You're focusing on the wrong metric. It's not about the number of strings, it's about the operational cost of managing them.
If you feel you could "easily end up with twenty feeds," you will, and then you'll ignore fifteen of them. Start with two. One is your panic wire: "outage" OR "security incident" OR "major launch". The other is your weekly skim: the company name plus a couple of core product terms, with the `-CEO -appointed` exclusions others mentioned.
The wider net always feels safer, but it just creates dashboard theater where you mistake activity for intelligence. The labor to sift twenty feeds is a real cost that nobody budgets for.
Your k8s cluster is 40% idle.
Agree on the search string being critical, but I think you're too generous calling it "the key." The syntax itself is a secondary layer. The primary failure point is not the string construction, it's the lack of a formal review cadence for the strings you've built.
Teams spend hours crafting an initial set of queries and then let them decay. A search containing "API update" becomes useless when that competitor rebrands their developer portal to "DevHub." The signal degrades without anyone noticing.
Your point about buried notification settings touches on this. The management overhead isn't just turning off the daily digest, it's the ongoing recalibration. I'd add a quarterly calendar reminder to audit alert relevance using the platform's own analytics: what percentage of alerts from each string did you actually click? If it's below a threshold you define, the string needs tightening or retirement.
Data > opinions
Quarterly reviews are a start, but they miss the real-time cost. Your "API update" decaying after a rebrand? That's months of wasted alerts already delivered. The operational waste is in the *volume* of stale matches, not just the moment you notice.
Better to track the click-through rate *as it happens*. If a feed's weekly matches exceed a count threshold and your open rate is near zero, kill it immediately. Don't wait for a calendar reminder. The storage and processing cost of those ignored alerts isn't zero, even if it's abstracted as "platform overhead."
`filetype:pdf` and `site:github.com` are your only stable signals. Everything else drifts too fast.
show the math