How it works

From a signature to sealed history.

One record's journey through the mesh — five stages, from a local signature to a signed epoch seal (every 60 seconds by default), post-quantum signatures at every step.

1 · RECORD 2 · GOSSIP 3 · WITNESS 4 · SEAL 5 · VERIFY ANY DEVICE sign() record ZONE COMMITTEE ⅔ ATTESTING STAKE EPOCH SEAL merkle: 0x4ad1… LIGHT CLIENT
FIG. 01 RECORD LIFECYCLE — SIGN / GOSSIP / ATTEST / SEAL / PROVE
60 sdefault epoch — each closes under a signed Merkle root
⅔ stakeattesting-stake finality threshold (independence weighting reported, not yet enforced)
64-levelrecord-inclusion proofs, checked offline — even in a browser tab
100%offline-capable record creation
1

Record — on any device, online or not

The work is hashed (SHA3-256), signed with post-quantum keys (ML-DSA-65 — FIPS 204, "Dilithium3"; an optional second SPHINCS+ signature is not yet bound to the signer — see known limitations), and stamped with an interval tree clock — causal time that needs no synchronized clocks and no connection.

2

Gossip — the mesh moves it

Records travel a peer-to-peer mesh over ElaraPQ, a post-quantum transport (hybrid X25519 + ML-KEM-768 key exchange, ML-DSA-65 authentication; peer keys are trusted on first use). Content-routed DHT delivery is designed to keep it efficient as the network grows — so far tested only on small networks, where flood-gossip covers delivery.

3

Witness — staked peers attest

Staked witnesses sign attestations, and a record settles once attesting stake passes two-thirds. MESH-BFT's design also weighs independence: witnesses that share an organization, subnet or location count less, so one operator's clones should not settle a record alone. That weighting is computed and reported today, but it does not yet gate finality, and it cannot help against identities spread across distinct organizations and networks — known limitations.

4

Seal — history becomes immutable

Epochs close over records with Merkle roots, chained seal-to-seal and periodically consolidated into super-seals and checkpoints. Rewriting one record means rewriting every seal after it — so an edit cannot hide from anyone who kept an earlier seal.

5

Verify — offline, against math

A light client downloads headers, not history: checkpoint skip-sync means no genesis replay. Any record's existence and position is verified with a Merkle proof (at most 64 hashes) against a sealed root — light enough for a browser tab (phones are not yet measured). The check is only as good as the root, so pin a seal you trust or check its anchors.

Verify a real record in your browser →

+ How MESH-BFT is designed to resist clone armies

Classical BFT counts weight. MESH-BFT's design counts independent weight: witnesses that share an organization, a subnet, or a geography are correlation-penalized before their stake is summed, so the committee on the left, with 100% of the raw stake, would not settle, while the diverse committee settles with the same total.

Status: the shipped code computes this weighting and reports it, but finality today counts raw stake — so, today, the clone committee on the left would settle; stake is what holds clones off, as in any proof-of-stake system (known limitations).

The aim: Sybil resistance by arithmetic, not CAPTCHAs.

SAME ORG · SAME SUBNET raw stake 300 → effective ~107 BLOCKED THREE INDEPENDENT ORGS raw stake 300 → effective 300 SETTLED ✓
FIG. 02 MESH-BFT DESIGN — CORRELATION-PENALIZED STAKE (NOT YET ENFORCED)

The math is in the papers.

Committee selection, seal chains, light-client proofs and the diversity-weighting design — written down and open to review. The proofs are sketches, not machine-checked; where papers and code differ, known limitations says how.

Protocol whitepaper    MESH-BFT paper