ROADMAP

Structural regime measurement for Ethereum, Polygon, Solana and Avalanche. Two-layer model, infrastructure structure versus economic demand, producing a certified, signed regime vector per chain. Four regimes: S1D1 · S1D2 · S2D1 · S2D2. Attestation v3 live. Calibration status per chain: ETH and POL validated, SOL and AVAX in calibration.

Structural measurement extended to Arbitrum, Base and Optimism, execution profile, sequencer response, blob saturation. Cross-chain Execution Context API live: L1 regime × L2 regime × Bridge state → certified on-chain execution context.

Python SDK deployed (pip install invarians), JavaScript integration examples available. labs.invarians.com launched as a substrate observability research surface (intentions and metric inventory, agentic load × substrate response). agentic.invarians.com deployed as a reference integration of Invarians within the Chainlink CRE stack.

L2 threshold calibration completed 2026-05-01 on Arbitrum, Base and Optimism (rhythm, continuity, sequencer_publish_latency). Statistical calibration on 30-day production baseline, ~3% D2 nominal rate, coherent with L1 FPR. Beacon participation threshold calibrated on Ethereum 2026-05-01. Event-detection validation pending Phase D (on-chain event detection via archive node replay, Q3 2026).

Bridge classification scope: variable-latency surfaces (Chainlink CCIP, Circle CCTP). Circle CCTP upgraded from health-probe proxy to per-message Circle ECDSA capture 2026-05-11 (capability_level: per_message_attested, confidence MEDIUM on the 10 EVM routes). Chainlink CCIP upgraded to capability_level: per_message_attested on 2026-05-12 (source-to-execute matched by bytes32 messageId on the 10 EVM lanes); DON multi-sig CommitReport capture is the next step.

The agentic pressure baseline is collecting on regime-distribution observations across L1 × L2 × Bridge pairs since 2026-03-30 and cannot be reconstructed retroactively.

Execution context exposed as a single panel covering variable-latency cross-chain surfaces. L2↔L2 via Circle CCTP and L1↔L1 via Chainlink CCIP rolled out through Q2.

One direction-agnostic call: GET /v2/panel. The agent composes the exact route it cares about from the panel's L1, L2 and bridge entries; Invarians exposes states, not directions. Each bridge entry carries a unified BS1 / BS2 state (or null when classification is deferred), with a type field distinguishing CCIP from CCTP.

Complete EVM centralised coverage reached late April 2026. Seven chains live, with cross-chain surfaces exposed through a single panel: L2↔L2 via Circle CCTP and L1↔L1 via Chainlink CCIP.

Panel exposes 20 variable-latency cross-chain surfaces: 10 Circle CCTP routes (upgraded 2026-05-11 to per-message Circle ECDSA capture, capability_level: per_message_attested) and 10 Chainlink CCIP lanes (upgraded 2026-05-12 to per-message capture via bytes32 messageId matching, capability_level: per_message_attested, with DON multi-sig CommitReport capture scheduled in the same Phase 01 window). Unified BS1 / BS2 nomenclature on panel.bridges[].state; type, capability_level, and crypto.anchor fields distinguish the protocol, the depth of signal captured, and the cryptographic anchor mechanism.

Unified API: GET /v2/panel, one call returns L1, L2 and every bridge at once. Each entry carries its own calibrated flag and capability_level, so agents reason about raw vs calibrated and aggregate vs per-message at a glance. Institutional grade cross chain coverage, live. Methodology published on the audit trail.

API surface restructured around three explicit primitives in a single signed payload: Primitive 1, Attestation (HMAC-SHA256 envelope, on-chain anchor slot reserved May 2026), Primitive 2, Regime (12 SxDx codes per chain via the structural × demand grid with signed direction extension), Primitive 3, Drift Signal (per-metric shift, shift_delta, shift_magnitude_delta over short ~10h and long ~30d EMAs, plus composite drift per axis).

Two new structural classifying observables: beacon_participation on Ethereum (Beacon Chain validator participation rate) and sequencer_publish_latency on each L2 (sequencer health, unblocks S2 detection on rollups). Both emit shift_available: false during their slow-EMA stabilization period (~30 days).

New endpoints: GET /v2/panel + POST /v2/verify. Tiered exposure via ?include=core|diagnostic|full. SDK Python invarians 0.7.0 on PyPI. v1.1.0 endpoints sunset 2026-06-30.

CCTP routes upgraded from health-probe proxy to per-message attestation tracking. Each USDC transfer between EVM chains is now captured individually with its Circle ECDSA secp256k1 signature, retrievable via GET /v2/cctp/attestation/{message_hash} and independently verifiable against Circle's published attester public key.

Real per-message latency P90/P99 per route. A pending queue handles the asymmetry between collector cycle (10 min) and source-chain finality (~13-19 min on Ethereum-anchored sources): no message is lost across cycles. Each bridge entry exposes a new capability_level field (per_message_attested for CCTP today, aggregate reserved for surfaces where per-message capture is not yet implemented) plus structured metrics and crypto objects. CCIP reaches the same per_message_attested level on 2026-05-12 (see next entry).

Confidence flag moves from LOW (proxy) to MEDIUM (per-message, EVM only) on 10 EVM CCTP routes. SDK Python invarians 0.8.0 on PyPI with client.get_cctp_attestation().

CCIP lanes upgraded from aggregate signals to per-message capture. Source CCIPSendRequested and destination ExecutionStateChanged are matched per message via the bytes32 messageId, yielding real send-to-execute latency per lane per direction. A pending queue handles messages whose execute lands later than the 10-min collector cycle, mirroring the CCTP pending model.

CCIP capability now matches CCTP's per-message depth. crypto.anchor remains null until DON multi-sig capture (next step). Captured messages retrievable via GET /v2/ccip/message/{message_id}. SDK invarians 0.9.0 ships with client.get_ccip_message().

Crypto-grounding of CCIP signals, respecting Chainlink's native cryptographic chain of trust (different mechanism from Circle's single-attester ECDSA).

CommitStore capture. CommitReport events emitted on the destination CommitStore carry the F+1 threshold-signed DON multi-sig over the batch Merkle root. Captured per batch, with per-message Merkle inclusion proofs derived from the batch tree.

API outcome. CCIP capability_level upgrades from per_message_attested to per_message_crypto_anchored, with crypto.anchor: "don_threshold_sig" on every CCIP bridge entry. The verification mechanism is intrinsically different from CCTP's single-attester ECDSA, and the API exposes that distinction honestly via the crypto.anchor enum.

Strategic alignment with Phase 02 CRE integration. The same DON threshold-signature mechanism Invarians will capture on CCIP bridges is the architecture that will host the Invarians attestation flow via Chainlink CRE in Q3-Q4 2026. Monitoring CCIP at DON-grade depth and operating Invarians attestation through a DON converge: bridge observability becomes operational reference for the path toward decentralised attestation. Capturing CCIP commit reports is dual-purpose by design.

A Model Context Protocol server exposing Invarians execution context as a native tool, callable by AI agents without human middleware or SDK integration. SDK addresses developers. MCP addresses agents directly.

An agent running on Claude, GPT or Chainlink CRE can call get_panel_v2() as a first-class tool in its stack, the same way it calls any other capability.

CCTP confidence flag moves from MEDIUM (per-message, EVM only, since 2026-05-11) toward HIGH as the production-grade 30-day window of clean per-message latencies accumulates (target ~2026-06-10). Tier 2 CCTP routes (AVAX↔ARB, BASE↔OP) added once observed throughput supports per-message calibration.

ETH↔SOL CCIP lane has been live since May 2025 but historical data on the Solana side is thinner (~12 months) and requires the Solana RPC pipeline (same dependency as CCTP Solana). Passive observation accumulates until classification can be committed, targeting Q1 2027. Pattern Reference enriched with directional patterns observed in production.

Solana and Avalanche structural calibration validated on internal production data, regime classification reaching full confidence on all four L1 chains. July 2026.

Signed regime classification reaches every chain after event-based backtest on documented incidents (sequencer halts, agentic cascades, censorship events). The 12-code grid (S1D1, S1D2+, S1D2-, S1D2±, S2+D1, S2-D1, S2+D2+, S2+D2-, S2+D2±, S2-D2+, S2-D2-, S2-D2±) has been live on Ethereum, Polygon, Base and Optimism since 2026-04-29 (statistical activation). Q3 2026 brings Solana and Avalanche into the same scheme alongside their L1 calibration, and refines the existing lower bounds via event-based validation. Arbitrum gets a multi-dim demand workaround (size + tx based) since the native sigma_ratio is degenerate on Nitro.

Coverage extended to third-party institutional bridges: LayerZero Enterprise, Across, with the same directional tuple (source regime × destination regime × bridge state). ETH↔SOL CCIP observation continues toward Q1 2027 MEDIUM confidence.

Pattern Reference enters production: documented historical frequencies per directional tuple, per bridge type, across 7 chains.

The Invarians API continues serving off-chain agents via Python and TypeScript SDKs, existing integrations unchanged. In parallel, a Chainlink Functions layer enables genuine DON-level computation: each node independently fetches structural metrics from its own RPC endpoint and reaches consensus on the regime vector without a single computation source.

CCIP delivers the DON threshold-signed regime cross-chain, an execution context produced by decentralized consensus, verifiable on-chain at the point of execution. CRE workflows extended to mainnet with the full L1 × L2 × Bridge execution gate.

The hybrid model operates both paths simultaneously: API for agents, DON consensus for smart contracts. The full calibration methodology and backtests have been public since Q2 2026 (agentnorthstar.com); Q4 closes the loop by moving the runtime trust from a single operator to a DON. Capturing CCIP commit reports during Phase 01 (Q2 2026) provided the operational reference for this transition: the same threshold-signature mechanism now hosts Invarians' own attestation.

CCIP observability, already live at per_message_crypto_anchored capability since Phase 01 (DON multi-sig CommitReport captured per batch, Merkle inclusion proof per message, RMN cursed status per chain), is now wired into the A2A coordination layer. A cursed CCIP lane freezes routing independently of statistical classification, invisible to any fee monitor.

This unlocks a unified cross-chain composite context across two patterns: L1×L1, (SxDx)_ETH × (SxDx)_SOL × BSx_CCIP, the structural regime of two L1 chains crossed with the operational state of the bridge connecting them; and L1×L2, (SxDx)_ETH × (SxDx)_ARB/BASE/OP × BSx_CCIP, extending existing rollup coverage to CCIP-grade bridge certification.

Designed for A2A agent coordination: Agent A passes a signed panel to Agent B before any cross-chain action. Both agents operate on the same independently verifiable ground truth, no trust required between them. API and SDK unified: panel.bridges[] exposes every CCIP lane alongside CCTP routes, same interface for L1 or L2 endpoints.

EVM lanes production-ready at launch. ETH↔SOL activates with observation-phase certification, calibration completing Q1 2027 as the lane accumulates sufficient structural history.

Invarians publishes updated structural baselines on every computation cycle, the quantified reference of nominal behaviour for each chain. Independent nodes receive these baselines, compute their own invariants directly from raw RPC data, and submit their regime classification. Nodes whose output matches consensus earn token rewards. Stake is required to participate.

Invarians retains the methodology, the historical baseline, and the role of reference publisher. The computation, and with it the trust, becomes distributed. Regime classifications are no longer certified by a single server. They are verified by an independent network against the same source data, each node signing its output with a registered Ed25519 key.

Calibration stops being a private asset. It becomes a public good, reproducible and verifiable by any participant.