One of my working rules says every change goes on a branch, however small. It lived in a document, nothing checked it, and it had been broken 6 times.
Last week it became a commit hook in all 7 of my repos. Commit rather than push, because by push time the work is already on main.
At first it enforced nothing. Every checkout was behind its remote, so git pointed at a hooks folder that didn't exist yet, while all 7 reported as protected.
My Proton Bridge server reports the SPF, DKIM and DMARC verdicts on a message, and points out a From domain that doesn't match the Return-Path.
It stays quiet about your own aliases. Mail through SimpleLogin always has a Return-Path that differs from the sender, so it says that rather than raising a flag.
A warning that fires on ordinary mail teaches you to ignore warnings.
Not affiliated with Proton.
https://considus.com/proton-bridge-mcp/
Two runs agreeing is normally the thing you want to see when you measure a page.
On the 1st of August I took 11 PageSpeed measurements of one URL, and two of the analyses agreed on First Contentful Paint to 13 decimal places. Lighthouse can't do that twice, it was one analysis handed back under two timestamps.
The fix was to count runs that match on everything as one, which understates the sample rather than overstating the certainty.
https://considus.com/journal/pagespeed-replays-cached-results/
Building an MCP server? Find out how long your client waits on a tool call. Mine gave up at 60 seconds, progress notifications or not.
My PageSpeed server's long checks all timed out, while the server carried on, finished, and posted the answer into a pipe nobody was reading any more.
The limit isn't in the protocol, so a server can't discover it. Long work now starts a job and hands back an id, collected by a tool that answers straight away.