I published, verified the version, and my users still got the old file
First thread in here, so I'll say what this forum is for, since that's the thing you're actually deciding about if you follow it.
Not release notes. One thing I got wrong, each time, with enough detail that you can go and check whether you have the same problem. The checklist has 4,352 items in it and nearly every one exists because somebody, usually me, found out the hard way. This is where I write those up while they're still embarrassing.
Here's the first, from Wednesday.
I shipped a release. Checked the version on the registry, correct. Checked the pinned CDN URL, correct, new data in it. All green.
prodcheck is live. Here's what it can't do!
It's live, after a week of getting it ready in public!!
So instead of the pitch, the three things it doesn't do.
It never tells you something is safe. The worst answer it can give is a confident pass, so it doesn't have one. Every item comes back as a finding with a file and line, as UNKNOWN, or as not applicable.
It doesn't run anything. It reads and reports. Nothing gets fixed while you sleep.
88% of the items work on any stack, which is another way of saying 12% don't. If your stack isn't one of the 26 with a supplement, some items won't mean much to you.
If you've used a checklist you didn't write: what made you trust it, or stop trusting it?
What broke in your production that no checklist warned you about?
Until this morning, the item count on this page was 20 behind the real list. It said 4,352. The list is at 4,372.
My own pre-push check makes every number in the repo and on the site match the data. Product Hunt isn't in the repo, so nothing was looking.
Fixed now. The check did its job everywhere it could see. The one place it couldn't was the page I'm launching from.
So, the day before it goes live here: what bit you in production that you never saw on any list?
If it isn't in the 4,372, that's the gap I most want to hear about.
🚀 prodcheck launches here tomorrow. What getting it ready taught me about followers
Tomorrow, prodcheck goes live on Product Hunt. It's the checklist I wished someone had handed me before my own launches: the things that break in production that nobody warns you about, sorted so you can check them before your users find them.
The lesson I learned too late while getting it ready is about followers, and it might save someone launching soon a week.
A thread in a topic forum like p/general grows your profile. Only a thread in your product's own forum shows the product's Follow button, and product followers are the ones Product Hunt notifies on launch day.
I found that out with my profile at 54 followers and this page at 2.
So if you'd like to know when it's live, the Follow button above this post is the one. That's the whole ask!
The guard I reviewed, approved, and never actually ran
I spent a day writing guards into a build config. Deploy by digest instead of by tag, refuse to deploy when a required value is empty, run the tests between install and build.
Then I created the CI trigger from the console wizard, because it was right there.
The wizard writes its own build steps inline, and a trigger with inline steps ignores the file in your repo completely. So every guard I'd written was reviewed, committed, and never ran once.
What makes it the worst kind of bug is that nothing failed. The builds were green. The deploys worked. The file sat in the repo looking like protection.
I found it by accident, while looking for something else, because the trigger had no filename set.
It's going into prodcheck as its own item. "The config you reviewed is the config that runs" is an assumption almost nobody checks, and checking it takes one command.
Does your CI trigger point at your repo's file, or carry its own copy?