A monkey, a light,
and a squirt of juice.

In the early 1990s, neuroscientist Wolfram Schultz was recording individual dopamine neurons in a monkey's brain, giving the animal a squirt of juice and watching the dopamine fire in response. As the monkey learned that a light reliably preceded the juice, something changed: the dopamine neurons stopped firing when the juice arrived and started firing when the light appeared instead, the signal that predicted the reward, not the reward itself.

That single observation rewrote what dopamine was understood to do. Dopamine isn't a pleasure detector, it's a prediction and learning signal, sometimes described as a surprise detector, that compares what you expected to happen against what actually happened. The dopamine loop is what happens when a predictive cue, a notification badge instead of a light, starts triggering that same response on its own, ahead of and independent of whatever's actually behind it.

Schultz's finding is now known as reward prediction error: dopamine spikes when something is better than expected, and it drops when something is worse than expected, or absent entirely. A reward that's fully anticipated barely moves dopamine at all, because there's no error to correct for, no gap between prediction and outcome. The strongest dopamine responses come from uncertainty, not from the reward's actual size.

This is also, not coincidentally, close to how modern machine learning algorithms are built to learn from error. Evolution appears to have arrived at a similar solution to the same problem, optimizing future behavior based on the gap between expectation and result, long before anyone wrote it down as an algorithm.

Once your brain has learned that a red badge sometimes precedes something rewarding, a like, a message, a match, the badge itself starts to function the way Schultz's light did for the monkey. It triggers anticipation and a mild dopamine response before you've opened the app, before you know whether anything good is actually there. The checking behavior is driven by the cue, not by a guarantee of reward.

That's what makes intermittent, unpredictable rewards more compelling than reliable ones. A guaranteed reward stops surprising the prediction system over time, but an uncertain one, will this notification be good or not, keeps the prediction error, and the dopamine response, alive indefinitely. It's the same uncertainty mechanism that makes slot machines and app notifications functionally similar at the level of brain chemistry.

Dopamine's predictive response happens fast and largely outside conscious awareness, closer to a reflex than a decision. By the time you've consciously registered "I should check my phone," the cue has already done its work. That's part of why simply deciding not to check tends to fail in the moment: the urge is generated upstream of the decision meant to override it.

This doesn't mean the loop is unbreakable, it means the intervention works better earlier in the chain, at the cue, rather than later, at the decision. Removing the badge, or the app that generates it, addresses the trigger directly instead of asking a slower, more effortful part of the brain to override a faster one every single time.

Muting notifications helps, but the checking habit often outlives the notification, once the loop is established, people check without a badge prompting them, purely on schedule or boredom. The cue has effectively become internalized, a habit rather than a response to any single external trigger.

Fella keeps selected apps blocked by default, so there's no badge to see and no app to open reflexively, whether the urge came from a notification or from habit alone. One 5-minute emergency unlock a day covers a genuine need, and the block returns automatically when the window ends.