Generic "brain training" apps often measure an abstract internal score that's hard to relate to anything real. Arithmetic is different — accuracy and speed on a sum are about as concrete as a cognitive measurement gets, and they map directly onto something you already do in daily life.
"Brain age 34" or a proprietary index doesn't tell you what to actually do differently. "82% accuracy on division, 94% everywhere else" does — it points at exactly what to practise next.
That specificity is why Math Games reports plain arithmetic stats instead of a synthesised score.
The strongest criticism of brain-training games is that improving at the game doesn't always improve anything outside it. In 2014, over 70 cognitive scientists signed an open letter through the Stanford Center on Longevity arguing there was no solid evidence that commercial brain-training games improve general cognitive ability or everyday functioning — a claim a separate group of about 133 researchers publicly disputed months later, and the argument over exactly which studies count as good evidence hasn't fully settled since.
Mental arithmetic mostly sidesteps that specific argument, because the skill being trained — fast, accurate arithmetic — is the same skill used splitting a bill or checking a discount. There's no transfer question to argue about when the practised task and the real-world task are the same thing.