> For the complete documentation index, see [llms.txt](https://docs.verio.network/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.verio.network/protocol/architecture.md).

# Architecture

A shared evidence layer between existing AI infrastructure and the parties that rely on it.

Verio treats evidence about AI as shared infrastructure rather than as a feature of any single platform. Models stay where they are published and inference stays with the providers that serve it. Verio standardizes how artifacts are identified, how executions are recorded, how records are checked, and how evidence connects to payment.

## Layers

| Layer                                                       | What lives here                                                                                                                                                                                                                                                                         |
| ----------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Parties that rely on AI outputs**                         | Applications and agents choose a verification class per request. Routers and marketplaces rank providers by evidence. Enterprises and auditors query lineage and receipts. Researchers and evaluators cite and reproduce results.                                                       |
| **Verio on chain** (contract modules on Base)               | Artifact Registry, Receipt Anchor, Staking and Slashing, Dispute Module, Settlement, Reputation Events. Supported by a public randomness beacon (drand) and receipt data availability.                                                                                                  |
| **Verio off chain** (services run by independent operators) | Registration client, provider adapters, verifier network, indexers and explorer. Content commitments, signed receipts, and evidence live here; weights and compute never move.                                                                                                          |
| **Existing AI infrastructure** (unchanged)                  | Model hubs and storage (Hugging Face, object storage, decentralized storage, local mirrors). Execution providers (inference APIs, enterprise clusters, decentralized compute networks). Evidence backends (hardware attestation, zero-knowledge proofs, refereed-delegation verifiers). |

## Design principles

**Artifact-first.** Registering artifacts, reading passports, and emitting signed receipts require no token and no migration, so any hub, provider, or application can join incrementally. $VERIO enters only where evidence needs economic backing: payment, collateral, and governance.

**Inspectable, off-chain verification.** Every verification mode states what it checks and which assumptions remain: a signature proves what a provider claimed, while a replay checks whether the result reproduces. The chain holds only compact commitments, identities, collateral, disputes, and settlement.

**Assurance scaled to risk.** Replicating every inference would erase the economics of shared execution, while checking nothing leaves high-value actions exposed. Applications choose an assurance class per request and see its cost in advance.

## On-chain modules

Six contract modules hold shared state, deployed on **Base**, an Ethereum layer-2 network with low fees, native USDC liquidity, the Ethereum Attestation Service as a system contract, and precompiles for BLS12-381 and P-256 signature verification. Anchoring a million receipts in batches of 10,000 costs well under one dollar at September 2026 gas prices.

| Module                   | Holds                                              |
| ------------------------ | -------------------------------------------------- |
| **Artifact Registry**    | Artifact IDs, commitments, parents, publisher keys |
| **Receipt Anchor**       | Batch roots, receipt counts, fee totals            |
| **Staking and Slashing** | Collateral, bonds, covered capacity                |
| **Dispute Module**       | Audit selection, replays, escalation rounds        |
| **Settlement**           | Escrow, holdbacks, upstream shares                 |
| **Reputation Events**    | Audits, disputes, slashing history                 |

## Randomness and data availability

Public randomness and data availability protect the audit process. Each batch is bound to the first **drand quicknet** round published a fixed delay after its commitment is final, so no provider knows which receipts will be audited when it commits them; beacon signatures are verified on chain. Full receipts are published to a data-availability network for the challenge window, and a selected receipt that is not revealed counts as a failed audit.

## Off-chain services

* **Registration client:** signs manifests from existing release pipelines.
* **Provider adapters:** translate API records into canonical receipts and batch them.
* **Verifier network:** replays sampled and challenged receipts, runs redundant execution and known-answer tasks.
* **Indexers and explorer:** serve passports and evidence, and mirror receipt data.

Each service can be run by independent parties.

## Core components

* [Artifact Passports](/protocol/artifact-passports.md)
* [Provenance Graph](/protocol/provenance-graph.md)
* [Inference Receipts & Execution Profiles](/protocol/inference-receipts.md)
* [Verification Classes](/protocol/verification-classes.md)
* [Verified Evaluation](/protocol/verified-evaluation.md)
* [Programmable Settlement](/protocol/programmable-settlement.md)