I like the simplicity of the default assumption that, if an operation is defined in metadata as bound to a type, it can be called on that type (it might return an error, but it's valid to call it).

The question then becomes, if I have an operation that isn't always supported, how is it described? We know how it's advertised in payloads, but that payload needs to point back to metadata to describe parameters, return types, etc.

The action/function advertisement is the pointer to metadata: the advertisement has the form of a URL fragment that identifies the action/function by its qualified name, and optionally by an additional list of non-binding parameter names to identify a specific overload

{
"@odata.context": "http://host/service/$metadata#Employees/$entity",
"#Model.RemainingVacation": {},
...
}

or

{
"@odata.context": "http://host/service/$metadata#Employees/$entity",
"#Model.RemainingVacation(Year)": {},
...
}

Today, by defining a bound action/function you are implying that you can construct the url for invoking the action or function, although the client may get an error if the bound instance is not in a valid state.

If the intention is to call out cases where an instance is in a state where it cannot be invoked, perhaps an instance annotation to call out that the action/function is not supported on a particular instance.

Perhaps we can use the existing syntax but specify a null value:

"#Model.RemainingVacation": null

Resolved as proposed 2016-10-13

Closed 2016-12-01, https://www.oasis-open.org/apps/org/workgroup/odata/download.php/59508/latest/odata-meeting-154_on-20161201-minutes.html