And there are lots of other factors to be concerned with than just how tolerant of target course errors our solution is. For instance against warships, I believe you want the John P Comwell shot from ahead of the target, because every second you wait to shoot is a second during which they might detect you. When they do, all that beautiful data you've collected is in the garbage and you don't care HOW accurate it was...
Also there's the error mitigating fact that the faster the torpedo's approach speed to the target, the smaller the error radius of the impact point. Using the roughly 110º torpedo track angle, the torpedo is approaching the target from 20º aft of the beam. So from your 46 knot torpedo speed you must subtract some of the speed of the target to get the torpedo approach speed.
Similar to the way speed is added to torpedo speed by the target's approach in the John P Cromwell technique, the approach speed of the torpedo must be reduced by the speed the target is receeding from you at point of impact. Well that is calculated by the cosine of the 110º torpedo track angle, .34 in round figures. So the 31 knot approach speed of the torpedo must be reduced by 34% of a 20 knot target's speed! That leaves the already anemic electric with an approach speed reduced by 6.8 knots! And with a target approached at only 24.2 knots, there is all kinds of time to ruin all your beautiful numbers.
The graph shows a very incomplete view of all the decisions you must make during an attack. I question their usefulness as a practical matter.