I did a similar experiment, and I found considerable nonlinearity in the early stages of discharge (first 1 or 2 KWH). What I ended up doing was to collect KWH and % data every KWH or so (it's not critical, just collect a dozen or two data pairs). It's not necessary to go all the way to zero, 10% or even 20% works fine. I only got to 40% before I reached home so I stopped there.
Then enter the data pairs into Excel and plot KWH consumed as a function of %. You can eyeball a straight line fit, see where the extrapolated line crosses the % origin. Better yet, use Excel to do a linear regression from 90% to whatever minimum you reached, and find the intercept.
I plotted from 90% to 40% (which is as far as I got) and the intercept was 19.2 KWH total battery capacity, right in line with our estimates. My eyeball straight line was a little higher, maybe 19.5 This was after SSN fix, so I have no reason to suspect this significantly affects the capacity.