What happens to a spending habit when you remove the uncertainty at the exact moment of decision? That question sits at the centre of a small but growing body of work on digital reward systems, and a recent analysis of disclosed-odds mechanics in games with randomised rewards offers a striking number: showing the probabilities of outcomes upfront was associated with a 31% reduction in late-night top-ups among players who had previously spent most heavily after midnight. The finding matters well beyond any single product, because it tests a long-held assumption in behavioural design — that uncertainty itself is a core driver of repeated spending, not merely a side effect.

The psychology of the near-miss and the variable reward

B.F. Skinner's work on variable-ratio reinforcement is the obvious starting point. When a reward arrives after an unpredictable number of responses, the behaviour that produces it becomes remarkably persistent — more so than when rewards are predictable. This is one of the most replicated findings in behavioural psychology, and it explains why so many digital products, from social feeds to collectible card systems, are built around randomised outcomes rather than fixed ones.

But the mechanism is often described too loosely. It is not simply that randomness feels exciting. It is that randomness creates a continuous stream of near-misses — outcomes that fall just short of the reward — and near-misses are processed in the brain in ways that resemble actual wins. Brain-imaging work by Luke Clark and colleagues at the University of Cambridge showed that near-misses in a gambling-like task activated reward-related regions and increased the motivation to continue, even though the participant had objectively lost. The near-miss is not neutral feedback. It is a signal to keep going.

This is where disclosed odds become interesting. If a player knows the true probability of a rare outcome, the near-miss loses some of its ambiguity. It is no longer "so close" — it is a statistically expected event within a known distribution. The uncertainty that fuels persistence is partially dissolved by the information itself.

Why late-night spending is a distinct problem

The 31% figure specifically concerns late-night top-ups, and that specificity is not incidental. Decision-making under fatigue is measurably worse. Research on sleep deprivation and risk-taking, including work published in The Lancet and replicated in multiple lab settings, shows that tired brains lean more heavily on immediate reward and less on long-term consequence. The prefrontal systems that normally apply the brakes are compromised.

Layer on top of that the fact that late-night sessions tend to be solitary, unstructured, and emotionally unregulated. There is no colleague to raise an eyebrow, no scheduled obligation to interrupt the loop. The conditions that produce impulsive spending are almost perfectly assembled between midnight and 3am: fatigue, low social friction, and a reward system designed to deliver unpredictable reinforcement.

So when an intervention cuts late-night top-ups by 31% while leaving daytime behaviour relatively unchanged, it suggests the mechanism is not general deterrence. It is specifically interrupting the impulsive decision pathway — the one that dominates when cognitive resources are depleted.

What the disclosed-odds study actually did

The study in question compared spending patterns across two conditions: one in which players saw randomised reward probabilities displayed before purchase, and one in which they did not. The headline result — a 31% drop in late-night top-ups — was concentrated among the heaviest late-night spenders, not spread evenly across the user base. Light spenders barely moved. This is a familiar pattern in behavioural interventions: the people most affected are those whose behaviour was most extreme to begin with.

Crucially, the effect was not driven by players quitting. Overall engagement remained broadly stable. What changed was the timing and size of top-ups, particularly among users who had previously shown a pattern of escalating late-night purchases. The information did not stop people from playing. It changed the quality of the decision they made at the point of spending.

This aligns with Kahneman and Tversky's foundational work on loss aversion and reference points. When probabilities are hidden, the mental reference point is the possibility of a win — a gain frame. When probabilities are shown, the reference point shifts toward the expected cost, and the asymmetry between a small certain loss and a large uncertain gain becomes harder to ignore. The frame changes the decision without changing the underlying odds.

The competitive dimension: skill, status, and sunk cost

There is a second force at work in these systems that has nothing to do with pure chance: competitive play. Many randomised-reward products layer in leaderboards, seasonal rankings, and status markers that reward cumulative spending or time investment. This introduces a sunk-cost dynamic that operates independently of probability disclosure.

Arkes and Blumer's classic 1985 research on sunk cost showed that people who have already invested in a course of action are more likely to continue it, even when continuing is irrational. In a competitive context, that sunk cost is amplified by social comparison. Quitting feels like conceding. The leaderboard does not care that the odds were disclosed.

This is why disclosed odds alone are unlikely to be a complete solution. They address the uncertainty that drives the near-miss loop, but they do not address the status anxiety that drives competitive escalation. An intervention that cuts late-night top-ups by nearly a third is meaningful, but it is addressing one lever among several.

What this means for design and regulation going forward

The practical implication is not that transparency is a silver bullet. It is that transparency at the point of decision — not buried in terms and conditions, not in a help menu, but displayed before the purchase — has a measurable effect on the most vulnerable spending patterns. The 31% figure is large enough to matter and specific enough to guide policy.

The forward-looking question is whether disclosure can be paired with other timing-sensitive interventions. If late-night fatigue is the vulnerability window, then design choices around session pacing, cooling-off periods, and default settings at night hours could compound the effect. The disclosed-odds result suggests that the decision environment matters as much as the decision itself. Change the environment — the time of day, the information available, the friction involved — and you change the behaviour without removing the product.

For anyone studying reward loops and decision-making under uncertainty, this is the interesting frontier. Not whether people can be informed, but whether information delivered at the right moment, in the right form, can interrupt a loop that was never really about the reward in the first place.