In 20w11a, MC-171079 was supposedly fixed, I assume that is what caused this. From my understanding of MC-171079, it didn't even report a valid bug and now the behaviour of comparators is inconsistent when there are multiple inputs to it.

Reference screenshot:

The bottom shows intended behaviour: A comparator with two inputs, one 14 and one 0, outputs 14. A comparator with inputs 14 and 13 outputs 14 as well (the strongest of the two inputs). And a barrel is shown to be conductive to redstone, because that matters at the top.

The top shows the bug. All containers are empty here and all levers turned on.
The right part is correct, an input signal of strengh 15 causes an output signal 15.
On the left however, an input signal 14 does NOT cause an output 14, but instead 0. If items were in the hopper, the signal strength of that would be output, no matter if it is more or less than 14.
The middle shows that the direct redstone signal from the lever gets ignored, the comparator only looks at the barrel in this case.

If a comparator is put behind a barrel and a barrel with items in it is placed on the other side, the contents of that second barrel are also ignored by the comparator, even though they could be read through a solid block.