Hi,

I examined the MathML markup in MathTestNative. Well done! It's great to see such a long test page - but it is also a reason that my feedback is partial, as I ran out of free time before I could inspect all examples.

Here are some suggestions for improvements:

• For \operatorname{} it looks like each individual letter in the argument becomes a separate <mi> element. It should be possible to group those into a single <mi> for the entire argument.
• There is a similar need of grouping in numeric scripts, e.g. the "Overbraces" example has ...^{5050} where each digit is in a separate <mn>. An alternative <mn>5050</mn> would be better for a11y. You may similarly want to group 0.5 into a single <mn>0.5</mn>.
• For function names, such as \min, \max, \liminf, \limsup, \dim... , they are usually better represented as an <mi> element, possibly with an invisible function application (U+2061) <mo> between the function name and the argument. If you keep them as <mo> you might want to add a form="prefix" attribute (which is sometimes there, but not always - e.g. \liminf does not have it).
• You can consider using that invisible apply character (U+2061) after other function names with known arguments (such as \sin, \cos...). Some accessibility tools take advantage of it, I am told, but it may also be used to provide correct spacing from the MathML operator dictionary. In Chrome <mi>sin</mi><mi>a</mi> will show up with no added spacing as sina, which is not ideal.
• There is a curious use of <mstyle> around spacing elements, as in <mstyle scriptlevel="0"><mspace width="-0.167em"></mspace></mstyle>. I am not sure that wrapper element is needed? The mspace ought to be successful standalone.
• The {smallmatrix} example has its fences wrapped in too many mrows I think, so it renders without stretching them in both Firefox and Chrome. Ideally you should be able to emit the <mo>(</mo> followed directly by an <mtable>, and trailed by <mo>)</mo>, and only wrap those 3 into a parent <mrow>.

I am not the most proficient MathML Core practitioner, so this may be a partial list (and you may want to double check with the spec text).

Firefox reports several parse errors in the mathml in

https://en.wikipedia.beta.wmflabs.org/wiki/Help:MathTestNative

Error in parsing the value '' for 'columnspacing' attribute of <mtable/>.  Attribute ignored.Invalid markup: Incorrect number of children for <munder/> tag. Help:MathTestNative
Invalid markup: Incorrect number of children for <msup/> tag. Help:MathTestNative
Invalid markup: Incorrect number of children for <munder/> tag. Help:MathTestNative
Invalid markup: Incorrect number of children for <msup/> tag. Help:MathTestNative
Invalid markup: Incorrect number of children for <munder/> tag. node.js:370:30
Invalid markup: Incorrect number of children for <msup/> tag. node.js:370:30
Invalid markup: Incorrect number of children for <munder/> tag. node.js:370:30
Invalid markup: Incorrect number of children for <msup/> tag. node.js:370:30
Invalid markup: Incorrect number of children for <munder/> tag. utils.js:166:21

A typical cause of the incorrect child count is empty tex groups.

For example the slightly nonsensical example

\underset{\mathrm{def}}{}

on the page is producing an <munder> with just one child. The mathml could be simplified simply removing the munder (corresponding to correcting the tex to \mathrm{def} without the spurious \underset) or if you want to model the tex exactly you need an empty <mrow/> (or <mi/> to match the empty tex argument so that <munder> has two children.

Math Alphabets rendered incorrectly. in

https://en.wikipedia.beta.wmflabs.org/wiki/Help:MathTestNative

The math alphabets such as \mathcal{ABCDEF} are rendered using <mi data-mjx-variant="-tex-calligraphic" mathvariant="script">A</mi> mathvariant is not in mathml-core so the end result depends on CSS and local font install. On Chrome I get Cambria Math ascii in italic style A and in firefox I get STix Two Math Ascii A in italic style, so not calligraphic/script in either case. It would be better to use the Unicode math alphabet codepoints, so in this case 𝒜ℬ𝒞𝒟ℰℱ (perhaps supplying stix two math as a fallback web font if no suitable font is locally available, but that applies to all math symbols)