It seems whenever busybox grep -w finds the pattern as a strict substring of a word but then sees that it's not enclosed by symbols that delimit words, the whole line is discarded. As a result, grep won't find the pattern when it occurs as a word later in the line, and hence won't match the line.

Examples:

# the good cases that work as expected:
echo "a" | busybox grep -w a
a
echo "b,a" | busybox grep -w a
b,a

# the bad cases which should match but doesn't:
echo "ab,a" | busybox grep -w a
<no match>
echo "ba,a" | busybox grep -w a
<no match>
echo "bab,a" | busybox grep -w a
<no match>

Shouldn't `grep -w ${X}` be equivalent to `grep -e "\<${X}\>"` (the latter works fine in busybox's grep)? At least that seems to be the case in GNU grep, and is explicitly the case in BSD grep according to its manual.