Gremlin using select in hasId returns incorrect result

I'm trying to execute a gremlin query where a saved vertex id is re-used later in a hasId clause. What I see is that when I put in the literal Id the answer is correct, however when I substitute the literal for a select('deployable_id') the answer is incorrect. Unfortunately in my real life example I can't put in the literal Id.

I would like to understand why this behavior is occurring, and also if there is a better way of doing this query that avoids this problem.

I am running gremlin against AWS Neptune, however I can also replicate this problem locally using just the gremlin console.

Steps to replicate the problem in gremlin console:

Set up a simple data set

graph = TinkerGraph.open()
g = traversal().withEmbedded(graph)

g.addV('deployable').property('name', 'd1')
g.addV('deployable').property('name', 'd2')
g.addV('library').property('name', 'l1')
g.addV('class').property('name', 'c1')
g.addV('class').property('name', 'c2')
g.addV('app').property('name', 'a1')
g.addV('app').property('name', 'a2')

g.V().has('name', 'd1').addE('ships').to(V().has('name', 'l1'))
g.V().has('name', 'd2').addE('ships').to(V().has('name', 'l1'))
g.V().has('name', 'l1').addE('includes').to(V().has('name', 'c1'))
g.V().has('name', 'l1').addE('includes').to(V().has('name', 'c2'))
g.V().has('name', 'a1').addE('deploys').to(V().has('name', 'd1'))
g.V().has('name', 'a2').addE('deploys').to(V().has('name', 'd2'))

g.V().has('name', 'a1').addE('loads').to(V().has('name', 'c1'))
g.V().has('name', 'a2').addE('loads').to(V().has('name', 'c2'))

Find the id of d1 using this query (it is always 0 as far as I can see)

g.V().has('name', 'd1').id()

Run the query with the literal id (ie the number 0)

g.V().
    has('name', 'd1').
    as('deployable').
    id().as('deployable_id').
    select('deployable').
    out('ships').
    project('library','total_classes', 'loaded_classes').
    by('name').
    by(__.out('includes').count()).
    by(
        __.out('includes').
        where(
            __.in('loads').out('deploys').hasId(0)
        ).count()
    )

This returns the correct result where loaded_classes = 1

==>[library:l1,total_classes:2,loaded_classes:1]

Now run the query which uses the select

g.V().
    has('name', 'd1').
    as('deployable').
    id().as('deployable_id').
    select('deployable').
    out('ships').
    project('library','total_classes', 'loaded_classes').
    by('name').
    by(__.out('includes').count()).
    by(
        __.out('includes').
        where(
            __.in('loads').out('deploys').hasId(__.select('deployable_id'))
        ).count()
    )

This produces an incorrect result where loaded_classes = 0

==>[library:l1,total_classes:2,loaded_classes:0]

The above example does have a solution (__.in('loads').out('deploys').has('name', 'd1')), however this solution also does not work in my real life example, and I am as yet unable to replicate this problem in a simple example.