This is a rather trivial request, but I would imagine trivial to implement as well. I just received a notable question badge on SU and when I look at my question list, it says 3 kviews, even though it only has 2500 views. It looks like you want to limit the view count to 3 characters and 2.5 would fit. So how about formatting kviews as n.n until they reach 10?
I like it. I keep watching the following, expecting that I am close to Notable Question for this question:
I checked and sure enough, a single decimal would fit nicely!
So I think this would be a great feature.
You can mouse over the number and the exact value will appear in the tooltip.
In my opinion I see that counter as a "milestone" meaning if I get a rounding to 2K it means my question has been viewed at least 2'000 times and then the question is in the 2'000 - 2'999 range.
Even mathematically the error could be bigger, in a user point of view, it seems that logic make sense, no?
For reference, these are the current inaccuracies
range : margin of error 0 - 1000 : 0 % 1001 - 1499 : 33 % (at 1499, it is off by 499 against 1499) 1500 : 25 % (reported as 2K, the error is 500/2K) .... : (consistently decreasing) - 9499 : 5.3% (at 9499, it shows as 9.9kV, so the error is 499/9499) 9500 - 10000 : 5.0% (at 9500, error is 500/10K) 10499 : 4.99% 10500 : 5 % .... : (consistently decreasing thereafter)
(max of 33% off). Using x.x (1dp) for between 1k and 10k will reduce the maximum inaccuracy to 4.8% within the 1K-10K range, which is exactly the same as the inaccuracy for 10499 -> 10K.
range : margin of error 0 - 1000 : 0 % 1001 - 1049 : 4.67% (49/1049) 1050 - 1149 : 0 % - 4.76% .... : (consistently decreasing) - 9949 : 0.49% (at 9949, it shows as 9.9kV, so the error is 49/9900) 9950 - 10000 : 0.5 % (at 9500, error is 500/9500) 10499 : 4.75% 10500 : 4.76% .... : (consistently decreasing thereafter)