ArcTable and Commitments

ArcTable and commitment backends support the semantic layer, but they do not replace it.

For the full design narrative, see System Design. For current package roles, see Runtime and Prototype Status.

ArcTable

ArcTable provides root-recoverable arcset persistence and materialization:

given root -> recover enough semantic state to answer queries and generate proofs

It belongs to the performance plane. It is allowed to use implementation-local storage prefixes, namespaces, indexes, or partitions. Those details are not part of canonical semantic identity, commitment inputs, ProofList, or VerifyRead.

Incorrect materialized state is rejected by verification or root recomputation.

Commitments

Commitment backends are stateless primitive proof engines over cell vectors. They do not know whether a cell vector came from a list node, a map binding vector, a radix node, or a bucket.

They are responsible for:

• commit
• prove
• verify
• update when the backend supports efficient local updates

They are not responsible for map key semantics, list range semantics, path resolution, application layout, or root publication policy.

The semantic packages add a narrow storage-free layer above those primitives: auth/semantic/list.Commitment and auth/semantic/mapping.Commitment expose Commit, ProveSlot, and VerifySlot for semantic slot representations. Runtime implementations then compose those single-step primitives with ArcTable materialization and tree/radix traversal.

Canonical ArcSets

The target representation is deterministic and semantic-layer owned:

CanonicalArcSet {
  kind: map | list
  entries: []ArcEntry
}

ArcEntry {
  coordinate: CanonicalCoordinate
  target: TargetRef
}

Entries are sorted by canonical coordinate bytes. A well-formed ArcSet has at most one target per coordinate. Conflicting bindings for the same coordinate are invalid inputs; equivalent duplicate input may be rejected or collapsed before canonicalization. Coordinates are encoded by list or map semantics, not by ArcTable.