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 trackers, and mempool scanners, every tool built on L1 economic signals — are now blind to L2 sequencer degradation, rollup congestion, and bridge backlog. An AI 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 three factual outputs in a single signed payload. The agent owns the execution policy, no primitive prescribes action.

Primitive 1, Attestation

Every panel response carries a HMAC-SHA256 envelope (payload_hash + signature + key_id + anchor) covering the full payload. Independently verifiable via POST /v2/verify. The signature makes the certified execution state provable, not just observable. Anchor populated on-chain from May 2026.

Primitive 2, Regime classification

A single panel covering every observed chain and bridge. Each panel entry carries its own status, timestamp, and regime code:

Per-chain regime is a 2-axis tuple on the SxDx grid: structure axis (S1 nominal, S2+ structural high, S2- structural low) and demand axis (D1 nominal, D2+ demand high, D2- demand low, D2± composition split), giving 12 signed codes per chain on both L1 and L2. Ethereum extends the S axis with beacon_participation; L2 chains extend it with sequencer_publish_latency (3rd structural classifying observable). Bridge state is independent: unified BS1 (nominal) / BS2 (degraded) across all variable-latency bridges (CCIP, CCTP); the type field on each bridge entry distinguishes the protocol. L1/L2 entries refresh over a 1-hour rolling window; bridge entries refresh every 10 minutes. Panels are independently verifiable at POST /v2/verify with body { payload, signature }.

Primitive 3, Drift Signal

Regime classification answers "what is the substrate doing now?" but not "is it about to flip?". The Drift Signal closes that gap. For every classifying observable, the panel exposes (in diagnostic mode) a MetricBlock with ratio (short EMA), ratio_long (long-term ~30-day baseline), shift = ratio - ratio_long (current deviation magnitude, signed), shift_delta (raw direction of value movement between cycles) and shift_magnitude_delta (whether the deviation is growing or shrinking). The composite drift object aggregates per axis (structural · demand, plus their delta and magnitude_delta), giving the agent a single fitness-for-action signal: substrate stable, mild drift, active drift, or imminent transition probable. Pattern Reference (frequency lookup over 32 historical execution contexts) is a complementary catalog, candidate primitive 4 in v2.x.

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 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 execution context on-chain is Chainlink Functions: a consumer contract requests the current execution context via a DON call, the DON fetches the signed panel, verifies the HMAC-SHA256 signature, and returns the relevant panel.l1[].regime and panel.bridges[].state on-chain. The result: execution conditioned on certified infrastructure state.

CRE fit: The Chainlink Runtime Environment (launched March 2026) orchestrates AI 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. AI 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 →