Transport Layer: Gossip-Erasure Protocol¶

Design Constraints¶

The Analog Signal Insight¶

Digital protocols are binary: packet arrives or it does not. This creates fragility — a single lost packet can break a transaction.

Niko's design uses the analog model: the signal can degrade at the edges, lose fidelity, arrive with noise — and still be readable. This is achieved through:

Erasure Coding¶

• Original message M is encoded into n fragments using Reed-Solomon or fountain codes (LT codes, Raptor codes)
• Any k fragments (where k < n, typically k ≈ n/2) are sufficient to reconstruct M
• You can lose up to (n - k) fragments and the message still arrives intact
• This is literally how analog signals work: redundancy in the signal provides noise tolerance

Fountain Codes (Preferred)¶

Fountain codes (LT codes, Raptor codes) are ideal because: - The encoder can produce an unlimited number of fragments from a single message - The decoder needs only slightly more than k fragments to reconstruct - No coordination needed between sender and receiver about WHICH fragments to send - The sender just keeps spraying fragments; the receiver collects until it has enough - This maps perfectly to the gossip model: spray fragments into the network, they bounce around, every node eventually collects enough to reconstruct

Gossip Protocol (Not Multiplex)¶

Why Not Full Multiplex¶

Full multiplex (every node sends to every other node simultaneously) is: - O(n²) in bandwidth — scales quadratically with network size - Requires coordination — which is the antithesis of steganographic invisibility - Compute-expensive — too much energy per node, violates U = V/(E×T)

Gossip (Epidemic) Routing¶

Each node, on each cycle: 1. Selects a small random subset of known peers (fan-out f, typically 3-5) 2. Sends them any fragments it has that they might not 3. Receives any fragments they have that it might not 4. Repeat

Properties: - O(n log n) total messages to reach all n nodes — scales beautifully - No coordination — each node acts independently, random peer selection - Guaranteed convergence — mathematically proven that epidemic spreading reaches all connected nodes with probability → 1.0 as cycles increase - Resilient to node failure — if a node disappears, others fill the gap through redundant paths - Low compute per node — each node only talks to f peers per cycle, not n - Indistinguishable from normal traffic — random HTTP/3 requests to random servers look like normal web browsing

The Bouncing Pattern¶

"Signals bounce around and get everywhere onto the network eventually."

This IS gossip. A fragment: 1. Is created by the originating node 2. Sent to 3-5 random peers via steganographic HTTP/3 requests 3. Each peer forwards to 3-5 of THEIR random peers 4. The fragment propagates through the network like a rumor 5. After O(log n) hops, every node has seen it 6. Combined with erasure coding: even if some forwarding chains die, enough fragments survive to reconstruct the message at every node

The Full Stack¶

Layer 4: Application     — Double-entry ledger, smart contracts, federation
Layer 3: Consensus        — Cross-verification of double entries
Layer 2: Assembly         — Erasure decoding (fountain codes reconstruct from k-of-n fragments)
Layer 1: Gossip           — Epidemic routing, fan-out f=3-5, random peer selection
Layer 0: Steganographic   — HTTP/3 (QUIC/UDP), WebAssembly runtime, encoded in headers/metadata/timing

WebAssembly Runtime¶

• WASM module runs in any browser as part of any web page
• User visits a website → WASM module loads → node joins the gossip mesh
• The module:
• Encodes outgoing ledger entries as erasure-coded fragments
• Embeds fragments in outgoing HTTP/3 requests (steganographic carrier)
• Extracts fragments from incoming traffic
• Reconstructs messages when k fragments collected
• Gossips fragments to random peers
• All of this happens invisibly inside normal web browsing
• No installation. No app. No visible software. Just a WASM module embedded in web pages that participate in the network.

Fallback Chain¶

Multiple carrier channels provide fallback redundancy:

1. Primary: HTTP/3 (QUIC/UDP) headers — fastest, highest bandwidth
2. Fallback 1: DNS query encoding — works even when HTTP is filtered
3. Fallback 2: SMTP header encoding — email always gets through
4. Fallback 3: WebSocket timing modulation — works through any proxy
5. Fallback 4: Image metadata (EXIF/XMP) on social platforms — works even on heavily censored networks
6. Emergency: Timing-based encoding in TCP ACK patterns — works on literally any internet connection

Each fallback is independent. Blocking one does not affect the others. Blocking ALL of them requires shutting down the internet entirely.

Performance Estimates¶

Source: @B_Niko, session v7, 2026-03-10