Much has been said about LLM context.

A language model without context produces generic responses, sometimes wrong, sometimes absurd. We call it hallucinating. The solution is known: give it context. Documents, history, fresh data. The model improves.

But there is another type of agent. Not an assistant that answers. An agent that acts.

It does not produce text. It executes transactions. It manages funds. It triggers protocols. And it too needs context. A different kind of context. The context of the infrastructure it operates on.

What an agent sees, and what it doesn't

A concrete example.

On Jun 20, 2024, the Arbitrum sequencer went down. For 37 minutes, transactions were no longer processed in normal order. Simultaneously, the basefee on Ethereum reached 1649 times its baseline level.

A rebalancing agent that triggers at that exact moment knows none of this. It observes an asset price, it sees an opportunity, it sends a transaction. The transaction lands in a degraded context: abnormal latency, unpredictable cost, reduced execution guarantees.

The result is not what was expected.

This is not a bug in the agent's logic. It is a decision made without infrastructure context.

The structural problem

Autonomous agents operate on public blockchains. These blockchains are not fixed rails. Their behavior varies: congestion, gas cost, validator activity, bridge state between chains.

These variations are measurable. They are structured. They follow regimes: stable or degraded states, recognizable and classifiable.

But this information is not natively accessible to an agent. There is no standardized interface today that tells it: "Infrastructure is in nominal regime" or "The Arbitrum bridge is degraded, wait."

An experienced human can read the signals. An agent cannot. Not without help.

The hypothesis Invarians is tracking Autonomous agents modify the infrastructure they operate on.

Not individually. At scale. Hundreds of thousands of agentic operations per hour, concentrated on certain chains, at certain moments, produce measurable pressure on execution regimes.

This is a hypothesis. Invarians has been tracking it since March 30, 2026, with live data across five chains: Ethereum, Arbitrum, Base, Optimism, Polygon.

If the hypothesis holds, agentic load is itself a context signal. Agents operating at scale create the deformation that subsequent agents will need to factor into their decisions.

Invarians calls this measure epsilon(t). It is the third primitive under construction, following L1 structural attestation and pattern reference.

What Invarians does Invarians classifies blockchain execution regimes in real time.

For each covered chain: a structural state across two dimensions, at every moment. These states are certified, timestamped, verifiable. Not floating scores. Qualified contexts.

The complete reference frame is L1, L2, and Bridge. The three layers that compose the infrastructure of a multi-chain agent. When all three layers are classified, the agent has a complete view of the environment it is about to act in.

This context can be consumed via API, integrated into an SDK, or exposed directly to agents via MCP.

The gap this fills

An LLM without context hallucinates.

An agent without infrastructure context acts in the dark. It may execute a transaction at the wrong moment, on a degraded network, through a bridge under stress, with costs that match no prior estimate.

This is not a logic error. It is a context error.

Without context, an LLM hallucinates. Without context, an agent destroys value.

Invarians — On-chain Execution Context for Autonomous Agents

invarians.com

Much has been said about LLM context.

A language model without context produces generic responses, sometimes wrong, sometimes absurd. We call it hallucinating. The solution is known: give it context. Documents, history, fresh data. The model improves.

But there is another type of agent. Not an assistant that answers. An agent that acts.

It does not produce text. It executes transactions. It manages funds. It triggers protocols. And it too needs context. A different kind of context. The context of the infrastructure it operates on.

What an agent sees, and what it doesn't

A concrete example.

On Jun 20, 2024, the Arbitrum sequencer went down. For 37 minutes, transactions were no longer processed in normal order. Simultaneously, the basefee on Ethereum reached 1649 times its baseline level.

A rebalancing agent that triggers at that exact moment knows none of this. It observes an asset price, it sees an opportunity, it sends a transaction. The transaction lands in a degraded context: abnormal latency, unpredictable cost, reduced execution guarantees.

The result is not what was expected.

This is not a bug in the agent's logic. It is a decision made without infrastructure context.

The structural problem

Autonomous agents operate on public blockchains. These blockchains are not fixed rails. Their behavior varies: congestion, gas cost, validator activity, bridge state between chains.

These variations are measurable. They are structured. They follow regimes: stable or degraded states, recognizable and classifiable.

But this information is not natively accessible to an agent. There is no standardized interface today that tells it: "Infrastructure is in nominal regime" or "The Arbitrum bridge is degraded, wait."

An experienced human can read the signals. An agent cannot. Not without help.

The hypothesis Invarians is tracking Autonomous agents modify the infrastructure they operate on.

Not individually. At scale. Hundreds of thousands of agentic operations per hour, concentrated on certain chains, at certain moments, produce measurable pressure on execution regimes.

This is a hypothesis. Invarians has been tracking it since March 30, 2026, with live data across five chains: Ethereum, Arbitrum, Base, Optimism, Polygon.

If the hypothesis holds, agentic load is itself a context signal. Agents operating at scale create the deformation that subsequent agents will need to factor into their decisions.

Invarians calls this measure epsilon(t). It is the third primitive under construction, following L1 structural attestation and pattern reference.

What Invarians does Invarians classifies blockchain execution regimes in real time.

For each covered chain: a structural state across two dimensions, at every moment. These states are certified, timestamped, verifiable. Not floating scores. Qualified contexts.

The complete reference frame is L1, L2, and Bridge. The three layers that compose the infrastructure of a multi-chain agent. When all three layers are classified, the agent has a complete view of the environment it is about to act in.

This context can be consumed via API, integrated into an SDK, or exposed directly to agents via MCP.

The gap this fills

An LLM without context hallucinates.

An agent without infrastructure context acts in the dark. It may execute a transaction at the wrong moment, on a degraded network, through a bridge under stress, with costs that match no prior estimate.

This is not a logic error. It is a context error.

Without context, an LLM hallucinates. Without context, an agent destroys value.

Invarians — On-chain Execution Context for Autonomous Agents

invarians.com