how caliper measures code region runtime?

Using caliper instrumentation to measure end-to-end runtime, I am confused by the following 2 cases.

Suppose the program is instrumented as following:

main()
{
cali_begin_string_byname('loop', 'main')
..

cali_begin_string_byname('loop', 'loop1')
loop1 code
cali_end_byname('loop')

...
cali_begin_string_byname('loop', 'loop2')
loop1 code
cali_end_byname('loop')
..
cali_end_byname('loop')
}

After data collection, I process the data with
cali-query -s time.inclusive.duration --table a.cali

My expectation is that aggregation of caliper entry "main" should be the end-to-end runtime.
But, I observe two cases:

1. for some apps, the result is as expected. The entries are like:
   main
   main/loop1
   main/loop2

2. for others, there is no "main" entry, but only entries like
   main/loop1
   main/loop2

I am not sure why I see the differences between case 1 and case 2.
Could anyone explain this?

Start with the shown cali_begin_string_byname and cali_end_byname calls, then reproduce both result shapes using cali-query -s time.inclusive.duration. Check Caliper's documentation for inclusive-duration and nested-region aggregation semantics; done means documenting why the main entry appears in some applications but not others.