Invarians Labs invarians.com ↗

Substrate observability

Blockchains under AI agent load. What deforms, how it recovers.

Research preview. This page lists the intentions and observable metrics that frame Invarians Labs research. Agentic load measurement is scoped, not yet operational. Substrate response metrics are produced by the Invarians main pipeline (api.invarians.com), not by Labs. No live data is published on this page. Critiques and reproducibility checks are welcomed.

Invarians Labs treats blockchain infrastructure as a material under load. The load comes from AI agents acting on-chain. Like any material, the substrate carries that load through one of four regimes:

Connecting observables to regimes is one of the open empirical questions Labs investigates. Persistent drift in structural rhythm without recovery suggests plastic. Brief deviations that recover suggest elastic. Sustained slow drift under continuing load suggests creep. Formalizing these mappings against ground-truth events is research in progress.

Labs studies the intersection between two sides. On one side, the agentic load: AI agent activity sources that apply pressure on the substrate. On the other, the substrate response: how the chain reacts, already measured continuously by the Invarians main pipeline. Five EVM chains in scope: Ethereum, Polygon, Arbitrum, Base, Optimism. EVM-only scope at Labs because the agentic load sources studied here are predominantly EVM-native (EIP-4337, x402, ElizaOS deployers, Virtuals on Base). Solana and Avalanche, also covered by the Invarians main pipeline, sit out of Labs scope until equivalent agentic sources emerge.

Two sides

Agentic load and substrate response

Agentic load

Sources of AI agent pressure on the substrate (scoped, not yet collected)

  • x402 payments USDC transfers via the x402 HTTP-402 payment protocol, per hour. Direct measurement of agent-to-API micro-payments.
  • EIP-4337 active senders Smart account sender count per hour via the canonical EntryPoint contract. Proxy of programmatic execution.
  • ElizaOS framework activity Contract calls originating from ElizaOS deployer addresses, per hour, on each EVM chain.
  • Olas / Autonolas registry Registry events from the Autonolas agent service network.
  • Virtuals Protocol agents Tokenized agent contract activity on Base.

Substrate response

How the chain reacts, measured by the Invarians main pipeline

  • Structural rhythm Regularity of block cadence, deviation from per-chain nominal.
  • Sequencer publish latency P97 L2 commit cadence on rollups (Arbitrum, Base, Optimism).
  • Beacon participation rate Validator activity on Ethereum L1.
  • Bridge total latency P97 End-to-end transit duration on CCIP and CCTP routes, from source finality to destination inclusion.

These observables feed the live regime classification on api.invarians.com (S2+ / S2- / D2+ / D2- on L1 and L2, BS1 / BS2 on bridges).

Labs does not run its own measurement pipeline at this stage. The substrate response metrics listed above are computed continuously by the Invarians main pipeline against per-chain rolling baselines, exposed via the api.invarians.com signed payload (regime + drift signal). Each metric is normalized per chain, since chain designs are heterogeneous: cross-chain comparisons reflect deviations from each chain's typical level, not raw values. The agentic load sources are listed as research intent. Their integration into a Labs-specific collector is not yet in production.

Open research

Open methodology. Open data. Critiques and reproducibility checks welcome.

A dedicated Labs discussion space is on the roadmap.