I'm going to take the time to go through each of the concerns here, but before I do let me give a general disclaimer about my position about all these statistics:
First, I'm not going into this being defensive. There were a couple good catches here that I will use to update my guide. However, since I will be throwing out some numbers here it is going to look like I'm trying to defend my points with math, and that can seem harsh. That is not my intention. It's just for clarity.
And second, be careful of mathhammer.
I don't know of an easier or faster way to kill your enjoyment of the game than to mathhammer it into the ground. By adhering to statistics and math over your enjoyment of the game you will often find yourself in situations where your statistically-perfect unit or army gets inevitably shafted by dice probability or your own tactical mistakes stemming from those numbers. There is a danger to succumbing to a general smugness that comes from thinking you are starting from a superior advantage. I am not saying anyone here correcting me is being smug, but I am speaking from personal experience that trying to math every move is a frustrating waste of time, taxes your enjoyment of the game, and makes you less fun to play against. This is not to say that you shouldn't learn and know your army and know its strengths and weaknesses--you absolutely should--but please make sure you are enjoying your time with the hobby first and foremost, and that everything else comes second. /soapbox
Okay, now on to my self-defense.
I used https://aos-statshammer.herokuapp.com/ just like you did. What likely skews our agreement here is unit size, using all available buffs on the warscroll, and the enemy save. Whenever possible I try to skew my statistics and math to real-world expectations and point-to-point comparisons.
In this case, a single model of Bladegheists vs a single model of Glaivewraiths, if you enable all the buffs on each (charged and ST on 'geists, charged on 'wraiths) is 25% more damage for the 'geists against a 4+ save. But at max unit size, 20 for the 'gheists and 16 for the 'wraiths, the 'geists will do 66.24% more damage assuming all the models could attack, which isn't a realistic scenario. However, 10 'ghests who can all attack, and 12 'wraiths who also all can attack, which both cost the same amount of 180 points, is a difference of 49.36708860759494 damage. Granted, I rounded up instead of down to 49.4%.
Exact numbers have a way of not being exact. As I said above, any change of the variables involved, sample size, or even replication size can skew the results. Very few people who mathhammer actually use statistically sound biases and frequently find themselves subject to something called "P-hacking" whether they mean to or not. In order to prevent that as much as I can for myself, I try to stick to comparisons that are likely to show up on the board or be a factor in people's lists. For example, point-for-point comparisons because most people will wonder what to do with 200 points and not 20 vs 16 models, and reasonable expectations of how many of those models will actually be doing something, which I will get into below. I'm just trying to offer "back of the napkin" math and do so only in areas where it illustrates the point.
It's 50% more damage that gets through the save based on the pior examples. Not your rolls, theirs. Which is a sliding scale. +2 to +3 is 50%. +3 to +4 is 25%, +4 to +5 is 20%. It's the end-result we're looking at here.
As for Spectral Lure, casting is two dice needing to meet or exceed a base value. In this case 6. The probability of a 6 or higher appearing on two dice is
or a total of 72.22%, like you say. Dropping that cast to a 5+ adds 11.11% to that pool, making it 83.3%. Again, just as you say. Until you factor in the unbind attempt. Those 72% and 83% chances are for your roll, now you have to see if an opponent can snipe it and unbind you. Assuming you made the exact roll you needed to make to cast each time (6) then your opponent needs to roll a 7 to beat your roll, no matter your bonus. They'll have a 58.34% chance to undo you. If you add this in, then your actual chances of getting the cast off is 72-58=14% for the 6 and 83-58=25% for the 5. So, you have a 25% chance at getting the spell through with the +1. So, if I were to correct it I would drop the 30% statement I made down to 25%, which I can do. And honestly, I got the original 30% from TellTaleNoob. I assumed I could shorthand it by stealing his stat.
And lastly, the 15 attacks statement. This one I need to go back and reword for sure. I don't mean that All-Out Attack will increase your damage by 50%. I'm trying to say that spending a CP on All-Out Attack will be worth more on units that can throw out more attacks in general, and that I personally don't like spending it on units that won't be throwing out at least 15, because 15 seems to be the breakpoint of reasonable attacks you can actually get to throw and is 50% more damage than if you threw any less.
In before that gets challenged, here is my math on it. Let's assume that you have Y unit with X number of attacks and no other buffs. You may or may not have various ways of increasing the number of X, like the Knight of Shrouds on steed's CA or Bladegheists charging. 10 Bladegheist attacks will average 2.96 wounds against a 4 save. 15 attacks will average 4.44 wounds. That's 50% more. For Spirit Hosts, let's say you had 12 vs 18 attacks. 12 is an average of 2.5 wounds and 18 is an average of 3.75 wounds. Again, 50%. This will hold true across most units that have ways of generating both below and above 15 attacks, making it the breakpoint. So, if you want to know where to spend a KoSoES CA, look for places to push the number of attacks to at least 15. And if you are looking for where to spend CA on All-Out Attack, don't waste it on units who will be attacking less than 15 unless it's all you've got (and aren't spending the CP on charge re-rolls instead).