Splendor
Reference

Serialization and Determinism

Rules for serialization, hashing, and deterministic IDs.

Serialization and Determinism

Splendor uses deterministic serialization, explicit identity fields, and content hashing so runs can be inspected and replayed reliably.

Serialization Rules

  • Public schemas use stable serialized field names.
  • JSON output is suitable for trace export, replay inspection, and API clients.
  • Trace sequence numbers are strictly monotonic within a run.
  • Unknown extension metadata must not influence authorization, verifier results, adapter selection, quota decisions, or replay authority.

Timestamp Handling

  • Runtime events store their original timestamps.
  • Replay and audit tooling use stored timestamps rather than wall-clock time.
  • Stable cross-language schemas use RFC3339 timestamps unless a local SDK field explicitly documents another representation.

Content Hashing

State nodes and trace integrity records use content hashes to make state and trace evidence tamper-evident.

Canonical hash strings use:

{algorithm}:{hex_digest}

State nodes and snapshots

State node identity is derived from parent linkage, state data, and metadata. Snapshots are content-addressed references that replay and handoff tooling can validate before use.

Trace integrity

Trace stores can chain event hashes so missing, reordered, or corrupted records are detectable during export, audit, or replay.

On this page