From the questions I got at Ordina I understood that we need to clarify our view on unknown data. First, you need to recognize that there are three types of time in Anchor Modeling.
Changing time (others: valid time, effective time)
The time when entities change states, attributes
change values or relationships change members.
Recording time (others: transaction time, assertion time)
The period of time during which information about
the domain was recorded in some kind of memory.
Happening time (others: user-defined time)
The time of an event taking place in the domain
When you stumble upon something that you believe is transitioning from having a value to becoming unknown, the way you model it depends on how you use this information. If it is of importance to track these changes, for example that a user can read out the string 'Unknown' for these values, then unknown values are part of your domain and should be modeled explicitly (as a string in an attribute or knot). The time when a value transitions to 'Unknown' is then changing time.
If there is no value in being able to track unknown data, then it should for the periods of time that it is unknown, not be a part of the domain. In this case, unknown data is represented by gaps in recording time. If you are looking at your data during such a period, that information is "not there", tucked away somewhere and not touched unless you rewind recording time. Note that rewinding is only possible in bi-temporal AM.