Another thing is why they are using different calculations than yours (i mean they are not doing the Xbar-R chart for this data) but instead only calculating the Std Dev, Xbar bar and straight getting the answer for UCL and LCL with the formula below :
UCL = CL + 3*S
LCL = CL – 3*S
Well the formulas are just the same, but the question here is: What does "S" stand for?
If we say: UCL = CL + Average_Range * A2
So
3S = Average_Range * A2
S= Average_Range* A2/3
S= Average_Range *(3/Sample_Size^0.5/d2)/3
reordering and simplifying:
S= Average_Range /Sample_Size^0.5/d2
being: ds_Within_Sample= Average_Range /Sample_Size^0.5/d2
UCL = CL + 3*ds_Within_Sample
LCL = CL – 3*ds_Within_Sample
There are diferent kind of estimates for different statistical calculus, but SPC uses ds_Within_Sample wich is the variation inside the subgroup
Theorically: ds_Total = ds_Within + ds_between
There are also other kind of charts like "Average and s", but it's another story, they are used for sample sizes >10

, and a big
NO,
they SHOULDN'T using Population Std Dev (N) or Sample Std Dev (N-1), don't believe all that can you seen on the NET, that site is cute, nice displays but WRONG.
One question that arrived to me is: ¿why 4 as sample size?, remember that samples are rational subgroups, not just to use a nice number but by a reason.
ACLARATION TO THE SITE'S CALCULATION:
http://www.wikihow.com/Create-a-Control-Chart
Is wrong in many ways, first the variation estimate,



, it's the stdev of all data (without subgrouping)


Second, the use of colors to make SPC look like a precontrol chart
Third, the use of rules to determine patterns without advice that most of them not always apply, depend on the process and variable behaviour.
But al least, they try to show how and it's a nice looking site.