I've been trying to develop a query that uses the sack, which seems to require turning off the bulking for accurate traversal merging. If I add the withBulk() step the query returns the following when trying to profile:

{
  "detailedMessage": "When specified, the profile()-Step must be the last step or followed only by the cap()-step.",
  "requestId": "5078e1e8-0671-4a7a-94df-4fead2650e19",
  "code": "InternalFailureException"
}

Update:

I'm not particularly familiar with the airports dataset, but I whipped out this query which seems to demonstrate my problem.

g.withBulk(false).withSack(1.0d,sum).V().has('code','SAF').out().in().barrier()
.project("airport","sack").by(__.identity().values("code")).by(__.sack()).order().by(__.select("sack"),desc)

The withBulk(false) with sack() usage is unfortunately very briefly touched upon in the documentation, so I can't exactly understand what is happening, but my guess is that traversal merging is not performed properly with bulking enabled. If I run the query as is, I get the expected output:

If I remove it I get a lot of duplicates:

As for Neptune actually allowing this configuration, I have no clue, I just noticed the difference which seems to align with the very brief and confusing explanation given in the TinkerPop docs: "There is now a single traverser with bulk of 2 and sack of 1.0 and thus, setting withBulk(false) yields the expected 1.0"