A local attention budget for recommendation feeds

I’ve been building a browser extension around one narrow idea: a local, user-configurable attention budget. It watches coarse signals—repeated feed-page loads, rapid scrolling, and continued interaction with recommendation cards—then adds friction after a threshold the user chooses. That could be a pause screen, a session summary, or a finite-view switch. Counters stay in browser-local storage; nothing goes to a platform or employer, and the permissions and network behavior should be inspectable.

Ordinary timers are easy to dismiss and often arrive late. A feed with no bottom removes the natural “I’m done” cue, so a deliberate pause may work better than another countdown. But this can become paternalistic fast, and the detector will be wrong: research, work, accessibility needs, news monitoring, or seeking support can all look repetitive. Overrides and site/mode exemptions are essential. Should attention controls be transparent, user-owned settings, built-in platform safeguards, or entirely our responsibility? Counterexamples and alternative designs welcome.

“Nothing leaves the device” needs to be a demonstrated property, not just a privacy promise. I’d want minimal host permissions, no remote configuration, and an obvious network inspection path. Local storage helps, but an extension can still exfiltrate data if its code or dependencies change.

I like the finite-view option more than a moralizing interruption. The UI should say what signal triggered it and offer “continue for 10 minutes” rather than imply the user has failed. Otherwise the tool becomes another voice competing for attention.

The stopping-cue point is well supported, but I’d avoid claiming timers generally fail. They may be useful for some people; this is better framed as a different intervention for feeds that lack endpoints. Any evaluation should separate reduced scrolling from simply annoying users into disabling it.

The hard part is detection, not the pause screen. Feeds mutate their DOM, scroll behavior varies, and “rapid scrolling” could describe someone trying to find one specific post. I’d start with explicit per-site adapters and make the default threshold conservative rather than pretending there’s a universal classifier.

I’m less convinced platforms will make safeguards user-owned when their incentives point the other way. A local extension is valuable precisely because it moves the decision outside the feed, but it also puts maintenance on users. Maybe browsers should expose a standard, inspectable attention-control interface.