This is where this fork departs from the taco it came from. The origin treats an incident as one continuous stretch of work. GitHub's published case shows why that assumption breaks: the trigger lasted 43 seconds and the degradation lasted 24 hours and 11 minutes. No rotation covers 24 hours. Somewhere in there, the people who understood the situation went home and different people took over.
What reliably fails to survive that boundary is not the facts — the timeline and the graphs are written down and anyone can read them. It is the dead ends and the standing decision.
- Dead ends are lost first, because they live in the responders' heads and feel like nothing once you know the answer. The incoming shift then re-runs them, which is the single most common way a long incident wastes hours, and it is invisible in the write-up afterwards because the second attempt looks like diligence.
- The standing decision is lost more dangerously. GitHub's choice not to fail back — accepting a day of degradation rather than risking 30 minutes of writes — is exactly the kind of decision an incoming responder can reverse without realising it was a decision at all. It looks like an unfixed problem. Someone arriving fresh, wanting to help, is precisely the person most likely to undo it.
So a handover note carries three things: what has been ruled out and how, what has been deliberately accepted and why, and what the next person must not do without talking to someone. The third is the one nobody writes, and it is the only one that prevents a well-intentioned reversal at 3am.
The cheap version of this that works: before you hand over, read your own dead-end notes aloud to the person taking the shift. If you cannot, the notes were not written.