. This is a nice change, because reviewing docs is the lowest priority thing 2ndline do, so they tend to build up. And when you see 50+ pages needing review, it puts you off even more. Spreading the load is part of solving the problem. The next part is making people care more about doing reviews, but I’ve not figured out how to achieve that yet.
I reviewed all of our “how to” docs and made sure that they were actually how-tos (and if not reclassified them as “learn” docs), and that they had an action-oriented title.
I also found some coverage issues in our docs:
There are lots of dev VM related docs which are almost irrelevant (and can be deleted when everyone is using govuk-docker) and some postgres docs which are only relevant to our old cloud environment (and can be deleted when the last few apps make it across to AWS).
There aren’t actually many docs on govuk-docker at all, that’s probably a gap which needs filling judging from the frequency of questions on slack.
Finally, I did manage to do a bit of writing, documenting how we can use SageMaker to A/B test different ranking models.
I’ve been on call this week, and it’s been the reshuffle this week! Rumours had it being a big one, but then it didn’t quite go as planned (or as feared?). So thankfully I didn’t get any calls on Thursday evening. But I’m still on call until 09:30 on Wednesday, so who knows what could happen.
I did fix one reshuffle-related bug, someone noticed that the “View all announcements” links on ministerial role pages was trying to filter using the
people facet, and not the
roles facet. So nothing was being found, because there isn’t a person called (eg)
prime-minister. Changing whitehall is always worrying, but it ended up only being a small fix.
I’ve been working on implementing the RFC, and have got to the point where whitehall can publish attachment metadata! Unfortunately, I got that implemented just before a reshuffle-related deploy freeze was announced, but I should be able to merge that on Monday.
Next I need to make content-publisher send the same metadata, then I can start phasing out the old way. I had a quick look at content-publisher, and it feels… kind of whitehally. Hopefully we’re not just replacing one whitehall with another.
I’m on leave for two weeks from Thursday next week. I don’t have any plans, I just had some leave to use up before April. I’m looking forward to waking up at noon and not having to do anything.
I read The Phoenix Project, which was an enjoyable enough read, in a way, but not quite the revelatory experience the hype had lead me to expect. The story was pretty contrived, and the main character was basically infallible and guided by a mysterious mentor who just so happened to have almost unlimited influence with the company execs. The IT Ops team are portrayed as poor victims of circumstance, their only failing being not having enough awareness of what work is going on. Once they get a handle on that, they whip those silly developers into shape, cause the security guy to have a mental breakdown (who somehow later comes back as an evangelist for the guy who caused it), and totally transform the build and deploy pipelines, including migrating from a on-premise datacentre with hand-managed VMs to an auto-provisioned AWS environment in a couple of weeks. It’s very much a story about how IT Ops are superheroes, and how everything wrong with IT is caused by someone else.
I’ve started reading The Unicorn Project which is the same story (or maybe a variant on it in a parallel universe) told from the perspective of the devs. Maybe it will be better, though the mysterious mentor has shown up again: apparently he runs a bar now (well, has run a bar for years)—in addition to being a devops expert, a business genuis, and vastly wealthy—and it just so happens to be the bar at which all the angry devs have been meeting to moan about work.
I played a game of Blades in the Dark this weekend1, and it was pretty good. The setting is a Victorian-era city in a world where ghosts, demons, and the like roam the Earth. You play as a gang of criminals in the city of Doskvol, which is a large and important city protected from the horrible nightmarish beings in the wasteland by a wall of lightning powered by demon blood.
In a four-hour session we got through character creation and a short heist. My character almost died, but I survived in the end. I think I’d be up for playing it again some time.
- pipe: use exclusive waits when reading or writing
- The Wall of Technical Debt
- Fizz Buzz in Tensorflow
- Haskell in Production: Riskbook
- A Bridge To Nowhere
- Upcoming stackage LTS 15 snapshot with ghc-8.8.2
- The Myth of the Barter Economy
- This Week in Rust 325
- Issue 198 :: Haskell Weekly