As one of those visual observers who used to get accurate estimates when I was active, I can tell you that those points are just the result of reporting your estimates to two decimal places.
It is not something strange. You don’t necessarily use 0.1 mag. steps to estimate variable star brightness.
That would be the Pogson method, but if you adopt the Argelander method, you can just use the number of steps that you consider useful given the brightness differences you observe, and then, knowing the magnitude difference between the two comparison stars (always taken with two decimal places, not one), you get a value for the step that may be anything: 0.1, 0.05, 0.03, etc.
I was a naked eye observer mostly focused on detecting small brightness differences.
I started observing constant stars in order to improve the accuracy of my observations. That’s how I discovered the variability of both delta Scorpii and delta Velorum: they weren’t shining as they were supposed to.
Delta Sco was only 0.09 mag. brighter when I realized something was going on, in June 2000.
After that, my main goal became detecting Be star outbursts and eclipses of systems without known periods 
I succeeded, detecting small amplitude outbursts of ome CMa, mu Cen and other GCAS stars. I also detected eclipses of some of those unsolved EAs and solved them (VZ PsA, KV CMa, V0438 Pup and others).
I attach a light curve of the early years of the del Sco eruption, that combines my observations with photoelectric and CCD observations made by other observers.
In order to detect these small brightness variations, you can’t set your magnitude steps to 0.1 and you can’t use comparison stars that are 0.7 mag. apart like in Jim’s example.
I used a dense set of comp stars spanning all the variable range and with magnitude differences that might be 0.1 or less in some cases.
You first check that you are able to see the differences between all of them, in a kind of “calibrate your eyes” exercise, and then you continue by comparing the variable against all of them until you are sure where it fits, that is, you choose the closest brighter and fainter comp stars.
I tried to have comp stars with similar colors in all sequences, because that minimized the errors. When you have a good sequence, good colors, similar magnitudes, and very close in the sky, you can get amazing results.
I understand that not everyone can get them. Experience, patience, an appropriate technique and good eyesight are required, and that’s where the observer’s responsability comes into play.
Before jumping to observe small amplitude variables visually and report your results to 0.01 mag., you need to prove yourself that you are able to do it. Otherwise you are adding noise to the database…
But we can’t prevent people to submit data like that if they think it is worth it. We might be losing useful information.
Most of my data would be useless if reported to 0.1 mag.
Everyone should select stars for their program that match the precision that they can get.
There will always be a suitable target for everyone.
In the case of del Sco, my comparison star sequence is bet Cru (1.26v), lam Sco (1.62v), eps Sgr (1.84), the Sco (1.86), sig Sgr (2.09), eps Sco (2.29) and alf Lup (2.30v).
(yes, there are some small amplitude variables, because at these magnitudes, it is very difficult to have completely constant stars, it is what it is…)
Naked eye observations have the issue that the comp stars may be spread across large sky areas and you may have to change comp stars depending on the time of the year, because some of them may be too low to be used at times. Even then, you can get very good results.
In short, what you may see in the gam Cas light curve are observations submitted by visual observers with two decimal places because they are using the Argelander method, that allows more flexibility than the Pogosn method.
If that is justified in the case of gam Cas, that’s another story.
gam Cas settled into a quiescent state after its gigantic outburst, and the fact that is behaving like a <0.1 mag. variable, with periodic variations with amplitudes even much smaller than that, makes visual observations not useful to study such behaviour.
However, these are irregular variables. gam Cas might wake up and have another outburst, so keeping an eye on it is important.
I would defer to each observer’s responsability to decide if their estimate precision deserves reporting of results to two decimal places.
From what I see in the light curve, it’s not bad at all.
Actually, if you report a 0.1 mag. variable to 0.1 mag. precision, you are probably just checking that it didn’t go into outburst. I would leave that star to peppers or to people who are confident they can get results better than 0.1 mag.
Well, it was too long a reply, but this has been my life for more than 20 years so I had to speak up!
Good observing.
Sebastian
You would hopefully have a good observer or two who where consistent, were long-term observers of the star, and all others could be judged against. One off observations that were an outlier compared to the good observers could easily be removed. Iterative process, some judgement involved. If you have a lot of data after the observer-based filtering you might be able to do some light statistical filtering.
I’m waiting for the first AI produced “visual” magnitude estimate! 
