Due several hints in the comments I modify my query. It turns out that I have some know-how to use calculus but almost no know-why regarding analysis. That is why I alter my question to is my assumption reasonable.
Why I ask? Recently I found two hints, term-wise integration would require some theoretical justification, first one in a comment here, another in an elderly inaugural lecture held in 1911 by Oskar Perron -- "Über Wahrheit und Irrtum in der Mathematik" (German: "About verity and fallacy within mathematics") -- containing the unambiguous statement (or warning?), that term-wise differentiation and integration will lead to wrong results if specific prerequisites lack.
For sure without any justification of any "specific prerequisites" I evolved as student the normal distribution as Maclaurin series, integrated it term-wise, and combined it (after a polinomial division) to following formula
$$Q(x)=\frac{1}{2}+x\frac{e^{-x^2/2}}{\sqrt{2\pi}}\sum_{n=0}^{\infty}\frac{x^{2n}}{(2n+1)!!}$$
(which nowadays you find also in Wikipedia)
Today, after reading chapter 6 of Abbott's "Understanding Analysis" I assume following conditions should be checked to ensure "usability" of the approach:
- is Normal Distribution convertable to a Maclaurin series? Yes, it is, since it's not "extremely flat" at $X=0$
- how big is the "radius of convergence" (or interval of convergence)? -- I did not find out yet
- when computed to an accuracy of 10..12 digits the remnant up to infinity (according Lagrange's Reminder Theorem) must be small enough to have no impact on the result
Question: is the verification of this three points sufficient? What else should be checked?
Note: commentator Sgg8 suggested Abel's theorem as only prerequisite. But I can't see (yet) the reasons.
Trivia: Some time ago I found by reverse engineering the firmware of an ancient HP21S pocket calculator that a. m. formula is used for input $|x|\le2.32$ while other models use approximations. (When HP used it too, maybe I was not all that wrong -- but who knows what they once checked before relying on it.) I assume the range $|x|\le 2.32$ is due to shortest execution time, thus the second algotithm is faster for $x\gt 2.32$