Why does a failure that lands just short of success make people try again, while a failure that lands nowhere near it often ends the session entirely? And why does the time of day seem to change how strongly that effect bites? The answer sits at the intersection of behavioural psychology, decision-making under uncertainty, and the design of reward loops — and it has real implications for anyone building products, training regimes, or competitive formats in the UK.

The anatomy of a near miss

A near miss is a specific kind of negative outcome: you lose, but the feedback tells you that you were almost right. Three matching symbols with the third one just off by a position. A quiz answer that was one word away. A sales call that got to the final objection before falling through.

The critical feature is not the outcome itself — it's the information the outcome carries. A near miss narrows the perceived gap between current performance and success. That narrowing does something unusual: it converts a loss into what feels like evidence of progress.

B. F. Skinner's work on schedules of reinforcement gave us the foundational vocabulary here. Variable-ratio schedules — where a reward arrives after an unpredictable number of responses — produce the highest and most persistent response rates of any schedule he tested. Pigeons pecking for unpredictable food rewards kept going long after fixed-interval schedules would have produced pauses. The unpredictability itself is the engine.

But near misses add a second layer. Research by Monica Wadhwa and colleagues, published in the Journal of Consumer Research, found that near misses increase motivation specifically because they trigger counterfactual thinking: the mind automatically simulates the version of events where the outcome flipped. That simulation is effortful but energising. It makes the next attempt feel more likely to succeed, not less — which is a cognitive illusion, but a remarkably durable one.

Why "three tries" matters

The "three tries" element in the pattern is not arbitrary. Three attempts sits in a sweet spot between two failure modes:

• One try produces no learning signal. The feedback is too thin to generate a counterfactual.
• Two tries generates a signal but not yet a pattern. The person can still attribute the first failure to noise.
• Three tries generates a trend. By the third attempt, the person has enough data points to construct a narrative about their own improving performance — even when the underlying odds haven't changed at all.

This is where Kahneman and Tversky's work on the law of small numbers becomes relevant. People routinely extract confident patterns from tiny samples. Three attempts is exactly the sample size at which a random sequence starts to look like a trajectory.

The persistence effect compounds because each near miss resets the counterfactual simulation. The person isn't persisting because they've forgotten the previous failures — they're persisting because each failure has been re-encoded as partial success.

The 9pm effect: decision fatigue meets reward sensitivity

The time-of-day component is the part most people miss. Persistence isn't a stable trait; it fluctuates with cognitive resources.

By late evening, two things have typically happened. First, executive function — the system responsible for inhibiting impulses and weighing long-term consequences — is depleted. Roy Baumeister's ego depletion research, despite later replication debates, established a robust general finding: self-regulatory capacity declines across a demanding day. Second, and more importantly for this discussion, the salience of immediate reward increases relative to delayed reward.

Kahneman's dual-process framing helps here. System 2 (deliberate, effortful, slow) is what tells you "this is a random outcome, stop." System 1 (fast, associative, automatic) is what responds to the near-miss feedback with "you're close." At 9pm, System 2 is tired. System 1 is not.

There's a third factor specific to evening hours: the social context of persistence changes. Late evening activity is typically solitary or low-accountability. There's no colleague to say "you've had three goes, let's stop." The feedback loop runs without external interruption.

What the research actually shows

A widely cited study by Luke Clark and colleagues at the University of Cambridge, published in Neuron, used functional imaging to examine near-miss responses. Participants who experienced near misses showed increased activation in the ventral striatum — a region associated with reward processing — and reported higher motivation to continue, despite the objective outcome being identical to a clear loss. Crucially, the effect was stronger in participants who scored higher on measures of problem-gambling severity, suggesting the near-miss response sits on a spectrum rather than being a universal constant.

That spectrum matters. The near-miss effect is not a bug that affects only "vulnerable" people. It's a normal feature of human reward processing that varies in intensity.

Designing for persistence without exploitation

Here's where the bridge to practical application becomes genuinely interesting — and where the ethical line needs drawing carefully.

The near-miss effect is a mechanism. Mechanisms are neutral. What matters is the intent behind their deployment and the interests they serve.

Legitimate uses of near-miss feedback:

• Skill acquisition. A music practice app that shows you "you were 0.2 seconds off the tempo" is giving you actionable near-miss feedback that genuinely accelerates learning. The gap is real and closeable.
• Fitness progression. "You were two reps short of your target" is a near miss that reflects actual physiological adaptation. The next attempt really is more likely to succeed.
• Competitive play. In ranked multiplayer formats, close losses drive rematches because the skill gap is real and the outcome is genuinely uncertain. This is why best-of-three formats feel more satisfying than single games — they generate meaningful near-miss information.

The critical distinction: near-miss feedback is honest when the gap it describes is real and closeable, and manipulative when it describes a gap that doesn't exist or can't be closed through effort.

A slot machine's near miss is manufactured. The reels are independent; the third symbol being "one off" carries no information about the next spin. A tennis player losing 6-4, 5-7, 6-4 has received genuine information about a closeable gap. The feedback looks similar. The underlying reality is completely different.

The 9pm design question

If you're building anything with a persistence loop — a learning platform, a training app, a competitive game — the 9pm data should be telling you something uncomfortable. Your users are most persistent when their judgement is weakest. That's not a feature to exploit; it's a risk to mitigate.

Forward-looking design responses:

1. Make the gap real. If you show near-miss feedback, ensure it corresponds to an actual, measurable, closeable difference. Manufactured near misses are the line you don't cross.
2. Instrument for time-of-day effects. Track persistence curves by hour. If your 9pm persistence is double your 3pm persistence, ask whether the additional attempts are producing better outcomes or just more attempts.
3. Build in natural stopping points. Three attempts is a natural unit. A format that says "best of three" gives the user a legitimate exit without feeling like they quit.
4. Surface the counterfactual honestly. "You were close" is motivating. "You were close, and here's exactly what was different" is motivating and informative. The second version respects the user's System 2 even when it's tired.

Where this goes next

The interesting frontier isn't whether near-miss feedback works — the evidence is clear that it does, and that it works harder in the evening. The frontier is building persistence loops where the persistence is earned: where the third attempt genuinely carries a higher probability of success than the first, and where the user at 9pm is making a slightly worse decision about a genuinely better opportunity.

That's a design problem, not a psychology problem. The psychology is settled. What remains is whether the people building these systems choose to align the feedback with reality — or to manufacture the feeling of closeness where none exists. The 9pm data will tell you which one you've built.