Nifty has asked me to pop in here and give a quick explanation of why the higher magicka total comes out ahead of simply adding extra spell damage in some cases or builds like this one, after I gave him the hint awhile back to look into it and a similar but not as refined general gear concept that was built out into his own great one.
I’ll keep it relatively brief: in ESO, every skill has a zero base value, or just shy of it. The entire damage, healing, or effect number you see in the tooltips is derived from the main resource cost in 99% of abilities’ cases, and the matching “power” rating (aka weapon damage, spell damage, or health). So, for example, a skill might scale at your main resource cost stat, multiplied by a coefficient such as 0.01327. It then is added to the power rating multiplied by its coefficient respectively which for the majority of skills ranges from an average of 9 to 11.5x the value of the main resource stat coefficient’s value, and a tiny (often either zero, or near-zero) extra factor that’s basically a remainder, and that is what you see in your tooltip. For example, the Templar’s Puncturing Sweep has a magicka coefficient value of 0.014x (sorry, but I’ll keep the full data obfuscated just a hair ), and a power coefficient of 0.155xx, with a “remainder” value that is simply added to that result of 6.2x. Plug in your character’s resource values after factoring in what you have after passives and all applied, and you’ll get the exact value you would see on your tooltip in-game. The ratios on these skills vary as noted, but in many cases depending on your build, you can gain extra damage/healing/etc. out of aiming for magicka, by outstripping the bonus you’d gain to the skills from simply stacking up the power value after passives and buffs. Additionally, you increase your resource pool to cast from in general to help sustain.
All of the coefficients in the game are reverse-engineerable by using a curve fit with R, a statistical data analysis framework, alongside a custom in-game addon (one is publicly available and open-source. I have my own creation) to log values to a SavedVariables file with. The general gist of how you go about this is to equip various gear that provides isolated stat gains across a variety of ranges and values, without losing the controlled measurement by altering, for example, spell power when aiming to find the magicka coefficient information on a set of skills. Similar to Wykkyd’s Outfitter, you script it to unequip and change gear on the character from a number of ones you define in advance, and have your tool do so, then have it log the values for each skill. After parsing it with any manner of programmatic way you decide to based on how you exported your data set, to a friendlier form to programmatically analyze it, you can then process it through R and get a confidence value for the results.
What this means could be written about for quite awhile, however the gist of it is that this “R-squared” number tells you how close, mathematically, you can be sure of the results that were calculated as a percentage. This is influenced by the smoothness of the curve, and can be helped towards “1.0000” (or “100% confidence, in other words, it is a definite) in some cases by adding additional data points for it to work with. Some numbers can’t be fit completely accurately, but typically, most users regard a confidence value of 0.98-0.985 as good enough to use for data analytics in a case such as this. An example of an R-squared number on one skill, might be “0.9841”, while most will be “1.0000”. The odd one or two that I’ve seen, have come as low as the “0.7” range, but only on the tetriary effects nearly exclusively (an example of this is Replenishing Barrier, which has a rather strong though imperfect R-squared confidence of ~0.981 for its shielding value, but the magicka restoration part only landed at ~0.67 for me).