Changelog

What shipped, and when

Dated milestones from the OmniNode-Protocol repository, newest first. Every entry maps to real work on the main branch.

Docs

Settlement & proof posture, corrected against implementation

Audited the public posture: verifier attestations and escrow settlement are live on SUM Chain mainnet; Halo2 zkML proofs verify off-chain and are deliberately not yet a mainnet-eligibility signal. The site now states both precisely.

Protocol

Settlement dispute write path — open and resolve

Chain-authoritative OpenDispute (funder-signed) and ResolveDispute (validator-quorum over a basis-point threshold of the active PoA set). No personal resolver key; non-signing validators still count in the denominator.

Protocol

Verifier self-claim submission

A verifier submits its own ClaimReward against the InferenceAttestation it holds. No coordinator-claim and no claim-on-behalf — self-claim only, enforced on-chain.

Engineering

Settlement-read RPC client + operator CLI

Base, consistency, and bonded read paths with multi-gate dormancy handling, plus a settlement-read operator CLI: status, session, claimable, and verifier identity mapping.

Proofs

Production-MLP Halo2 proof guide

Byte-deterministic 16→32→16→8 fixed-point proof with a re-derived circuit/VK identity check that refuses to load on any drift. p95 verify 19.6 ms; prove ≈ 2.3 s; no trusted setup.

Release

v0.1.0 — first signed release

Multi-arch Linux binaries (x86_64 + aarch64), each in default and submit variants, with cosign keyless signing (Sigstore Fulcio + Rekor) over the release checksums.

Milestone

Phase 4 complete — pipeline parallelism is live

Two Apple Silicon Macs split a model in half and run real autoregressive inference across a LAN: native GGUF-to-MLX bridge, decentralized RAM pooling, and pure-QUIC tensor routing, end-to-end with zero central servers.

Engineering

Native GGUF-to-MLX bridge replaces HuggingFace

Weights load directly from .gguf via Apple's mx.load() with a custom GGUF-to-MLX key mapping; architecture is inferred entirely from GGUF metadata. No mlx_lm.load(), no HuggingFace Hub.