How we choose the expressions

Selection and labels

The initial editorial seed contains 180 named faces across ten collections. These are selected for recognizable expressions and practical copying, rather than generated in bulk to inflate the inventory.

A label suggests a use or feeling; it is not a universal translation. The same face can mean different things in different conversations. We keep decorations and large text art separate from the core face catalog.

Exact text matters

Spaces, combining marks, fullwidth characters and backslashes can be significant. Copy actions preserve the catalog string. We check for control characters, missing labels and duplicate identities before building the site.

Source inputs are reviewed separately from public data. The original collected compilations are not published as downloadable archives. Source permissions remain an owner review before commercial publication.

Emoji data

The emoji picker uses the official fully qualified Emoji 17.0 sequences and names, with the Unicode license retained. Version updates are deliberate, not automatic downloads from a floating latest address.

Corrections

Send corrections to [email protected] with the page and expression name. We review changes to text, meaning and categorization before updating the catalog.

Separate the public inventory from private experiments

The public library is versioned data, not a live stream of every face someone makes in an editor. A visitor can save a custom version locally without changing the source entry or creating an indexed page. Original packs reuse existing art references where appropriate and keep new authored text distinct from the main catalog. This prevents a pack membership from inflating the count of unique public entries. A duplication check inside the project is useful for maintenance, but it does not prove that a face has never appeared elsewhere. We avoid that global originality claim and describe the actual selection and editing behavior instead.

Write usage notes that answer a real question

An entry’s category should help a visitor choose a tone, not pretend to translate every symbol into one universal meaning. Happy, shy or annoyed readings can change with the sentence and relationship. Usage notes therefore emphasize context, readable words and destination checks. A shrug needs its backslash; a table flip needs enough room for its object; a multiline cat needs preserved spacing. Those differences deserve specific guidance rather than one interchangeable paragraph under every collection. The guides and English page explanations are drafted with AI assistance and checked against the implementation, without invented expert biographies, usage statistics or first-hand experience claims.

Keep display limitations visible

The site preserves approved source text when copying, but a destination’s font, markup rules and Unicode support can still change its appearance. A newer emoji sequence may split on older software, and a complex face can show a missing glyph or wrap onto another line. We do not label that as an automatic-copy failure without checking the sequence. The troubleshooting guide separates these cases and offers a manual fallback for rejected clipboard writes. Text-art notes explain code-block support as destination dependent. Practical accuracy includes stating what the tool cannot control, especially when a visual preview looks convincing on the device used to create it.

Review additions before making them public

More inventory is useful only when entries remain safe, recognizable and distinguishable. Build checks reject unsafe controls and inspect route links and structured-data syntax. Backup and model-output validators are separate trust boundaries, because private imported text and optional generated recipes should not become executable markup. The generator selects curated parts in its initial integration rather than unrestricted output, and it does not auto-publish results into packs. Live model quality remains unverified. Source permissions, final operator facts and production policy details also remain release gates. An attractive page or a clean build is not proof that those editorial and deployment decisions have been resolved.

Make corrections traceable

When reporting a problem, identify the collection, expression name and relevant page. For a source issue, include supporting information that distinguishes a specific design from a generic character shape. We can then assess the actual item rather than guessing from a broad description. Updates to source strings need care because a small character replacement can affect copying, saved references and a picture’s alignment. Updates to descriptions should reflect real behavior, not merely make a page appear newer. The current preview documents remaining source and operator reviews openly; final public publication requires those facts to be resolved and the relevant policies updated.