It seems odd to me to arbitrarily restrict the alphabet if the only requirement is that the data has to be readable by a machine through an OCR system. They could have easily used the Latin alphabet to encode arbitrary bit strings.

How do you pronounce 64656164626565766573?

You don't? It's machine-readable.

> alphabet if the only requirement is that the data has to be readable by a machine

its the only requirement because

1. it's only meant for OCR

2. it's clear that it won't be used by only OCR but at least also human interacting with the system, potentially phone calls passing this information by voice, and anyone who can't pronounce the original spelling. For fairness if you e.g. didn't sing at all in your life are somewhat tone deaf and now are expected to pronounce a asian you probably have to spend days or more until you can do so (just as a extreme example).

What's interesting is that the transliteration of symbols is not even remotely uniform in icao 9303. There are multiple recommended transliterations of each name, and it definitely goes only in one direction: national script -> MRZ transliteration. It is not possible to go the other direction.

Take a look at the spec if you are interested: https://www.icao.int/publications/documents/9303_p3_cons_en.... The transliteration tables start on page 24 and are scattered throughout the document.

It's not intended to round-trip, it's intended to be roughly human-readable without knowledge of the original script. It's pretty close to the system Olympics used, with the Wikipedia example of Hämäläinen -> HAEMAELAEINEN being well known as a gold medalist cross-country skiier.

Newer versions of the transliteration encourage stripping diacretics, so that would be HAMALAINEN. Much more readable to native speakers, but obviously loses information.

> encourage stripping diacretics

I wouldn't recommend it as there are official tranformations from the countries which uses diacretics and they most times are not to strip them. It's kinda another case of people forcing stuff onto other cultures. And if you do bussiness in some of the countries in some industries you might even get into legal trouble if you apply that.

Take it up with the spec, then. That is the recommendation:

> Section 6 of the 9303 part 3 document specifies transliteration of letters outside the A–Z range. It recommends that diacritical marks on Latin letters A-Z are simply omitted (ç → C, ð → D, ê → E, ñ → N etc.), but it allows the following transliterations: [...]

You said

> you might even get into legal trouble if you apply that.

We're talking about passports, this seems not relevant. For passport-related use such as travel, you use the form of the name written on the passport, exactly as-is.