Serialization and Determinism

This document defines the serialization and hashing rules required for reproducible kernel runs.

Serialization Rules

• All core objects derive serde::Serialize and serde::Deserialize.
• JSON and CBOR round-trips must preserve semantic equality.
• TraceEvent.sequence is strictly monotonic within a RunId.

Timestamp Handling

• TraceEvent.timestamp is captured at emission time and persisted verbatim.
• Replay and audit tooling must use stored timestamps instead of wall-clock time.

Content Hashing

ContentHash uses blake3 for deterministic digests. The canonical string format is:

{algorithm}:{hex_digest}

Snapshot IDs

SnapshotId is derived directly from the snapshot bytes:

SnapshotId = ContentHash::blake3(state_bytes)

State Node Hashing

StateNodeId is derived from a serialized StateNodeHashInput payload:

{
  parent_ids: [StateNodeId],
  data_hash: ContentHash,
  metadata: StateMetadata
}

The payload is serialized with serde_json and hashed with BLAKE3. Any change in ordering or field representation will produce a new node ID.