The last twenty minutes of a session look the same almost everywhere.
To see why, we can start at the other end. Pick a random moment from the first twenty minutes of a session and there’s about a 10% chance the table is in combat. Move that moment later into the night and the odds climb, steadily, until about forty minutes before the end, where the odds peak at 28%. In the last twenty minutes they fall back to 18% (queue session scheduling?).
How high the odds climb depends on how long the night runs. Two hour sessions peak around 25%. Four hour sessions peak around 34%.
Longer nights start slower too.. in the opening minutes of a five hour session the odds of a fight are under 6%, in a two hour session about 11%. So the length of the night changes where it starts and how high it climbs.
It doesn’t change where it lands. In the last twenty minutes the odds sit between 16% and 20% whether the session ran ninety minutes or five hours. Longer nights open slower and peak higher, but every length of night winds down to the same place.
One more thing happens across the night.. the talking slows down. Tables run at about 120 words a minute in the first hour and about 110 past the five hour mark, and the decline is gradual the whole way, still going while the odds of a fight are still rising. Fewer words, more dice?
We can’t see from a transcript whether the last twenty minutes are the DM landing the plane or the table running out of evening. We can see that it happens on the short nights and the marathons alike (and then everyone goes home).