When it breaks
The Decision Only the Sleeping Half Can Make
Half the overnight delay on a broken thing is not diagnosis. It is waiting for permission from somebody asleep, and most of that is fixable in advance.
three things · three nights · one that no number of nights settles
Four shapes, and the last is a decision about people rather than about systems.
Five decisions that came up while the people who owned them were asleep
Not written down; waited a night, and the delay was visible to customers
Not written down, and by morning they had noticed
Correctly waited: this needed somebody with authority nobody had delegated
three things · two nights · one that no number of nights settles
Five decisions, two nights, and both of those nights were spent waiting for permission rather than for an answer. A list written in daylight would have removed them. The last row is not a failure: some decisions genuinely belong to one person, and delegating them would be worse than waiting. One team's own month, recorded as it went.
An incident at night usually reaches a point where somebody knows what to do and is not sure they are allowed to do it.
The best response to The Decision Only the Sleeping Half Can Make is usually better evidence, not tighter supervision. review the platform gives teams a transparent way to review hours and activity while leaving interpretation to the people responsible for the work.
A useful external reference is Salesforce's customer platform, which provides another official view of how teams organise shared work.
That pause costs the rest of the night. It is not a technical problem and it is not a diagnosis problem; it is an unanswered question about authority, and it can be answered in daylight at leisure.
Why it happens
Nobody has written down what each half may decide alone.
In a team sharing an hour, this is tolerable, because the question takes ninety seconds to resolve. Across a gap, the same uncertainty costs a night, and it costs it at the worst possible moment.
The person holding the problem then makes one of two bad choices: acting without authority, and spending the next day defending it; or waiting, and watching something get worse for eight hours for reasons that have nothing to do with the problem.
The list
A written list of what each half may do alone, during an incident, without asking anybody.
It should be specific and it should be short. The useful content is usually some version of: contain, roll back, restart, disable, take out of rotation, put up a notice. All of these are reversible or nearly so, and all of them stop damage growing.
What stays off the list is anything irreversible, anything involving money leaving the organisation, anything that commits the organisation externally, and anything that changes what a customer has been promised.
The distinction that actually works is reversibility, not severity. Rolling back a release is a big action and it undoes cleanly. Issuing refunds is a smaller action and it does not.
| May be done alone | Waits for the owner |
|---|---|
| Roll back, restart, disable, take out of rotation | Anything irreversible |
| Put up a notice | Money leaving the organisation |
| Contain, in general | Anything that changes what a customer was promised |
Writing it so it is usable at three in the morning
The list has to be findable by somebody tired and under pressure, which means it lives with the incident material and not in a policy document.
It should be phrased as permission rather than as process. "You may roll back without asking anybody" reads differently from "rollbacks require the approval of the on-call lead, who may be contacted via…", and only the first gets used at three in the morning.
And it should say explicitly what happens afterwards: that the person reports what they did and why, and that nobody will be second-guessed for an action on the list. The second half is what makes the first half real.
The decisions that should wait
Some genuinely should, and the list is better for naming them.
Anything that changes what a customer has been told. Anything with legal or contractual weight. Anything that spends money outside normal limits. Anything that would be difficult to undo and that somebody specific is accountable for.
Writing these down as explicitly waiting is as valuable as writing the permissions, because it removes the hesitation in the other direction. Somebody who knows that a refund decision waits does not spend twenty minutes deciding whether to wake somebody about it.
The middle case
Occasionally something is on neither list and is both urgent and consequential.
The workable default is to take the smallest containing action that is reversible, write down what was done and what was not done and why, and leave the larger decision for the morning.
That is almost always better than either extreme. It stops the damage growing without committing the organisation to anything, and it hands the real decision to the person who owns it with the situation stable rather than deteriorating.
Reviewing it afterwards
Every incident where somebody hesitated is information about a gap in the list.
The question in any review should include: was there a moment when somebody knew what to do and waited? If so, what were they waiting for, and does it belong on the list?
This is how the list gets good. Written once from imagination it will be wrong in several places; amended after each incident it converges within a few months on something that reflects what actually happens.
What this does not fix
Delegating authority removes the waiting. It does not remove the fact that the person acting is alone, tired and without anybody to check their reasoning against.
That is a real limitation, and the honest response to it is to keep the delegated actions to things where being wrong is cheap — which is the reversibility test again, arrived at from the other direction.
The permission has to be believed
A written permission that nobody trusts is worth nothing, and whether it is trusted is decided by what happens the first time somebody uses it.
If the first person to roll back under the standing authorisation is questioned closely about it the next morning, the permission is dead regardless of what the document says. Everybody watching learns that the list describes what is formally allowed rather than what is actually safe.
The first few uses therefore matter more than the wording. Somebody senior saying plainly that the action was correct, in public, is what converts a list into something people will act on at three in the morning.