Not saying it shouldn't, but at times it almost appears that the evidence is forced to fit a preconceived notion, rather than taken at face value...
If you've read math beyond algebra, into functions like y=x^2 or similar, you can picture how it's useful and possible to fit the theory after the data.
Say you have a coordinate grid where you have two points known to exist:
(0,0) (the origin)
(1,1)
To make some useful predictions you assume that there is a function that can explain those two points, you see that you can have any function (there are more functions as well, but lets stick with these for simplicity) of the form y=x^n, where n is bigger than 0.
You can then exclude (under that assumption) that you won't have to look for points beneath the line of y=x (on the positive side of the y-axis) since all other functions diverges faster than y=x, that would save a whole lot of trouble looking where you had the best chance of finding some more points of interest.
Say you spend some time to find additional points, in the predicted area, and you find some:
(2,8)
(3,27)
You can conclude that there is indeed a function that would fit this in the manner you assumed, y=x^3.
During this, you've constructed a model based on available data, searched for more data and made the model more exact to fit the newly arrived data.
There are an infinite amounts of functions that can explain those points but additional points would exclude more and more functions until there are highly unlikely functions (as in very unnatural looking, example below) and, if you're lucky, a couple of easy functions.
Example:
You have gathered a lot of points (101 in this case);
(0,0)
(1,1)
and so on until
(100,100)
There are still an infinite amount of functions that can explain this but I'll focus on the two main contenders:
Function 1:
y = 0 when x = 0
y = 1 when x = 1
y = 2 when x = 2
and so on, each point defined with nothing in between until
y = 100 when x = 100
Function 2:
y = x
I gather more points, I find 100 more points, in between the 101 I already have:
(0.5 , 0.5)
(1.5 , 1.5)
(2.5 , 2.5)
and so on until
(99.5 , 99.5)
This will result in function 1 either scrapped completely, or modified, and function 2 to still work. I choose to rework function 1:
Function 1:
y = 0 when x = 0
y = 0.5 when x = 0.5
y = 1 when x = 1
y = 1.5 when x = 1.5
y = 2 when x = 2
y = 2.5 when x = 2.5
and so on, each point defined with nothing in between until
y = 99.5 when x = 99.5
y = 100 when x = 100
Function 2:
y = x
We can continue this all day long, eventually you'll end up with something that needs to be reworked every time something new shows up or something that is fairly exact and rarely has surprises for you.
This isn't an exact analogy but I hope I helped some
(If I made some mistake, don't hesitate to mention it)