The RAC expermint: progress report - what is happening? Credit awarding not stable?

Here I am again with another update of the experiment. Some strange things are going on that probably influence the way we are rewarded in ways I did not know off. If you missed it here are the previous posts about it:

First observations of my RAC prediction experiment
Some progress on my RAC estimation experiment
The RAC experiment - progress & request for help

and related: The math behind BOINC RAC

Let's start with the data right away:

There is a clear trend surfacing. Credit reward is not one to one related with the hardware. This is a server running 24/7, not a single task has been aborted and all tasks were confirmed. I didn't expect this to happen and to be straight with you I can only guess as to why RAC/s is fluctuating slowly. We can see that over the coarse of a several days the RAC/s goes from one level to another. The most obvious being around 18-23 September and now lately from 29th September.

The last trend really amazes me. It's a downward trend without me changing anything. The peak RAC was 1491.93 and now the RAC is 1432.92. That's a drop of 59 or around 4%.

Obviously there must be a reason. I think however that I do not have the data at hand to detect a trend and investigate it further. I'm currently thinking of BOINC host normalization.

From the wiki:

The second normalization is across hosts. Assume jobs for a given app are distributed uniformly among hosts. Then the average credit per job should be the same for all hosts.

and

The host normalization mechanism reduces the claimed credit of hosts that are less efficient than average, and increases the claimed credit of hosts that are more efficient than average.

It is especially that last sentence that keeps me wondering. Suppose a big BOINCer leaves the project or joins the project. Could that influence the awarded credit. Let's take a simple example. Only my host is BOINCing on a project. A big fish with more and more powerful hardware joins. It is obvious he should be rewarded more credit. I always thought this would happen because the tasks get crunched faster. But apparently these more powerful hardware is assigned more credit for the same task as well. And my reward goes down. If this big fish leaves the project the reverse happens. I'm now more efficient than average and my reward goes up.

I read something on Slack about @dutch moving his hardware back to VGTU. It could explain the trends though I'm not sure to what extent. I have tried to contact him but up until now he seems to be to busy.

If my assumption is right this does influence the gridcoin rewarding mechanism. It means host with more powerful hardware are rewarded twice. Once for crunching faster and once because of the host normalization. I have no idea how big that 'double' reward might be but on top of that the magnitude will be bigger and the GRC reward will be bigger. This does seem like preferring certain hardware over other hardware more then I originally thought.

Please share with us your thoughts and feedback!

Proud minnow, eager to grow.