Skip to Content
DocsVersion History

Version History

v0.2.0 (Current)

Adds the types needed for structured, order-sensitive data such as rule tables. Every v0.1 schema loads, encodes and fingerprints exactly as before; this is proven against all production and documentation schemas with a frozen copy of the v0.1 implementation.

Language

  • int: signed integer within ±(253 − 1), with optional min / max bounds
  • list<T>: ordered list; order is part of the canonical form, the fingerprint and the wire
  • map<string,T>: string-keyed map of any type
  • Named types in a types section: record, variant (exactly one field present) and enum
  • Optional fields inside records; recursive types (guarded)
  • Rules and conditions as data: predicate trees are variant values, not expressions

Compiler and library

  • check(value, schema): type checker with JSON-pointer paths
  • canonicalJson(value, schema), typedFingerprint(value, schema), versionedId(name, fp)
  • Extended Vein (LE magic, 32-bit lengths) for schemas using v0.2 types
  • lodec check, lodec canonical, lodec fingerprint -s <schema> [--id <name>]
  • fingerprint and loadLode no longer need Bun-only APIs

Fixes

  • v0.1 Vein encoding throws for values over 65,535 bytes instead of writing a truncated length

v0.1.2

Stable release of the LODE language and lodec compiler.

Changes since v0.1.0

  • Improved error messages for schema validation
  • lodec diff command for comparing Vein binaries
  • Documentation portal with versioned docs

v0.1.0

Initial release of the LODE language and lodec compiler.

Features

  • Schema definition format (.lode files)
  • Vein binary encoding/decoding
  • Four primitive types: string, bool, string[], map<string,string>
  • Deterministic SHA-256 fingerprinting with canonicalization
  • Verification (decode + fingerprint + compare)
  • lodec CLI with compile, decode, fingerprint, verify, diff commands
  • Forward compatibility (unknown fields skipped by decoder)

Schemas

  • model-layer-v1.lode — Model pool definitions for the StackNet registry

API

// Core functions compile(source, schema): Uint8Array decode(vein, schema): Record<string, any> fingerprint(obj): string verify(vein, hash, schema): boolean loadLode(pathOrObj): LodeSchema

Wire Format

  • Magic bytes: 0x4C44 (“LD”)
  • Header: 4 bytes (magic + version + count)
  • Fields: id(1) + tag(1) + length(2) + data(N)
  • Byte order: big-endian

Roadmap

Planned

  • Raw bytes type (bytes)
  • Schema import/reference mechanism
  • Streaming encoder for large payloads
  • Schema-to-TypeScript and Rust code generation

Floats are not planned: use scaled integers (basis points, milliseconds), which are exact in every language.

Last updated on