System overview

Architectural thesis

CoNET separates concerns that are often collapsed into one “decentralized network” label:

1. Where do decentralized network, storage, and compute resources come from? — L0, the permissionless cloud resource plane.
2. How does encrypted application traffic move without using a stable public origin as the application identity? — Layer Minus, the wallet/OpenPGP-addressed privacy-routing protocol built on L0.
3. Where does shared, production state settle? — L1 CoNET Blockchain (EVM PoS, chainId 224422).
4. How can application activity scale into specialized parallel ledgers? — L2 CoNET-DLE.
5. How are wallet-addressed applications opened and published? — The web3:// Application Protocol defines locators, caller-signed requests, correlated encrypted responses, persistent streams, and browser adaptation on top of Layer Minus.

Operational facts in this book come from current source code, deployed contracts, and live endpoints. Destination architecture is labeled separately from production join paths.

Responsibility by layer

L0 — decentralized cloud resources

L0 is the permissionless decentralized cloud resource plane. Participants may contribute traffic forwarding, ciphertext storage, service hosting, CPU/GPU capacity, and other measurable resources. GB is the active resource unit. L0 uses ordinary TCP/IP as its underlay and does not claim that every resource provider is trusted, independently operated, or able to reconstruct an application.

Any node may be malicious. Applications must add recipient encryption, fragmentation, redundancy, verification, and recovery rules appropriate to their data. Those are not automatic properties of resource contribution. The intended AI direction keeps models, data, and agents as independent wallet-addressed roles. See Permissionless cloud and zero-trust applications, Privacy-first Decentralized AI, and the whitepaper.

Layer Minus — privacy routing on L0

Layer Minus uses L0 resources to forward application ciphertext by wallet-linked OpenPGP identity. Wallets, OpenPGP keys, and on-chain route bindings replace a stable origin IP as the durable application route name. A sender posts recipient-encrypted traffic to entry A. A recipient listens through entry C. Both are distinct from mailbox B. This is the A → B / C → B mailbox model.

The carrier is ordinary HTTP between nodes and HTTP or HTTPS from clients to entries. HTTP-shaped traffic reduces reliance on a distinctive custom protocol, but does not guarantee indistinguishability from all Web traffic. TCP/IP remains the underlay. Entry and mailbox roles reduce direct endpoint exposure only when operators and identifiers are sufficiently separated; they do not prove absolute anonymity, censorship resistance, or immunity to traffic analysis. How to use Layer Minus explains how applications select wallets, encryption targets, and envelope contents.

web3:// — application protocol

The web3:// Application Protocol is an application-layer contract: URI grammar, exact wallet/tag resolution, caller-signed requests, correlated encrypted responses, persistent application streams, host-adapter security, and failure semantics. It is not a new Layer Minus wire command; Layer Minus continues to route encrypted application envelopes over L0 resources.

The canonical product and platform model is web3:// under Applications. Linux servers and clients can use conet-l0d; browser and native clients on Windows, macOS, Android, and iOS implement the same protocol contract. conet-l0d is a Linux runtime, not a separate application family.

L1 — shared state and settlement

The CoNET Blockchain is the production EVM network with chainId 224422. It anchors network identity, validator and Guardian state, canonical assets, the cross-chain Treasury, and application settlement. The AAC proof-driven settlement read-only Shadow is production-approved and deployed for observation, but AAC custody, mint/release, and miner-vote cutover remain closed.

Validator consensus, Guardian participation, and application state are related but distinct concerns. In particular, “stealth” describes the objective of reducing publicly exposed network topology; it is not the name of a separate consensus algorithm. Do not label ValidatorDepositRedeem.totalStakedValidatorCount() (~475) as the L1 Beacon set, and do not treat the Guardian registry (~472) as that census either. Those two figures are the L0 / VDR-managed scale. Beacon validator_index values already pass 2000. See L1 decentralization.

L2 — specialized parallel ledgers

CoNET-DLE specifies Decentralization Clusters and parallel atomic ledger classes for asset, storage, and trade activity. The design moves high-frequency application events away from one globally serial execution lane while retaining explicit archive, finality, and settlement rules. EIP-155 Chain ID is unique for the DLE plane (CoNET-DLE Testnet 0x44c45 / 281669). User-visible Group ID for the first archive group is that group’s L1 register transaction hash. Lab M6 added a second live group; its user-visible Group ID is that group’s L1 registerLiveGroup transaction. Neither identifier is CoNET L1 224422.

The L2 section is a digest of the English whitepaper (revision 2026-08-18) and normative specifications. It states design maturity separately from L0/L1 production status. A public lab explorer at https://dle.conet.network/ inspects isolated Archive health, testnet 0x44c45, the bootstrap Group ID hash, lab M6 Clusters = 2, and a non-green 30-day clock chip (pilotStartedAt=2026-08-18T09:53:58.092Z; clock ≠ qualification). The L1 Global Archive Routing Registry is deployed; that does not launch Archive Certificate or asset ingress, and it does not yet register the second lab group.

For an ERC-20 issued elsewhere, entering CoNET L1 and activating DLE are different transitions. Treasury route execution can establish a canonical CoNET representation; DLE asset admission additionally requires its own pool, oracle, registry, gateway, conservation, and release gates.

One composition, end to end

Consider a paid social post:

1. Identity and delivery: the publisher and reader use wallet-linked identities; Layer Minus moves encrypted payloads through L0 entry and mailbox resources.
2. Rights and payment assets: ownership, access rights, and settlement assets can be anchored on L1.
3. High-frequency events: reads, tips, boosts, or revenue shares can be modeled as L2 application events and periodically settled according to a DLE ledger class.

The broader stack supports very different products, but a product need not depend on every layer. SilentPass focuses on network access, CoNET Chat on relationship-private wallet communication, Beamio on wallet and merchant workflows, and web3:// on wallet-addressed application access.

A second composition is wallet-addressed hosting: a browser or native client resolves a web3:// target, signs an Application Protocol request, and submits ciphertext through a healthy Layer Minus entry. A Linux host can use conet-l0d to verify the requester and map the logical port to a loopback Web, API, AI, or TCP application service. Other server runtimes may implement the same protocol contract.

A proposed miner-matched order-book exchange illustrates another composition: externally issued ERC-20s first enter through explicit Treasury routes; admitted canonical assets can then be represented in signed limit orders; miner matchers coordinate deterministic fills; and a contract, rather than the matcher, controls final asset movement. The exchange remains a design study, not a shipped DLE application.

How CoNET differs from adjacent designs

The comparison below is architectural, not a throughput or anonymity benchmark.

Security and engineering boundaries

• End-to-end encryption is recipient-specific. Entry and mailbox nodes must not be trusted with application plaintext.
• Routing metadata still exists. An entry sees the connecting client; a mailbox knows which route it serves. The design limits correlation by separating those roles.
• A successful entry request is not proof of delivery. Mailbox persistence and recipient acknowledgements provide the stronger delivery signal.
• Short sessions trade state for overhead. Fetch-and-Close reduces persistent-flow exposure but is not an anonymity protocol; Chat SSE remains a traffic fingerprint.
• Wallet identity does not solve Sybil resistance alone. Stake, admission, rate limits, or application policy are still required.
• Backend must not trust client-supplied identity headers. A host adapter must strip colliding fields and inject only verified values.
• Specification is not deployment. DLE's normative design, a live L1 endpoint, the DLE lab explorer, and the Application Protocol draft have different evidence levels. The explorer does not make Archive Certificate or asset ingress live.

Evidence map