GigaBerlin's VINs are sequencially increasing and the subset that are exported to Norway are public (at e.g.
Tesla Registration Stats), so the "German Tank Problem" is perfectly suited to estimate GigaBerlin's production:
en.wikipedia.org
With the expectation that the production rate at GigaBerlin is non-constant, the total production is less interesting than the recent rate of production on e.g. a weekly basis. As such one can base the estimate on the most recent D days - where for those D days the effective maximum observed VIN is the actual maximum observed VIN minus the maximum observed one D+1 days ago.
Because of the random nature of the day when a new maximum VIN is observed, the estimated rate of production is strongly dependent on the choice of D. Further, with a too small D, the fluctuations in the estimate are less likely to reflect actual production rate fluctuations and more likely to be an effect caused by the random nature of the maximum VIN occurrence. Lastly, with a too large D the estimate is at risk of being impacted by actual changes in the production rate during those D days.
To overcome this, one can compute the estimate for all meaningful values of D, e.g. D=7,8,9,... up to some limit where the production rate can be assumed to not have changed (e.g. some time after GigaBerlin's 2nd shift started). One can then make a histogram of the production estimates over D.
Based on this approach (for any histogram bucket size in the range 50 - 150) the average weekly production rate at GigaBerlin in the past 5, 6 and 7 weeks is about 1750.
Some expected, systematic errors in this approach:
1) There is some latency in a VIN being produced at GigaBerlin before it can be observed in Norway - basically the time to transport and register the vehicle there,
2) It is possible that certain production days are allocated to Norway, while others are not - e.g. if a whole train is going to be filled up with Norway exports. In such a case (a "Norway wave") the observed VINs will indicate a fluctuation in production rate which is not real but rather caused by a sampling bias. If this actually happens it seems to be on a timescale not discernible among the noisy samples.