I cannot reproduce the issue reported in #2159 with my test phone.
I can apply the patch to dev, commit 47450c1.
It applies cleanly, builds with no problem and the ICE screens are unchanged.

If there's an additional test that would be helpful, I could try again.
(I ran for about 6 hours with 3 carb entries.)
Do you know of specific configuration I could try to reproduce to make the inconsistency show up?

Based on my understanding of the maths, to reproduce the issue prior to the patch you need:

1. A carb entry where ICE is slightly larger than minimum rate absorption but not hugely larger.
2. To set this up on a test phone you could try adding an insulin entry and have glucose flat. Add a matching carb entry at a similar time. You could enter this at a time close to the earliest time on the portrait view glucose change graph to make subsequent steps shorter.
3. Carb entry should initially show dynamic carb absorption - ie ‘predicted’ curve will follow ‘observed’ curve - on both graphs.
4. Wait until the start time of the carb entry moves off the left of the portrait view graph. At a certain point, the sum of minimum rate absorption over time for the entry will be more than the sum of ICE (because prior to the patch, ICE is truncated at the left edge of the graph). Now the predicted curve will follow minimum rate absorption and be below ICE on the portrait view. But on landscape view, the it should still show dynamic carb absorption (because the landscape view uses a longer series of ICE values).