VER 1.0 — The Visual Embedding Recordstable
§3 Architecture
VER is organized as four layers. Layers 1–3 correspond exactly to the thesis's three layers. Layer 0 is new — and it is what makes Layer 1's determinism claim true.
- Layer 0 — Identity & Normalization. CPNP-1 resolves render-affecting metadata into the Canonical Buffer and computes the identity gradient.
- Layer 1 — Canonical Visual Embedding(s). One or more vectors computed from the Canonical Buffer only, each bound to a declared Embedding Space. Metadata availability MUST NOT be a precondition for Layer 0 or Layer 1.
- Layer 2 — Metadata & Provenance. Byte-exact preservation, reconciled normalized view, availability states, trust tiers, privacy redaction, acquisition chain, C2PA binding.
- Layer 3 — Contextual Embeddings. Optional text-derived vectors, provenance-tagged by source class, never fused into Layer 1.
The identity gradient. Every Record carries up to four tiers of identity, each answering a different question:
| Tier | Artifact | Survives | Fails under / Role |
|---|---|---|---|
| Byte | content_hash — SHA-256 of Asset bytes | Exact copies only | Any re-encode or metadata rewrite. Role: dedupe, C2PA binding. |
| Pixel | pixel_hash — SHA-256 of Canonical Buffer | Metadata stripping and rewriting | Recompression, resize, crop. Role: render identity. |
| Perceptual | PDQ / pHash digests | Mild recompression, resizing | Crops, heavy edits, adversaries. Role: near-duplicate clustering. |
| Semantic | Canonical visual embedding | Most benign transforms | Model change, adversarial perturbation. Role: similarity, retrieval. |
Design principle (normative restatement of the thesis): Preserve everything. Require nothing but pixels. Resolve what rendering requires. Declare every space. Trust only signatures.