Object encoding not used for nested query arguments

[HawkOnPK]

We have a nested graphql query

steps: [
  {timeframe: datetimerange}, 
  {
    output: [
      {timeframe: datetimerange},
      {outputid: true}
    ]
  }
]

where the datetimerange of type DateTimeRange is registered as a scalar with its own encoding function.

The encoding function is called correctly for the first "timeframe" argument, but is never called for the second, for which none of the encodings are called, even though it is of the same type (even same object) as the first argument.
The first argument is thus correctly encoded into a json, the second instead appears as (empty) fields of the base class {_lower: {},\n_upper: {},\n_bounds: \"[)\",\n_empty: false}.

If only the first argument is given and the second skipped, it works.
If only the second argument is given and the first skipped, it works as well.
It does not differ, whether the values given to the two "timeframe" arguments are the same object of different ones.

[HawkOnPK]

Further observation:

The DateTimeRange contains MyDateTime as class HawkDatetime extends Date for lower and upper.
The following query also fails, since "testdt" is not encoded in the nested "output" query, but is successfully encoded if used in the outer "steps" query. Still, if the "timeframe" argument is skipped in both places and the "testdt" argument is instead used in both places it works.

steps: [
  {timeframe: datetimerange}, 
  {
    output: [
      {testdt: MyDateTime.now},
      {outputid: true}
    ]
  }
]

Calling the filter arguments in order steps.timeframe and output.testdt does not work. But steps.testdt and output.timeframe does work.

Since the Date object is then encoded as {} and the range as {_lower: {},\n_upper: {},\n_bounds: \"[)\",\n_empty: false}, it looks like zeus is exploring the object and recursively its fields without regard to the schema.