SPECIFICATION · AXIA.ORIGIN.V0 · DRAFT

Axia Origin Layer

Reconstruct the history of a piece of content. First-seen registration and derivative detection via cryptographic + perceptual hashing.

This profile does not assert ownership.

The Origin Layer records which entity registered which content first inside AxIA Protocol. It does not assign copyright, authorship, or legal ownership. Every API response includes an explicit disclaimer to that effect.

1. Scope

The Origin Layer (axia.origin.v0) lets verifiers answer two questions about a piece of content:

The response carries a single classification field with one of three values: original, derived, unknown.

2. Algorithm

Exact (SHA-256)

Authoritative for tamper detection. Unchanged content always matches.

Perceptual (dHash 64-bit)

Image-only in v0. Compares Hamming distance ≤ threshold (default 35/64).

First-seen ordering

Earliest created_at wins. Ties broken by lowest distance.

MUST — Implementations treat SHA-256 as authoritative for tamper detection. Perceptual similarity MUST NOT override an exact-hash match.

SHOULD — Verifier UIs display the three states using neutral language (Original / Derived / Unknown) and SHOULD NOT display raw hashes or distances in the primary surface.

3. API

4. Perceptual hash format (dHash)

For images: scale to 9×8 grayscale, compare each pixel to its right neighbour (8 comparisons × 8 rows = 64 bits), pack into 8 bytes encoded as 16 lowercase hex characters.

This is intentionally simple, deterministic, and library-free. Hamming distance over 64 bits is small enough for a sequential scan up to ~10⁵ rows; pgvector indexing kicks in at scale.

5. What this is not