Negative differences with no unit

[mpeter50]

I have built flashbanch at commit d783b1b with the make command, and running it as sudo ./flashbanch -a /dev/sda as the readme recommends results in the following:

align 34359738368       pre 986µs       on 1.01ms       post 1.87ms     diff -412331
align 17179869184       pre 833µs       on 1.19ms       post 875µs      diff 335µs
align 8589934592        pre 690µs       on 1.01ms       post 1.44ms     diff -49389n
align 4294967296        pre 1.13ms      on 1.43ms       post 1.67ms     diff 32.7µs
align 2147483648        pre 582µs       on 935µs        post 1.63ms     diff -173123
align 1073741824        pre 1.21ms      on 1.39ms       post 1.61ms     diff -18714n
align 536870912 pre 980µs       on 1.1ms        post 986µs      diff 114µs
align 268435456 pre 650µs       on 1.09ms       post 1ms        diff 263µs
align 134217728 pre 1.53ms      on 1.53ms       post 1.15ms     diff 187µs
align 67108864  pre 3.42ms      on 3.24ms       post 1.97ms     diff 549µs
align 33554432  pre 1.07ms      on 1.18ms       post 1.21ms     diff 43.4µs
align 16777216  pre 1.31ms      on 1.96ms       post 2.01ms     diff 299µs
align 8388608   pre 995µs       on 1.2ms        post 1.01ms     diff 199µs
align 4194304   pre 3.38ms      on 2.87ms       post 1.07ms     diff 641µs
align 2097152   pre 1.75ms      on 1.59ms       post 1.62ms     diff -93720n
align 1048576   pre 1.04ms      on 1.45ms       post 1.44ms     diff 210µs
align 524288    pre 956µs       on 842µs        post 1.56ms     diff -416828
align 262144    pre 1.5ms       on 1.51ms       post 1.14ms     diff 187µs
align 131072    pre 1.07ms      on 941µs        post 1.59ms     diff -387973
align 65536     pre 1.64ms      on 1.75ms       post 1.34ms     diff 256µs
align 32768     pre 1.42ms      on 1.19ms       post 1.7ms      diff -371078

Multiple lines have negative values without a unit in the diff column

Using --count and --blocksize with different values does not help.

I have found a similar report 10 years ago on the linaro-dev mailing list, it did not get responses.

The test was run on a SanDisk Ultra microSDXC UHS-1 SD card, 128 GB in size, which I bought new earlier this year.
By default it had some kind of FAT partition starting at 16 MB.

Did I do something wrong? Should I run it differently?