Inconsistent word and character boundaries in ICU - unicode

Question

I'm using ICU's break iterators for both characters and words as described in here. I expect the output of character break iterator stops more frequently and the break-points be a superset of that of word break iterator. For instance, if I pass abc, I get a, b, and c from character break iterator while I get abc from word break iterator.
Now, I have a Thai string as ด้าน้ำ. The problem is that the behavior of these two break iterators are inconsistent. Given the length of the above string is 6 in Unicode, I get these results from ICU 61.1 on MacOS:
Word boundaries:
[0, 5)
[5, 6)
Character boundaries:
[0, 2)
[2, 3)
[3, 6)
As you can see, character break operator breaks the word in [3, 6) (which seems correct), while word break operator breaks it in [5, 6). Here's a small Python3 code which uses PyICU to repro the issue:
import PyICU
def wordBreakIterator():
return PyICU.BreakIterator.createWordInstance(PyICU.Locale("th"))
def charBreakIterator():
return PyICU.BreakIterator.createCharacterInstance(PyICU.Locale("th"))
def printBoundaries(txt, bi):
bi.setText(txt)
start = bi.first()
try:
while True:
end = next(bi)
print("[{}, {})".format(start, end))
start = end
except StopIteration:
pass
if __name__ == "__main__":
text = u'ด้าน้ำ'
print("Word boundaries:")
printBoundaries(text, wordBreakIterator())
print("Character boundaries:")
printBoundaries(text, charBreakIterator())