[chrisdengso]

Follow-up to #10. The quick-start analyzed a sentence a reader cannot tell is spoofed — good demonstration of the attack, poor demonstration of the library, since input and output look the same on screen.

Now opens with visibly styled text:

const r = analyze('Ｇｅｔ 𝐅𝐑𝐄𝐄 ⓒⓡⓨⓟⓣⓞ now');

r.normalized; // 'Get FREE crypto now'

Three different styling systems (fullwidth, math-bold, circled) in one sentence, with now already ASCII and returned untouched — so the before/after is obvious at a glance.

Shows the whole signals object, all five keys, instead of a single field — the previous example only ever showed confusable_word, which made the other four invisible to a reader skimming the API.

The pixel-identical case is now the follow-up, where it reads as the punchline it deserves to be:

lookalike === 'paypal'; // false
[...lookalike].map((c) => c.codePointAt(0).toString(16));
// ['440', '430', '443', '440', '430', '6c']  — Cyrillic, except the final 'l'

Adds an all-signals example — one message carrying a Cyrillic lookalike, an intra-word script mix, a zero-width space, stacked marks and a NUL byte, with all five firing.

Every value above is asserted in test/readme.test.ts (262 tests pass).

Two things worth noting, both caught by that test file:

• 'раураl' === 'paypal' does not typecheck as a literal comparison (TS2367, "no overlap") — the snippet needed an explicit : string annotation to be copy-pasteable.
• I nearly used 𝐁𝐈𝐆 in the headline example; its skeleton is BlG, since UTS #39 folds capital-I to lowercase-l. Avoided capital-I words.

Merge this before #11 ("chore: version packages") so 0.2.0 publishes with the current README.