Terraform Applied Cleanly and Measured Nothing
A successful apply means the checks were written, not that anything is being measured. Then the dashboard went red on every panel except one — and that one exception was the entire diagnosis.
k6 Only Reports LCP for the Last Page You Close
Sixty-one monitored URLs reported time to first byte. Nine reported Largest Contentful Paint. The nine weren’t a random sample — they were the last page of each check script, which is the entire explanation.
Ten Browser Checks, Sixty-One URLs, and One Terraform Loop
A vendor cap of ten browser checks per tenant, undocumented on the pricing page, against sixty-one URLs. The workaround turned out to be the cheaper design — and the labels you assign never reach the metrics you want to filter.
We Designed a Monitoring Stack, Then Decided Not to Build It
Three architecture decision records planned a self-hosted performance-monitoring stack in detail. The fourth one priced it against the market and killed it — because the servers were never the expensive part.
The File Watcher That Looked Like a Permissions Problem
A sudo prompt, a macOS privacy dialog, and an EMFILE crash — three failures on a locked-down Mac that all look like missing admin rights. None of them is.
Your File Watcher Deleted the File You Just Saved
Mirror a folder with delete-on-delete enabled, hit Save in your editor, and watch the target file disappear. The bug isn’t in the watcher — it’s that saving a file, on most editors, is a delete followed by a rename.
The WordPress Performance Profiling Toolkit We Actually Use
Not a plugin roundup. Four tools that actually found something on a real WordPress VIP network, what each one is blind to, and which symptom sends you to which one.
What ‘VIP-Style Multisite’ Actually Means for a Performance Bug
Every fix in this series landed at a different layer — application code, shared plugin, edge config, vendor dashboard. The scoping decision matters more than the fix itself, and it’s rarely obvious from the bug report.
Automating Index-Existence Guards So a Deploy Can’t Race Its Own Migration
The runtime existence check that kept one index rewrite safe to deploy in either order was a one-off. Here’s the same pattern turned into a reusable pre-deploy CI gate for a whole network.
The 465-Line Diff That Was Actually a 10-Line Fix
A one-line exception guard turned into a 465-line diff because a formatter had rewritten the whole file first. The fix wasn’t the guard — it was separating the reformat from the change before anyone had to review it.