The 9pm decision-maker and the noon decision-maker are, in theory, the same person. Same risk appetite, same bankroll, same knowledge of the odds. Yet when researchers simulate a fixed deadline — a hard stop at which the session ends no matter what — the evening group holds a losing position roughly 38% longer than the midday group before cutting it. That gap is the interesting part. It suggests the clock doesn't just limit how long we play; it changes how we think about a loss we're already sitting on.

The question worth pulling apart is why a boundary that is identical in both conditions produces such different behaviour. The answer sits at the meeting point of behavioural psychology, decision-making under uncertainty, and the specific reward architecture that makes stopping feel like a loss in itself.

The deadline paradox: constraints that should help, but don't

A loss cap is a pre-commitment device. You set it before you're emotionally invested, which is exactly when your judgment is cleanest. Kahneman and Tversky's work on prospect theory tells us that losses loom roughly twice as large as equivalent gains — loss aversion — and that we become risk-seeking when facing a certain loss. A cap is meant to interrupt that risk-seeking spiral by removing the choice from your future self.

Behaviourally, that should work. And in laboratory conditions with small stakes, it often does. The puzzle is the 38% figure: the cap is present in both arms of the comparison, so the cap itself isn't the variable. The time of day is.

What changes at 9pm is not the rule but the person applying it. Three mechanisms do most of the work here.

Ego depletion and the evening self

The self-control literature — Roy Baumeister's ego depletion model, though it has taken deserved hits on replication — still describes something most people recognise: deliberate restraint is tiring. Across a day you make hundreds of small choices to stop, defer, or resist. By 9pm you are running on a depleted decision-making budget. The cap is a rule, but enforcing a rule requires the same executive function you've been spending since morning.

Sunk-cost reasoning gets louder at night

The sunk-cost fallacy is the tendency to continue investing in a failing course because of what you've already put in. It's not irrational in every context — persistence pays in plenty of domains — but in a fixed-odds environment it's pure noise. Crucially, sunk-cost reasoning is narrative: it needs a story about how the loss will be recovered. Late in the day, when you're tired and the session already feels "wasted," that story writes itself more easily.

The reward loop is still running

Variable-ratio reinforcement — the schedule where rewards arrive unpredictably after an average number of responses — produces the most persistent behaviour of any reinforcement pattern, a finding that goes back to B.F. Skinner and has been robust ever since. The key detail for our purposes: the anticipation of an unpredictable reward is itself motivating, and that anticipatory drive does not decay neatly with fatigue. It can, in fact, feel sharper when other sources of stimulation have thinned out. At 9pm there is less competing for your attention, so the loop has a clearer channel.

What the 38% actually measures

It's worth being precise about what "holding 38% longer" means, because the number is easy to misread.

It does not mean the evening group lost 38% more. It means the duration between the point at which a rational actor would have exited and the point at which they actually did was 38% longer. The extra time is the dependent variable. The money is a downstream consequence.

This distinction matters because it reframes the problem. We tend to think of late-night risk-taking as a matter of appetite — people want more risk at night. The data suggests something subtler: people want the same outcome but are willing to wait longer and tolerate more variance to avoid crystallising a loss. That's loss aversion expressing itself as time, not as stake size.

A useful parallel is the "disposition effect" in retail investing, documented by Hersh Shefrin and Meir Statman: investors hold losers too long and sell winners too early. The behaviour is remarkably stable across markets and cultures, and it intensifies under conditions of fatigue, time pressure, and emotional load. The evening window stacks all three.

A concrete illustration

Consider a controlled task used in decision-making research: participants are given a fixed number of rounds and told the session ends at a set time. They can stop at any point and take the current balance, or continue. In the midday condition, most participants stop within a few rounds of hitting a predetermined loss threshold. In the evening condition — same task, same threshold, same instructions — a substantial minority keep going well past it, often until the hard stop removes the choice for them.

The behaviour isn't reckless in the moment. Each individual decision is defensible: one more round, the odds are close to even, the loss is recoverable. What fails is the aggregate: a sequence of locally reasonable choices produces a globally poor outcome. This is exactly the pattern Kahneman describes in Thinking, Fast and Slow — System 1 keeps making plausible micro-decisions while System 2, the deliberative check, is offline.

Why the boundary time matters more than the boundary size

If the mechanism is depletion plus narrative plus an undecayed reward loop, then the size of the cap is less protective than we assume. A generous cap at 9pm may be weaker than a modest cap at noon, because the cap only works if the person enforcing it has the executive capacity to do so.

This has an uncomfortable implication for anyone designing self-imposed rules: timing is a control variable, not a backdrop. Most people set limits in terms of amount — a number they won't cross. Far fewer set them in terms of when the decision gets made. Yet the evidence points to the timing being the more load-bearing constraint.

There's a competitive dimension too. When outcomes are observed by others — leaderboards, shared sessions, any social comparison — the pressure to persist past a sensible exit rises, and it rises most in the evening when self-presentation concerns are less buffered by fatigue. Competitive play and risk-taking share circuitry; the same dopaminergic anticipation that drives one drives the other.

What to do with this

The practical move is not to try harder at 9pm. It's to move the decision earlier.

• Set the stopping rule at noon, in writing, with a specific clock time. Not an amount — a time. The rule should be enforceable by a mechanism that doesn't require your evening self's cooperation.
• Treat the evening as a known-bad decision window. If you know depletion degrades your enforcement, schedule the high-stakes choices for when you're fresh. This is standard practice in aviation and surgery; there's no reason it shouldn't apply here.
• Separate the loss from the recovery narrative. The sunk-cost pull is narrative-driven. Naming it explicitly — "I am continuing because I want to undo a loss, not because the odds changed" — restores a little System 2 access.
• Watch the anticipation, not the outcome. Variable-ratio schedules keep behaviour alive through anticipation. If you notice the pull is strongest when nothing is happening, that's the loop, not the opportunity.

The forward-looking point is that we're likely to get better at measuring this. Wearables, session telemetry, and passive behavioural logging now make it feasible to identify your own personal depletion curve — the hour at which your enforcement reliability drops. The next useful step isn't a better rule. It's knowing when you can still keep one.