ADB Nexus draws a live RAM graph from cat /proc/meminfo on the connected phone. Two kinds of bad input are possible there, and the parser handles them in opposite ways. MemAvailable missing is normal on old kernels (the field arrived in Linux 3.14). The app estimates it from MemFree + Buffers + Cached + SReclaimable - Shmem. MemAvailable present but larger than MemTotal means the output is corrupt, so the sample is dropped. One unit test feeds MemTotal 10 kB and MemAvailable 11 kB and expects nothing back. The easy version falls back to MemFree in both cases. Then garbage from the device draws as a believable line and nobody questions the graph. If your fallback fires on impossible values as well as missing ones, your dashboard can't tell a bad reading from a real one.