The hotel night shift has a quality that makes it hard to plan: for most of those hours nothing happens, and in the few moments when something does, the person is alone. A plan built on the average of those hours is systematically wrong, because the average describes neither state well.
It is not a longer reception shift
Whoever covers the night is doing three different jobs that during the day belong to three people.
There is reception proper: late check-ins, calls, requests from rooms. There is the night audit, administrative work with a precise time window that cannot be deferred. And there is responsibility for the building: security, alarms, emergencies, the decision a duty manager would take in daylight.
Three jobs with different skills resting on one person. It is why the night shift cannot be covered by whoever happens to be free: it is covered by whoever has been trained for it, and that narrows the assignable pool far more than most rotas admit.
The constraint that breaks the week is the next day
The recurring mistake is not in the night itself, it is in what surrounds it.
Someone coming off at 07:00 is not free: they are asleep. Their rest falls while everyone else is working, and a plan that gives them an afternoon shift the same day leaves them a few hours of broken sleep. On paper the limits hold, in practice they do not.
The opposite rule to retail applies: whoever works the night does not work the following afternoon, and a run of nights closes with a genuine block of rest. The statutory thresholds, tighter than most people remember, are in night work, and the gap between shifts is in rest between shifts.
How many nights in a row
There is no number that works everywhere, but there are two extremes that both work badly.
Scattered nights, one here and one there, are the worst of all: the body never adapts and every night is like the first. It is the configuration that produces the most night audit errors.
Long runs, six or seven consecutive nights, are better tolerated in the moment but need an equally long recovery, which almost no rota genuinely allows for.
Short ordered rotations, three or four nights followed by a block of rest, are the compromise that holds up best. The general reasoning on patterns is in 6x2 and 5x2 rotations.
The night count has to be kept
The last point, and the most neglected. When nights always land on the same two people, the reason is almost never favouritism: it is that they are the only ones trained, or the only ones who do not push back.
A per-person counter, visible to everyone, changes the conversation. And it surfaces the real problem, which is nearly always training: if only two people can cover the night, you do not have a fairness problem, you have a continuity problem.
In Sked Solve the per-person night count and the rules about rest after a night shift are constraints on generation rather than checks afterwards. See how it works at Sked Solve.