How Invarians works

Most agents act on surface signals.
Invarians measures the substrate underneath.

Signed execution context (S_x^±·D_x^±)L_x×BS_x

Gas price, token price, fee monitors. None of these reveal whether the chain is structurally healthy at the moment of execution. Sequencer slowdowns, validator pressure, bridge stress: all invisible. Invarians publishes the missing layer.

See how it works

Fee monitor view

gas : 25 gwei ✓

price : $3,200 ✓

slippage : 0.04% ✓

mempool : flowing ✓

all clear · execute

Invarians view

L2 regime : S2+D1 ⚠

rhythm shift : +0.27 ⚠

seq publish : 95s lag ⚠

cctp route : BS1 ✓

structural stress detected

1The nominal

Each chain has its own normal.
It is built, not assumed.

Invarians constructs the nominal window per chain from invariants derived from physical metrics of the substrate: structural rhythm, continuity, sequencer publish latency, beacon participation, demand ratios. The metrics are what the chain produces; the invariants are what Invarians computes from them.

No fee, no price, no narrative. The metrics are observed directly on the substrate, chain by chain. The invariants derived from them are recalibrated continuously on a rolling 30-day baseline.

Ethereum has a different nominal than Polygon. Arbitrum's cadence is structurally different from Base and Optimism's. Solana and Avalanche are in active calibration, scheduled for Q3 2026.

The same per-chain logic extends to variable-latency cross-chain surfaces: each Circle CCTP route and each Chainlink CCIP lane has its own calibrated nominal.

nominal_window live · 5 chains

2The drift

The nominal moves
when the protocol evolves.

Drift is structural. It is what the protocol itself does to the chain. Hard forks, EIP implementations, validator client upgrades, network parameter changes: each one reshapes the baseline.

Ethereum post-Dencun is statistically not the same chain as Ethereum pre-Dencun. Drift is intrinsic to evolving systems. It must be tracked, not assumed away.

Invarians runs two timescales in parallel: a short EMA (~10h) captures the local pulse, a long EMA (~30 days) captures the structural baseline. When the long curve moves, the nominal has drifted.

Drift moves the window itself. Divergence is the chain stepping outside. Different questions, different timescales.

nominal_drift 2 timescales

3Divergence

When the chain
steps outside.

A divergence happens when the chain's current state falls outside its calibrated nominal window. Not before. Not later.

The nominal stays where it was. The chain is the one that moved. Invarians captures this moment and qualifies it: which axis fired, in which direction, how persistent.

The agent owns the decision on what to do with it. Invarians qualifies, the agent acts.

Quantitatively, divergence is measured by the shift (section 6).

divergence_event qualified

4Polarity

Stress comes
in two directions.

A signal can deviate two opposite ways. Above the upper threshold (+) or below the lower threshold (-). Polarity makes the distinction explicit.

S2+ means rhythm elongated, sequencer slowing down. S2- means rhythm too short, sequencer overproducing. Same label without polarity, opposite root causes.

For an AI agent the policy must distinguish them. Twelve signed codes per chain instead of four. The substrate has direction.

signed_regime · S1, S2+, S2− 12 codes per chain

5Calibration & thresholds

The boundary moves
with the chain.

Thresholds are not constants. They are the output of continuous calibration on a rolling 30-day window. P97 statistical, falsifiable, reproducible.

Both upper and lower bounds are calibrated where applicable. A chain can deviate in two opposite directions, and each direction has its own boundary.

Recalibrated daily so the threshold tracks the drift, not against it. What is normal for Ethereum today is not what was normal six months ago.

thresholds_p97_rolling recalibrated daily

6The shift

A signed measurement
of where the chain is.

The shift is the quantitative measurement of divergence: shift = ratio_short − ratio_long. A positive shift means the chain is above its 30-day baseline. A negative shift means below.

Where drift is structural (the protocol itself moves), shift is dynamic. Today the dominant driver of shifts is increasingly agentic pressure: bots concentrating activity, AI agents executing within tight rhythms, automated cross-chain orchestration that pushes the chain off its baseline without changing the protocol.

From the shift, two derived signals form the trend: shift_delta answers which direction is the value moving?, shift_magnitude_delta answers is the deviation amplifying away from nominal, or reverting toward it?

The attestation is published on a regular cadence, not as a real-time stream. The trend signal is what guides the agent between cycles. It tells the agent whether the regime is sticky or about to flip back to nominal, and lets it anticipate rather than only react.

shift · shift_delta · shift_magnitude_delta live

7Agentic pressure

Blockchains will deform
under economic pressure of AI agents.

AI agents are starting to execute on-chain at scale. RWA settlements, treasury operations, cross-chain orchestration, automated liquidity routing. The pressure they exert is structurally different from human-driven activity: tighter rhythms, repeated patterns, concentrated load.

This pressure is becoming the dominant driver of chain deformation. Where Dencun reshaped Ethereum's nominal in months through protocol change, agentic pressure will reshape it continuously, through concentrated economic activity that the substrate has never been calibrated for.

Invarians' bet: an observability layer that captures this deformation continuously becomes critical infrastructure as agents scale. The substrate is changing. The agents need to see it.

structural_deformation · agentic_load projection

Putting it together Framework · Phase B.rheo

A blockchain behaves
like a material under load.

Invarians applies the language of materials science to the substrate. The chain carries a load σ, the economic and computational pressure exerted on it, increasingly driven by AI agents. The chain reacts with deformation ε. Four regimes organize all observed responses.

Elastic

Reversible shift

σ varies, ε returns

The substrate bends under σ and returns to its baseline. The shift is reversible. The nominal holds.

Plastic

Structural drift

σ recedes, ε does not

The substrate consolidates the load. The nominal has drifted to a new baseline. Non-reversible under current conditions.

Creep

Slow stress accumulation

σ stable, ε drifts

Continuous low-grade σ produces silent ε accumulation. Neither a clean reversible shift, nor a consolidated drift yet. Irreversible build-up over time.

Rupture

Failure boundary

ε diverges, no return

The substrate fails to absorb. Drift and shift no longer apply, the chain is down (deep sequencer or consensus failure). Not observed to date. Detection protocol is part of the Labs roadmap.

Most agents act on surface signals. Invarians measures the substrate underneath.

Each chain has its own normal.It is built, not assumed.

The nominal moveswhen the protocol evolves.

When the chainsteps outside.

Stress comesin two directions.

The boundary moveswith the chain.

A signed measurementof where the chain is.

Blockchains will deformunder economic pressure of AI agents.

A blockchain behaveslike a material under load.