Development
The blog that couldn't publish on a schedule
A batch pipeline design that started with a five-minute substrate check: future-dated posts weren't hidden, they were 404s, and nothing ever rebuilt the site.
Dev log
blog-manager
Development
Debugging a hidden checkbox on Medium's real UI
A zero-sized invisible input behind a styled toggle, a disabled state that silently eats clicks, and verification against the live page instead of my own logs.
Development
Adding a Listmonk newsletter integration to blog-manager
Why a newsletter batch needed its own model instead of reusing the 1:1 post-sync pattern, plus Listmonk's odd auth scheme and a Brakeman quirk.
Development
The runbook that lied twice
A one-line docs fix turned into three meanings of one hostname, a renamed runner directory, and automation reintroducing the bug it was meant to prevent.
Development
Medium kept dropping my hero image, so the bridge now pastes it in
Two theories disproved at one manual import each, a paste that saved dimensions but no src, and the native file-chooser path that finally attached the image.
Development
Backfilling reality into a syndication tracker
Prod said zero Medium posts. Rake tasks fixed that in minutes, but the reconciler crashed everywhere and CI failed three separate ways.
Development
The syndication tracker that couldn't answer its own question
Missing status pills hid the real signal, a system test would have merged silently red, and firewall rules killed every outbound container connection.
Development
Reconciling Medium cross-posts blog-manager never staged
An RSS feed that silently caps at 10 items, a synthetic-key backfill that only upgrades, and 76 of 78 titles matching without any fuzzy logic.
Development
Proving the backup actually works
"We have backups" and "we have backups that work" are different claims. A file-level restore, a query against the restored data, and a stray copy cleaned up along the way.
Development
Three lines of config, an afternoon of verification
Uncommenting three Rails SSL flags took ten minutes. Reading the framework source and testing two plausible but wrong review findings took the rest of the day.
Page 2 of 6