Live operational scope, May 2026 · API v2.0

The method demonstrated on public and proprietary data.

Every calibrated threshold Invarians uses in the live panel is derivable from public BigQuery datasets or from open collector schemas. Full pipeline published: extraction query, calibration script, expected output. API v2.0 exposes three primitives in a single signed payload: Attestation (HMAC), Regime (12 signed codes per chain), Drift Signal (per-axis composite).

Invarians captures strictly physical metrics from L1 chains, rollups and variable-latency bridge surfaces (CCIP, CCTP), substrate signals on structure and on demand. Narrative and economic metrics (gas price, token price, TVL) are excluded by design.
A nominal is computed per chain, divergences from that nominal are detected against a published calibration, and the result is exposed as a signed attestation to AI agents. Agents gain structural awareness of the chains they operate on and can schedule execution inside the most nominal windows to secure their transactions. Invarians is focused on cross-chain workflows for DeFi risk dashboards, DAO treasury operators, intent solvers, and Chainlink CRE integrators, not daily trading. Institutional RWA settlement is a Phase 02-03 target, gated on the hybrid Chainlink Functions DON deployment.
View operational scope → How to reproduce

01: The reproducibility problem

A signed attestation is only as trustworthy as the pipeline that produced it.

Any closed signal infrastructure eventually asks for trust on faith. Methodology without runnable pipeline, published numbers without public data source, live service without cross-check path. Evidence replaces all three with a single commitment: anything in the live panel is reproducible end to end by a third party on public data.

Risk 1
Methodology without runnable pipeline
A document that claims thresholds but cannot be executed. A third party has no way to confirm the numbers came from the method described.
Risk 2
Thresholds without public data source
Numbers published on proprietary extractions. The reader is asked to trust the extraction as well as the calibration. No independent falsifiability.
Risk 3
Live panel without cross-check path
Signed attestations are verifiable against the signer, but the parameters embedded in the signer become a black box. Closed stack under HMAC.
The answer
End to end pipeline on public data
Extraction query on a public dataset, open calibration script, published expected output. Anyone can re-run the pipeline and confront the numbers served by the live panel. That is what this page packages.

02: Live operational scope

Two scopes, all fully reproducible.

Exactly what is calibrated, functional today, and backed by a runnable reproduction kit. No overclaim, no roadmap dressed as evidence.

Ethereum L1 regime
● Certified, HIGH confidence
✓ 12 signed codes since 2026-04-29 (S1, S2+, S2- × D1, D2+, D2-, D2±) ✓ tau structural rhythm + continuity, M1 = 5.07 ✓ pi demand, sigma + size + tx with calibrated lower bounds ✓ Public BigQuery, crypto_ethereum ✓ 34 697 windows, 2020 to 2024 ✓ TPR 100 percent on 4 canonical events ✓ FPR tau + pi = 1.23 percent
Open reproduction kit →
Polygon L1 regime
● Certified, MEDIUM confidence
✓ 12 signed codes since 2026-04-29 ✓ tau M1 = 12.60, pi M1 = 3.59 (Phi = 720, production-aligned) ✓ TPR 100 percent on 4 of 4 events ✓ FPR documented at 14.57 percent (elevated, transparent) ✓ Public BigQuery, crypto_polygon ✓ 28 744 windows, 2020 to 2023
Open reproduction kit →
L2 rollups, ARB / BASE / OP
● Certified, MEDIUM confidence
✓ 12 signed codes since 2026-04-29 (multi-dim demand: sigma + size + tx + complexity + gas_complexity) ✓ sequencer_publish_latency calibrated 2026-05-01 on batch_gap_seconds (3rd structural axis) ✓ Halt thresholds: ARB 600 s, BASE 480 s, OP 1800 s ✓ FPR 30d = 0 percent on all three chains ✓ Method: LAG window function on L1 inscription timestamps, no invariant-cadence sampling bias ✓ Documented historical halts (ARB 2024-12-21, OP 2025-09-14, BASE 2024-11) clear thresholds by x4 to x6
Reproduction kit: calibration_log entry #034 (available after repo sync)
Variable-latency bridges, CCIP / CCTP
● Phase Preliminary, since 2026-05-04
✓ BS1 / BS2 unified nomenclature on 10 Circle CCTP routes (P97/14d on circle_api_latency_ms, confidence LOW) ✓ 10 Chainlink CCIP lanes exposed raw, classification deferred until sustained throughput ✓ RMN cursed (binary protocol-level safety override) available on every CCIP lane regardless of calibration → Lifecycle: P97/25d MEDIUM ~2026-05-29, P97/30d HIGH ~2026-06-03
Open reproduction scope →
Public data only 3 primitives: Attestation + Regime + Drift Signal Updated 2026-05-05  ·  M1 = metric stability score. CCTP preliminary thresholds committed 2026-05-04 on P97/14d circle_api_latency_ms (10 routes, confidence LOW, lifecycle to MEDIUM and HIGH within ~30 days). L2 sequencer thresholds committed 2026-05-01 by envelope on the 30-day batch_gap distribution.

03: Proof

Live stress events feed, since the baseline started 2026-03-17.

Every hour, Invarians classifies the substrate of Ethereum and Polygon L1 chains and the Arbitrum / Base / Optimism L2 rollups. Each contiguous block of non-nominal regime is captured as a stress event with multi-axis classification: which axis fired, what it threatens, how long it lasted, whether it correlated across layers. Composite L1×L1 events (Ethereum × Polygon stressing simultaneously) are flagged separately. The table below is the full archive, refreshed every 60 seconds, no curation, no hand-picking. Variable-latency bridges (CCIP, CCTP) are not yet emitted as stress events here; the detector extension lands at CCTP confidence HIGH (~2026-06-03) plus the CCIP RMN cursed binary override.

—
Total events
—
Ongoing now
—
Cross-layer composite
—
Stress events (non-noise)
Chain
Layer
Severity
Auto-refresh every 60 s. Last update — —
Started Chain Layer Regime Severity Duration Layer scope Trigger
Loading stress events…

Three canonical moments from the archive worth pointing out.

★ 2026-04-10 07:00 UTC
Base and Optimism both in S2D1 simultaneously.
Substrate anomaly on Base and Optimism at the same hour while Ethereum L1 stayed nominal. An L1-only execution context would not have surfaced this at all.
★ 2026-04-18 14:00 UTC
rsETH cascade, the canonical D2± signature.
On Ethereum L1, sigma and size moved up while tx_count dropped sharply: fewer transactions, but each one much larger. With the legacy 4-state grid, the hour read S1D1 (nominal). With the 12 signed codes (live since 2026-04-29), the same conditions would have emitted S1D2±, the composition asymmetry signature of an agentic concentration cascade (DeFi liquidations, MEV searcher dominance, stablecoin depeg arbitrage). A multi-dim demand axis catches what a scalar threshold misses. Article in preparation.
★ 2026-04-27 19:59 to 2026-04-29 03:31 UTC
Optimism soft sequencer slowdown (48h, x2 cadence). Drift Signal beats regime.
18 batch_gap observations >= 648 s clustered over 48h on OP, cadence shifted from typical ~6 min to ~11 min (x2). Not a halt: max 732 s stays below the 1800 s S2+ threshold by design. The regime grid, calibrated for halt-only, does not flip. The Drift Signal exposed in API v2.0 (Primitive 3) is the right mechanism for this class of event: sustained positive shift on the structural axis, capturable by shift_magnitude_delta once the slow EMA stabilizes (~30 days post-launch). Canonical case study reserved for empirical validation of the Drift Signal. Article published.
The classification, the multi-axis taxonomy and the derived severity rules are documented in the calibration repository. The same logic that produced this archive runs hourly on the live substrate, with no curation between detection and display.

04: How to reproduce

Five steps, same pipeline Invarians runs internally.

The reproduction path is identical across the L1 backtest scopes, only the dataset differs. For Ethereum L1 and Polygon L1, the source is BigQuery public data. The variable-latency bridge reproduction kit (CCIP / CCTP) publishes when calibration reaches confidence HIGH (~2026-06-03).

01
Pick a scope
Ethereum L1 regime or Polygon L1 regime. Each has its own folder under Data with a self-contained README. The variable-latency bridge kit (CCIP / CCTP) is in Phase Preliminary; the full reproduction publishes at confidence HIGH (~2026-06-03).
02
Pull the data
For Ethereum, open BigQuery console, paste the extract SQL (extract_eth.sql), run, export CSV. For bridges, either run the open collector for at least 25 days against the three L2 RPCs or load a provided sample into Postgres.
03
Run the calibration
For Ethereum, run python backtest_eth.py. The script lives in the calibration repository and takes the exported CSV as input. For bridges, execute calibrate_native_p97_30d.sql as a transaction on the Postgres instance holding the signals.
04
Read the output
Ethereum produces a backtest_results CSV with per-window classification across the 12 signed codes (S1, S2+, S2- × D1, D2+, D2-, D2±) and aggregate TPR / FPR. Bridges produce a BEFORE / COMPUTED / AFTER audit showing the P97 thresholds applied per chain. L2 sequencer thresholds reproduce by running the LAG window function on `ans_l2_adapter_signals.l1_block_timestamp` and comparing the 30-day distribution to the published thresholds.
05
Confront to expected_thresholds.md
Each scope folder holds an expected_thresholds.md listing the exact numbers the live panel uses. Match within numerical noise confirms the pipeline. Divergence beyond noise is a published falsification, which is the intended use of this kit.
Every Invarians attestation carries calibration: { version, methodology_hash } in its payload. The methodology hash points to the published method that produced the thresholds. Evidence is the path to re-derive those thresholds independently, with no credentials required for the Ethereum scope.

05: Out of scope, in archive

What is not on this page, and where to find it.

Invarians' methodology work is broader than the current live panel. To keep evidence strictly aligned with the operational scope, chains and bridges not yet served in the live panel are not shown here. They remain public in the calibration archive.

This page
Live operational evidence
Ethereum and Polygon L1 regimes, Arbitrum / Base / Optimism L2 regimes, and the variable-latency bridge surfaces (Chainlink CCIP, Circle CCTP). Calibrated, functional, and reproducible end to end on public data. Anything claimed here is runnable against the live API v2.0 service.
Bridge scope · Phase Preliminary
CCTP lifecycle to confidence HIGH
CCTP routes carry preliminary BS1 / BS2 classification on Circle attestation API since 2026-05-04 (P97/14d, confidence LOW). Lifecycle to MEDIUM (~2026-05-29) and HIGH (~2026-06-03) per methodology.md §13. The stress-events detector extension to bridges (CCTP BS2 + CCIP RMN cursed override) lands at HIGH.
In archive
Methodology validation and parallel tracks
Solana backtest (proxy v1 for sigma, tx_count unavailable, partial). Avalanche calibration pending (M1 insufficient). CCIP statistical classification deferred to Q3-Q4 2026 (distribution dégénérée actuelle, 8/10 lanes saturate at cap; RMN cursed binary override remains the actionable signal). All documented at the calibration repository.
Archive reference : github.com/agentnorthstar/calibration. Methodology section 4.4 lists the per-chain parameters, section 12 (signal taxonomy) documents what is strictly excluded by design (narrative and economic metrics).

06: Documents backing the live scope

Three documents, all canonical sources.

The minimal set of public documents a reproducer needs to verify the two live scopes end to end. Everything else, including methodology validation on chains not yet functional and the Polygon backtest archived pending bridge calibration, is in the calibration repository.

METHODOLOGY
Signal architecture
OFFLINE and ONLINE pipeline, SxDx framework, per-chain parameters section 4.4, M1 stability score, signal taxonomy section 12 (physical only, no narrative).
BACKTEST ETH
Ethereum 2020 to 2024
34 697 windows. TPR 100 percent on 4 of 4 canonical events (DeFi Summer, NFT Mania, The Merge, Shanghai). FPR tau + pi = 1.23 percent.
GITHUB
Full repository
Scripts, SQL queries, calibration log, protocol watch, archive backtests for chains not yet functional.

Reproduce. Falsify. Feedback.

Run the pipeline.
Publish falsifications.

Public data sources, open SQL queries, open thresholds. Any reproducer with a GCP free tier or 25 days of collector samples can confront the numbers published here.

Data kit → Open an issue invarians.com