ROADMAP

Structural regime measurement on Ethereum, Polygon, Solana and Avalanche: infrastructure structure × economic demand, four regimes (S1D1 · S1D2 · S2D1 · S2D2), one signed vector per chain. ETH and POL validated; SOL and AVAX in calibration.

Structural measurement extended to Arbitrum, Base and Optimism (sequencer response, blob saturation). Cross-chain Execution Context API live: L1 × L2 × Bridge → certified on-chain context. Python SDK (pip install invarians), labs.invarians.com and agentic.invarians.com (CRE reference integration) launched.

L2 threshold calibration completed on a 30-day production baseline (Arbitrum, Base, Optimism; rhythm, continuity, sequencer_publish_latency), coherent with L1 FPR. Bridge scope opened to variable-latency surfaces (Circle CCTP, Chainlink CCIP) at per-message depth. Agentic pressure baseline collecting since 2026-03-30 — not reconstructable retroactively.

Execution context exposed as one panel. A single direction-agnostic call, GET /v2/panel, returns L1, L2 and bridge states; the agent composes the route it cares about. Invarians exposes states, not directions — each bridge entry carries a unified BS1 / BS2 (or null).

Full EVM centralised coverage: seven chains, 20 variable-latency lanes in one panel (10 Circle CCTP + 10 Chainlink CCIP), unified BS1 / BS2 on panel.bridges[].state. Each entry carries its own calibrated flag and capability_level, so agents reason about raw vs calibrated at a glance. One call: GET /v2/panel.

API restructured around three signed primitives in one payload: Attestation (HMAC-SHA256 envelope), Regime (12 SxDx codes per chain via the structural × demand grid), Delta (per-metric drift over short ~10h and long ~30d EMAs). New endpoints GET /v2/panel + POST /v2/verify, tiered via ?include=. SDK 0.7.0.

CCTP V1 upgraded from health-probe proxy to per-message attestation: each USDC transfer captured with its Circle ECDSA signature, independently verifiable against Circle's attester key. Confidence LOW → MEDIUM on 10 EVM routes. SDK 0.8.0. Retired 2026-05-27 in favour of CCTP V2 (corpus preserved read-only).

CCIP lanes upgraded from aggregate signals to per-message capture: source and destination matched by bytes32 messageId, real send-to-execute latency per lane per direction. Reaches per_message_attested; crypto.anchor stays null until DON multi-sig capture (next step). SDK 0.9.0.

Full migration to CCTP V2 ahead of Circle's V1 sunset (2026-07-31), across 10 ETH-paired routes. Standard mode (finality-bound, ~13-30 min) is the panel default for the RWA audience; Fast is subordinated in observed_fast_mode. New nonce retrieval key, plus a confounded_by_iris_downtime flag that breaks the "BS2 because my sensor is down" tautology. The planned latency cold-start was superseded on 2026-06-01 by event-based structural classification (BRIDGE_STATE_STRUCTURAL_v1.2); CCTP V2 routes are now live BS1/BS2. SDK 0.11.0.

Bridge state moves from a statistical threshold to a structural reading of corridor invariants: BS1 / BS2 now answer "are the corridor's invariants holding", not "is latency above a percentile". Adds stuck-message detection on CCTP Standard (invariant I5, 48 h cap, mechanically derived). Methodology v1.2 signed and anchored.

A Chainlink CRE workflow reads the signed Invarians bridge state and emits an execution verdict — ACT / DEFER / WAIT_FOR_DATA — inside the real CRE runtime. Validated on a single node; multi-node DON consensus is the next step. The consumer pattern (DON fetches the Invarians attestation over signed HTTPS) is the path toward Phase 02.

Delta calibration shown to be chain-type-exclusive (ETH-ARB and ETH-OP corpora, 648-config grid, Benjamini-Hochberg FDR: disjoint precursor sets that don't transfer). The composite drift block is replaced by an explicit per-chain precursors[] array carrying full calibration metadata. Method and three independent tests in the research note: Delta calibration is chain-type-exclusive, 2025.

Crypto-grounding of CCIP: capture the F+1 threshold-signed DON multi-sig on CommitReport (per batch, with per-message Merkle inclusion proofs). capability_level moves per_message_attested → per_message_crypto_anchored, crypto.anchor: "don_threshold_sig". Dual-purpose: the same DON mechanism is what hosts the Invarians attestation in Phase 02.

A Model Context Protocol server exposing Invarians execution context as a native agent tool — an agent on Claude, GPT or CRE calls get_panel_v2() as a first-class capability, no SDK middleware. SDK addresses developers; MCP addresses agents directly.

CCTP confidence moving MEDIUM → HIGH as the 30-day clean window accumulates (target ~2026-06-10); Tier 2 routes (AVAX↔ARB, BASE↔OP) added on sufficient throughput. ETH↔SOL CCIP under passive observation (thinner Solana history, RPC-pipeline dependency) toward Q1 2027.

Solana and Avalanche structural calibration validated, bringing regime classification to full confidence on all four L1s. The signed 12-code grid (live on ETH, POL, BASE, OP since 2026-04-29) generalised to every chain via event-based backtest on documented incidents. Coverage extended to third-party bridges (LayerZero, Across). Pattern Reference enters production: historical frequencies per directional tuple across 7 chains.

A Chainlink Functions layer enables DON-level computation: each node fetches structural metrics from its own RPC endpoint and reaches consensus on the regime vector, with no single computation source. The hybrid model runs both paths at once — API for agents, DON consensus for smart contracts — and CRE workflows extend to mainnet with the full L1 × L2 × Bridge gate. This moves runtime trust from a single operator to a DON; capturing CCIP commit reports in Phase 01 was the operational reference for the transition.

CCIP observability (at per_message_crypto_anchored since Phase 01) 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 composite context — L1×L1 (SxDx × SxDx × BSx_CCIP) and L1×L2 — passed as a signed panel from Agent A to Agent B before any cross-chain action, same verifiable ground truth, no trust required between them. EVM lanes ready at launch; ETH↔SOL completes Q1 2027.

The decentralised network is named INVAR. Invarians publishes structural baselines each cycle; independent nodes, bonded and rewarded in existing assets (ETH, stablecoins), compute their own invariants from raw RPC data and submit regime classifications, rewarded when their output matches consensus (each node signing with a registered Ed25519 key). Invarians keeps the methodology, baseline and reference-publisher role; computation, and with it trust, becomes distributed. Calibration stops being a private asset and becomes a public good, reproducible and verifiable by any participant. No native token exists, and none ever will.