I think though, if you are looking at a model to predict how a team will score the 70/71/72 points or whatever for top 4, it doesn't need to be as realistic in a prediction sense, as its simply not possible due to the massive role that luck plays in single results and seasons as a whole, and the further down the table a team is, the bigger role luck plays. So for me, as long as its logical, its fine.
Keeping with the golf analogy, and you may have said something similar in these threads but I've only skimmed back and from what I've seen you've covered all bases, but the par score of a hole is not actually the average score of the hole. Some par 4s will have an average of say 3.8, and others will have an average of 4.3, but a par 4 for both still makes sense. So in terms of teams finishing in 4th, My guess is that history will show that from the bottom 3, although par is 18 points, the average might be 15ish, from mid 10, whilst par is 48 points, the average will be around 42ish, and from the top 7, whilst the par score is 5, the average is 14ish. But to attempt to predict where points will be dropped/gained would not make any logical sense and would be futile attempting to do so, so something that simply makes sense is fine.
I've had a bit to drink, so not sure if I've made sense there
As a rule of thumb...the more simple the 'model'...the more useful it is for general application.
The beauty of Prof's model is that it doesn't attempt to pick when/ where/ against whom a side will drop points...the model would need to be so complicated...so sophisticated...that ya average bear couldn't follow it.
Besides...each factor you built into a model would be subject to error at every point.
I'm an engineer...we're the kinds of guys who build engines...get men to the moon...
If you're designing an engine...it's 'possible' to build a mathematical model...which could determine the Vector of every single atom inside the chamber at every point in time...but we don't do that.
We make assumptions that a single point remains constant over a given time...then just work out what goes in...what comes out.
Prof's model is a bit like this way of working...all the various/ individual 'factors' which have a bearing on performance are ignored...and we just take as our 'point in time' the status of current form at this moment [after each round of games].
In the early part of the season [say week one] you only have a small amount of data to build the trend line with.
The data which goes in week by week is 100% correct and becomes more reliable to build the trend line on...
And it is only about 'trends'...not hard n fast 'predictions'.
I estimate that by week 25...it's about +/- 3% accurate.