LITEPAPER

Before EIP-4844, high L2 activity produced blob-equivalent calldata pressure on L1, creating a weak but measurable fee signal. After Dencun (March 13, 2024), L2 batches post via blob transactions at a structurally lower cost floor. The mechanical coupling between L2 congestion and L1 basefee was severed.

The consequence is systemic: fee monitors, gas oracles, and mempool scanners — every tool built on L1 economic signals — are now blind to L2 sequencer degradation, rollup congestion, and bridge backlog. An autonomous agent executing a cross-chain transaction during a sequencer outage receives no warning from any existing infrastructure signal.

Invarians measures what has no economic signature on L1. It observes structural behavior across three layers — L1 chains, L2 rollups, and bridges — and classifies each independently.

Today, no execution pipeline incorporates this structural signal. The structural layer is absent from every on-chain agent architecture. That is the gap Invarians closes.

Invarians delivers two factual outputs. The agent owns the execution policy — neither primitive prescribes action.

Primitive 1 — Proof of Execution Context (PoEC)

A signed JSON attestation per chain, at sub-minute latency. Each attestation certifies the current infrastructure state:

The four regimes — S1D1, S1D2, S2D1, S2D2 — are measured independently for L1 and L2. Attestations expire after one hour and are independently verifiable at the /verify endpoint. Each attestation certifies one thing: the infrastructure regime at execution time — nominal, or a measured deviation from it.

Primitive 2 — Pattern Reference

The historical record of each execution context. There are 32 possible contexts (4 L1 regimes × 4 L2 regimes × 2 bridge states). For each, the Pattern Reference records which historical events occurred, at what observed frequency, anchored to the public on-chain record where available. Not prescriptive — frequency data and event documentation only. The agent reads the current PoEC, looks up its context in the Pattern Reference, and applies its own policy. Frequencies are calibrated from historical backtest and recomputed as new observations accumulate — the reference evolves with the data, not fixed.

Invarians signals have been backtested against four years of public blockchain data (BigQuery). Two events establish the core empirical case.

The system operates as a pipeline of independent services, each responsible for a single measurement layer. No service reads from another's output — classification is decoupled from collection.

Eight production services run on dedicated VPS infrastructure (Hetzner), writing to a Supabase backend. The oracle API exposes the current attested state and a /verify endpoint for signature validation. All classification logic uses adaptive baselines calibrated per chain — L2 thresholds are independent of L1 parameters.

Coverage: ETH, POL, SOL, AVAX (L1) · Arbitrum, Base, Optimism (L2).

Invarians runs off-chain today. The natural path to bring the Proof of Execution Context on-chain is Chainlink Functions: a consumer contract requests the current PoEC via a DON call — the DON fetches the attestation, verifies the HMAC-SHA256 signature, and returns l1_regime and bridge_state on-chain. The result: execution conditioned on certified infrastructure state.

CRE fit: The Chainlink Runtime Environment (launched March 2026) orchestrates autonomous agents executing cross-chain operations. An agent orchestrating cross-chain execution needs to know whether the target chain is structurally nominal before committing. Market data feeds cover price and liquidity. Invarians covers the infrastructure layer that precedes both — the structural context no fee signal can see post-EIP-4844.

Integration reference →

ETH and POL are production-ready with validated calibration on 4 years of public data. L2 thresholds require a 30-day clean baseline before event-based validation — systematic rollout in progress, Q2 2026. Confidence expands with each validated historical event. Calibration status is exposed directly in the API response — verifiable, not asserted.

The methodology behind Invarians comes from structural field instrumentation — anomaly detection from non-economic signals, calibrated against physical ground truth rather than market consensus. The same logic that detects subsurface stress before surface failure detects sequencer degradation before fee monitors react.

Applied to blockchain: measure structure, not price. The post-EIP-4844 decoupling between L1 basefee and L2 activity was not an engineering discovery — it was the confirmation that structural measurement was the only remaining signal.

In a world where AI can generate any data, the only trusted signal is one grounded in physical measurement — not inferred, not estimated, not produced by a model. Autonomous agents need to know the infrastructure state before acting on-chain. If that state comes from an uncertified API or an AI-inferred estimate, another agent can fabricate or corrupt it. Invarians produces what cannot be fabricated: infrastructure state computed from finalized on-chain data, signed cryptographically before delivery, verifiable independently.

From measurement to proof →