Grit,
I did a little more work on your ladder target. It is usually a bad idea to throw away a data point, such as your shot 14, without a very substantial justification, and this analysis winds up showing why or simply that it didn't help to do so.
What I did first was take the camera angle distortion out of your target in Photoshop. I then pasted that into a CAD program and used the measuring tape in the image to scale a pair or rules with 0.1"� minor divisions to facilitate locating the bullet hole centers for data entry. I located the vertical and horizontal scales so their zeros are at the bottom most and left most holes positions, respectively.
Next I wanted to get some kind of indication of how repeatable your bench hold was? Horizontal and vertical group size are not interchangeable because wind and barrel vibration affect the two differently. But since you hadn't yet identified an accuracy load to group, the only indicator available was the extreme spread of the position of the horizontal holes fired after the wind died down. I made a circle of that diameter and centered one over every hole location. This means the hole could have been made anywhere within that circle, though the probability gets lower out at the edge. Nonetheless, you can see (images below)shot 14 could have fallen much closer to a horizontal line with 12 and 13 and 15 if you put it at about 4:30 on the circle. Since that is within the hold circle guess, you don't know that it would not have been an accurate load?
Going to the graphs, I first plotted horizontal and vertical dispersion connecting the holes (yellow line) so the firing order was more apparent to the eye. I then plotted spline curves both including (red) and excluding (blue) shot 14, and you can see, even though shot 14 adds a bend, the trend line you would eyeball through the middle isn't too different. I also used the most extreme horizontal position jumps (2, 8, 12, 14, and 17) to pick shots to eliminate, on the assumption this was an indicator of a wind gust or a shooter induced error (heartbeat coincident with shot break, etc.), and plotted a spline curve that skips all these points (brown). This line is much smoother on the horizontal and appears to reveal the wind trend, moving the shots to the right as the wind drops off. When I went back and plotted a spline curve of this same edited shot population on the vertical, it, too, is smoother, but still not really affecting the overall trend.
You will note that shots 2, 8, 12, 14, and 17 deviate in opposite directions in about equal quantity off the horizontal, so this is why the trend error tends to cancel out. Moreover, despite the shot count being too low for great confidence in the results, it is nonetheless an indication that the error in dispersion is random.
So, I concluded there really wasn't any good reason for eliminating any of the shots in the ladder. All seemed to fall within reasonable random error expectation. So, the final graph is a least squares 5th order polynomial fit of the vertical dispersion for all 17 shots I did using Mathcad's regress(vx,vy,n) function. Higher order polynomial fits are tricky, because you can start forcing data fits that aren't real. However, in this case the 5th order fit had almost the same shape as the 4th order fit except it reduced a vertical offset from the first and last shot. It also eyeballed through the middle of the shot 12-15 range better. So I deemed it a legitimately better fit. Third order and below didn't fit well.
The surprise is that the sweet spot looks like it actually fell between shots 12 and 15 in this ladder. This represents a 0.9 grain span of powder charge. I have had bullet and powder combinations that remained accurate over a ±1¼ grain charge span in a .308. The same rifle with the same powder proved accurate over only a ± 0.2 grain charge span with a different bullet weight. So you have to figure out whether the trend lines and data here had enough resolution to get you where you want to go? I would not only try firing 10 rounds at a charge weight in between the shot 13 and 14 charge weights (as labeled on the graph), but I would also fire a new ladder starting at the shot 12 load, and incrementing the charge in 0.1 grain steps up to the shot 15 charge weight. Let's see what happens in the middle?
You could also run a ladder of 0.1 grain increments in the shots 3-5 range to see whether that is a real or a random shot bunching? While the data arrangement I put together here looks promising for shots 12-15, the 0.3 grain increment may just be lacking enough resolution, which would throw these results off (you could have other sweet spots hidden within these if you are on a tight spot, though having one that tight isn't very useful for practical conditions of changing temperature).
A final comment on benching: It isn't as easy as it looks like it should be to get good bench rest results; not even with all the right equipment. A sporting weight rifle in particular may not vibrate the same way in human supported positions as it does of the bench. How much difference it makes is load sensitive. So if the gun is for field use, take Boots Obermeyer's advice and check the loads in real shooting positions to see what groups best for you there, as well? For a given rifle, some loads will be more consistent between the bench and field positions than others, or even between different field positions. It is worthwhile trying to develop a position-insensitive load for the field. Besides, this means getting to buy more bullets and powder and do more shooting.
Nick