Metadata trust boundaries

Anyone can mint an Atomical containing any bytes with any declared type and any claimed name. Nothing in a payload is checked by anyone before it reaches your screen.

What a payload can claim, and what it proves

The payload says	It proves
name is a well known brand	Nothing
image is a famous artwork	Nothing
legal grants you rights	Nothing
links point to an official site	Nothing
The content type is an image	Nothing. Verify the bytes
The ticker matches a project	Only that a name was allocated to some Atomical

Rules for a renderer

1. Choose the content type yourself from the bytes. Do not trust the declared type.
2. Restrict the set of types you will render at all.
3. Enforce a size limit before decoding.
4. Serve media from an isolated origin with a restrictive content security policy.
5. Never inline payload content into your own document.
6. Sanitise any text before display, including names, descriptions, and link labels.
7. Verify the digest before serving.

Rules for a user interface

1. Show the resolved Atomical ID next to every name.
2. Mark unresolved name requests as candidates.
3. Never present metadata as verification.
4. Warn when a displayed name is visually confusable with another. See Unicode and IDNA.
5. Show the data source, its revision, and its freshness.

Rules for an indexer

1. Bound payload size and nesting depth before parsing.
2. Reject or quarantine payloads that fail to parse rather than partially applying them.
3. Record the digest of every stored file.
4. Keep the raw bytes exactly as revealed. Optimised derivatives are separate artefacts.
5. Never let payload content influence routing, authentication, or storage paths.