How about batteries that increase in voltage after stopping load?
All batteries do this. A source sags under load. When the load is removed, the battery rebounds.
I have three 12 volt (nominal) LiFePo batteries wired in series. Together they provide 36 volt (again nominal voltage).
12.8*3 = 38.4V
I have no chart for estimating percentage of battery charge remaining for a 36 volt system
But you've discovered you can use the LFP charts for individual 12.8V batteries.
so I disconnect at the middle Anderson connector thus breaking the possibility of the batteries interacting while not in use. I measure each battery separately.
Batteries in series can't interact in any way. There is no need to disconnect. Simply measure each one individually.
These are in a ebike. Batteries were all purchased at the same time and two have sequential serial numbers and the third is 4 batteries away. End of ride voltage 12.95 volts, two minutes later 12.97 volts, overnight batteries measure 13.13 volts. All voltages between batteries are "spot on" the same.
With LFP the bounce is pretty quick, and as you're discovering, it's not a reliable indicator as very small voltage changes result in big swings in SoC.
Generally speaking for LFP, a 5 minutes rest is probably good enough. 30 minutes max. If these were lead acid, you'd want to wait 2 hours minimum, 10+ hours preferred.
Which number is correct for estimating remaining charge?
Since you're choosing to use an inherently unreliable means of tracking SoC, you get to pick. Personally, with something like an e-bike, I would want a conservative value, i.e., give it a 5 minute rest and estimate your SoC.
This is why battery monitors that actually count current in and out are vigorously recommended.