What if the moment that most reliably improves a risky decision isn't a warning, a cooldown, or a bigger disclaimer, but a small, honest acknowledgement that you got the last three calls right? That's the question behind a striking internal finding doing the rounds among product and behavioural teams: when users were shown a skill badge every third decision they made, the rate at which they took high-variance, low-information options fell by 31% by late afternoon. The mechanism isn't magic and it isn't moralising. It's feedback architecture — and it sits at the exact seam where behavioural psychology meets interface design.

The afternoon problem: decision fatigue meets loss aversion

Anyone who has spent a working day making repeated choices under uncertainty knows the pattern. You start sharp. By mid-afternoon, you're pattern-matching on fumes.

This isn't a metaphor. The research on ego depletion is contested, but the broader finding — that the quality of decision-making degrades across a long session of repeated choices — is robust enough to design around. Kahneman and Tversky's work on loss aversion established that losses loom roughly twice as large as equivalent gains, which means that as the day wears on, the psychological cost of a bad call rises while your capacity to evaluate the odds of one falls.

The result is a predictable drift. Early in a session, people evaluate options on their merits. Later, they gravitate towards choices that feel decisive — high-variance, big-swing, low-information picks that offer the emotional relief of action rather than the discipline of analysis. The "risky picks" in the 31% figure aren't necessarily reckless in absolute terms. They're just worse than the same person would have chosen at 10am.

The badge intervention targets exactly this drift, and it does so without ever telling the user what to choose.

Why a badge every third decision outperforms a warning

The instinctive fix for deteriorating decisions is a nudge: a pop-up, a red banner, a "are you sure?" modal. These fail for well-documented reasons. Warning fatigue sets in fast. More importantly, warnings are negative signals that arrive after the decision has already been framed, and they implicitly position the user as someone who needs correcting.

A skill badge does something structurally different. It's a positive, earned signal, and — critically — it's delayed. Not after every decision, but after every third.

That delay matters more than it first appears. Behavioural research on variable-ratio reinforcement, most famously associated with B.F. Skinner's work on operant conditioning, shows that unpredictable reward schedules produce the most persistent behaviour. A badge that arrives on a fixed, predictable cadence (every third call) sits in a useful middle ground: frequent enough to feel connected to your behaviour, irregular enough that you can't game it, and — crucially — not so frequent that it becomes wallpaper.

Consider the difference in what each signal teaches:

• A warning says: you were about to do something wrong. It's retrospective and it's aversive.
• A badge says: your recent judgement has been sound. It's retrospective too, but it's affirming, and it reframes the next decision as a chance to maintain a standard rather than avoid a punishment.

The second framing is what changes the arithmetic. Under loss aversion, the fear of losing an established streak of good judgement is a far stronger motivator than the fear of an abstract bad outcome. You're not protecting yourself from a loss. You're protecting a record.

The three-decision window is doing real work

Why three? Because one decision is too noisy to mean anything, and a badge every single time collapses into a participation trophy. Three is roughly the minimum number of choices needed for a person to perceive a pattern in their own behaviour — enough to feel like the badge reflects something real about how they've been thinking, not just that they showed up.

It also creates a natural rhythm. You make a call, you make another, you make a third, and then you get a small signal that says: that run was good. The fourth decision now begins from a slightly different psychological starting point than the first did.

What the 31% actually represents

The headline number — a 31% reduction in risky picks by 4pm — is best understood not as a measure of caution but as a measure of consistency. The people in the intervention group weren't choosing more conservatively than their morning selves. They were choosing more like their morning selves.

That's the genuinely interesting part. The badge didn't make anyone more timid. It reduced the gap between someone's best judgement and their late-afternoon judgement. In decision-science terms, it shrank the within-person variance, which is usually a far more valuable outcome than shifting the average.

There's a useful parallel here with research on implementation intentions — the finding, associated with Peter Gollwitzer, that people follow through on goals far more reliably when they've pre-specified the when and how rather than just the what. A skill badge functions as a lightweight, recurring implementation intention. It doesn't tell you what to do. It quietly reminds you that you have a standard, and that you've been meeting it.

Designing the signal so it doesn't backfire

Not every badge works. The 31% figure comes from a specific configuration, and it's worth being precise about the conditions that made it hold.

It must be earned, not given

A badge that appears regardless of behaviour is noise. The effect depends on the user believing, at some level, that the badge correlates with something they actually did. This is why the intervention measured decision quality rather than mere activity — a badge for "you made three decisions" would be worthless.

It must never lecture

The moment a badge carries an implicit instruction ("great job avoiding risk — keep it up!"), it converts from affirmation into direction, and the user's defences go up. The strongest version is almost affectively flat: a clean, quiet marker that says three in a row, and nothing more.

It must not be farmable

If users can trigger badges by making trivial choices, the signal decouples from the behaviour it's meant to reinforce. The three-decision window helps here, but the underlying quality metric has to be robust enough that it can't be gamed by low-effort activity.

Where this points next

The most interesting implication isn't about any single product. It's that the timing and framing of feedback may matter more than the content of the feedback itself. We've spent years building better warnings, clearer disclosures, and more explicit risk communication — all of which assume that people make poor decisions because they lack information. The badge finding suggests something adjacent and more useful: people often make poor decisions because they've lost the thread of their own competence, and a small, well-timed reminder of that competence is enough to restore it.

The obvious next experiment is cadence. If three decisions is good, is five better? Does the optimal window shift with session length, or with the complexity of the choices involved? And does the effect hold when the badge is purely private versus when it's visible to others — because competitive framing, as the research on social comparison keeps showing, can cut both ways.

For anyone designing systems where people make repeated choices under uncertainty — trading interfaces, triage tools, editorial queues, anything with a long session and a real cost to bad calls — the practical takeaway is worth acting on now. Stop treating late-session decision quality as a fixed trait. It's a variable, and a modest piece of feedback architecture appears to move it by nearly a third. The question isn't whether to build the badge. It's what your version of "three good calls" looks like, and whether you can measure it honestly enough to mean something.