“Does this match what I said I’d do?”
Intent-based focus, and why a list of bad sites is only half of it.
A blocklist knows YouTube is YouTube. It can’t know you’re there for a tutorial on the thing you’re building. You know, though, and you said so when you started.
A list says where you are, not what you’re doing
Site lists are about places. Distraction is about fit. The same site can be work at ten o’clock and a hole at twenty past:
- A YouTube tutorial on the tool you’re learning, and the three videos after it.
- The Swift subreddit, and the one with the cat pictures.
- LinkedIn to message a candidate, and LinkedIn’s feed.
- Facebook’s Ads Manager, and Facebook.
A list has to pick one answer per site. Pick “blocked” and it stops your work; pick “allowed” and it stops nothing. So lists grow exceptions, and exceptions grow into a second job.
Start with a sentence
A watch in SqwarkSelf starts with what you’re doing, in your own words: “marketing and ads”, “pricing page copy”, “Q3 report”. No project picker, no categories. When you drift, the card hands it back to you: This isn’t marketing and ads.
Saying what you’ll do helps on its own. Psychologists call specific plans like this implementation intentions, and a 2006 meta-analysis of 94 tests found they help people reach their goals (Gollwitzer & Sheeran, 2006). A 2025 meta-analysis of 642 tests by the same researchers found them effective again, with a smaller average effect than the 2006 review and signs of publication bias; effects were larger for plans in the “when this happens, I’ll do that” form (Sheeran, Listrom & Gollwitzer, 2025). A sentence and a timer isn’t quite that, but it’s the same idea: decide in advance, specifically.
How the check works, in order
- Always allowed wins. Anything on your always-allowed list (an app, a site, words in a window title) is work, every time.
- Rules settle the obvious. Your own entries, then a short built-in list: x.com and Instagram are social, LINE is chat, Netflix is video. No AI and no network for any of it.
- Your key, for the rest. An unknown site, or one that can go either way (YouTube, Reddit, LinkedIn, Twitch), and you’ve added your own Jev key: one question to TypeSafe’s Jev model, about half a second, comparing the window with your sentence. The answer is remembered, so the same page isn’t asked about again during the watch.
- Only what you ruled out can be blocked. Work and “other” never are, however badly they seem to match your sentence.
Never block on a guess
A model that’s wrong in the blocking direction costs you work; wrong the other way, it costs you ten minutes. So the bar is set high on the side that blocks. A poor match with your sentence never blocks anything on its own. Jev can file an unknown page under something you ruled out only when it’s very sure, and it can let a ruled-out page through when it clearly matches what you said. If it’s wrong anyway, “This is work” is one click.
No key, no answer in time, or an error: the rules decide alone, and nothing is blocked on a guess.
What goes to Jev, and what never does
Only the undecided cases, and only with your key: the app’s name, the window title, the site and path (never the query string) and your sentence. Password managers, banking, Google, Apple and Microsoft sign-in pages, mail, Messages and private windows in Chrome, Arc, Brave, Edge and Firefox are never sent. (Safari doesn’t say which of its windows are private, so SqwarkSelf can’t tell.) More in the privacy guide.
The short version
A list says where you are. A sentence says what you’re doing. The first is easy to check; the second is the one that matters. SqwarkSelf uses the list for the obvious cases and, with your own Jev key, your sentence for the rest.
