Unicode, IDNA, and confusable names

Names can contain characters beyond ASCII. That is useful and it is dangerous, because two different strings can render identically.

The attack

Someone registers a name that renders exactly like one you trust, using a different code point. You read it, recognise it, and pay. The bytes were never the same.

Three different strings

A reader sees one word. A comparison sees three:

Appearance	What differs
A Latin lowercase a	The ordinary code point
A Cyrillic lowercase a	A different code point that renders the same in most fonts
A Latin a with a combining mark	Two code points that compose to one glyph

Normalisation

Normalisation maps different byte sequences that mean the same thing to one form, so comparison works. Two rules follow:

1. Normalise before comparing. Two names are the same only after both are normalised the same way.
2. Keep the exact original bytes. The authoritative name is what was minted. The normalised form is for comparison, not for display or for identity.

Never store only the normalised form, and never display it as if it were the name.

Authoritative against display

Form	Use
Exact minted bytes	Identity, storage, and what a validator saw
Normalised form	Comparison, deduplication, confusable detection
Display form	What a reader sees, escaped and marked when risky

What a product must do

1. Reject names that fail validation for the target rule set, with a clear reason.
2. Normalise before comparing, never before storing.
3. Detect mixed scripts within one name and warn.
4. Detect confusability against names the user already uses and warn prominently.
5. Escape names everywhere they are displayed, including in link text and error messages.
6. Neutralise bidirectional control characters so a name cannot reorder surrounding text.
7. Show the Atomical ID next to any name the user is about to act on.

Error reporting

A rejected name deserves a specific reason: which character, which rule. A generic failure teaches the user nothing and pushes them to guess.

The Universe read model surfaces Unicode and IDNA errors explicitly rather than returning a generic validation failure. See NFT and Realm API.