Below is a subpart of a response query via the rollup endpoint, where I have just replaced the keys with a random value. The sum between the childs' doc_count and the sum_other_doc_count is by far not consistent with parent's doc_count and at the same time the parent's doc_count is consistent with the expected count.
{
"key_as_string": "2018-06-01T00:00:00.000Z",
"key": 1527811200000,
"doc_count": 1786083,
"seriesDoi": {
"doc_count_error_upper_bound": 0,
"sum_other_doc_count": 554401,
"buckets": [
{
"key": "key1",
"doc_count": 5
},
{
"key": "key2",
"doc_count": 5
},
{
"key": "key3",
"doc_count": 4
},
{
"key": "key4",
"doc_count": 2
},
{
"key": "key5",
"doc_count": 2
},
{
"key": "key6",
"doc_count": 2
},
{
"key": "key7",
"doc_count": 2
},
{
"key": "key8",
"doc_count": 2
},
{
"key": "key9",
"doc_count": 1
},
{
"key": "key10",
"doc_count": 1
}
]
}
}