Well, evidence from the delivery threads shows that is not the way Tesla actually allocates and fills windows. For example
Werkelijke wachttijd - Model S ordered Dec 3rd, delivery Mar 17th. And here
Werkelijke wachttijd - Model S is another one who ordered in Jan 13th and got delivered exactly the same day.
First of all data points that are picked up from the delivery monitoring thread
after the fact is not an evidence of how Tesla operates in general, just an evidence how it operated for these two data points. As I am sure you are aware Tesla had some quarters that it operated in a way that maximized deliveries of cars produced within the quarter, which indeed included batching that you've described, but during other quarters, there was no batching and orders were uniformly filled in as they came.
Additionally, Tesla certainly did not operated like that during
all of the period between the updates, i.e. November 11th through January 30th, just during portion of it, as accumulation of the orders was over
before the end of this period on 01/30/16. (see data series below for illustration)
And finally, even when Tesla operates in batching mode, it does not apply to some periods because they fall within the time when Tesla allocates production to the specific region, so there is no accumulation of orders within these periods for this specific region.
So this is, I believe, is one of the problems with your method.
Another one is the fact that you assume that there is information on deducing what waiting time is between the days when projected delivery was updated. You assign a two week period as the end of the delivery window, and then keep that date fixed as time passes after the day of the projected delivery, which in effect shortens wait time as we move in time from one date of the update to the next one.
The problem is that this is a conjecture, and not a real data point. That is why in my method I assume that we do not know what wait time was between the dates the projected delivery was updated, I only consider data points that we have concrete information about - Projected Delivery on the Date of Update only. Than we can draw a trend line between the two consecutive points corresponding to the Dates of Update only. The overall trend line can be added on top of that, for longer periods that include multiple data points for several Dates of Update.
Take the period that we discussing. It starts with the Projected delivery of March on November 11th, and goes on till the next Date of Update on January 30th, with projected delivery date Late april. By assigning two week period as a window for delivery, you are saying that cars entered in production queue after November 11th are delivered by the middle of March, and then, cars entering production queue on and after the date of the next update, January 30th are delivered in the second half of April ("late April"). The problem that this approach results in modeling that assumes that no cars were delivered in Europe in second half of March and first half of April, which is inaccurate.
To contrast this with my method, it is based on assumption that we have knowledge about the delivery estimate only at two dates corresponding to the dates of update: that on November 11th projected delivery was March 1st ("March"), and on Janaury 30th projected delivery was April 16th ("end of April"). Those are the only points with known data, and they are connected by the trend line.
The example of the guy ordering halfway through January, delivered beginning of March shows clearly there is nothing fictitious about that 6 weeks delivery time.
Of course it is not. The 6 week delivery is absolutely certainly
IS fictitious.
First, the example you are referring to - of a guy ordering on January 13th and receiving on March 17th - had 9+ weeks of wait time, not the fictitious 6 weeks.
Second, as you are well aware the in transit/European assembly time is 6 weeks (and you recently even indicated that it is now more like 8 weeks), so how a
TOTAL wait time for European deliveries can be 6 weeks? It is fictitious wait time indeed.
So, in conclusion, because your method assumes that we know what the wait time between the Dates of Update is, it produces inaccurate results, as shown in examples laid out above.
That is why in my method, I assume that we have only discreet data points, indicating estimated delivery on the Dates of Update only,
and, once again, it is based on the following assumptions:
- The wait time is accurate only on the day the update was made
- Reference to a month means the 1st of the month (June means June 1)
- Reference to a "late" month means 16th of the month (Late June means June 16)