⬡ OpenChainGraph · Avalanche L1 / Evergreen · 5 nodes · ART-459, ART-494 through ART-498
OpenChainGraph · Avalanche L1 / Evergreen

Avalanche L1 / Evergreen · Verifiable Workflow Guide

If you came here from the Settlement Finality Classifier and it told you FINALITY_RISK_MOVED_OFFCHAIN, this is why: on Avalanche, finality is sub-second with no reorg tier, so the optimistic-rollup challenge-window math this suite ships elsewhere — the seven-day dispute window ART-321 checks, the validity-proof timing ART-290 checks — simply does not apply. There is no window to wait out. That does not mean an Avalanche L1 deployment carries no settlement risk. It means the risk relocated to three places nobody sells receipts for: who is authorized to sign a cross-chain message on the source chain's behalf, who is permitted to transact or deploy on the destination chain, and whether the chain itself stays funded. The five workflows below cover those three places.

Written for the ITGC, internal-audit, and treasury functions of an institution running or joining a permissioned Avalanche L1 (an Evergreen deployment), and for the engineers building one. Everything runs locally in your browser: no backend, no signup, no data leaves your machine. Every result carries a hash, so a counterparty or an examiner can rerun the same workflow on the same inputs and confirm they get the same answer. This is an independent project by Post Oak Labs, not an Avalanche product.

ART-459 · ART-494 → ART-498 · 5 nodes ICM / AWM quorum risk Precompile SoD Fee runway / depletion ASC 850 / IAS 24 related-party Client-side · Zero PII
Why this family exists
The finality question doesn't apply here — three others replace it

Avalanche's institutional pivot centers on Evergreen: a permissioned-L1 suite that lets a bank or consortium run its own subnet-EVM chain with protocol-level access controls, while still settling cross-chain via Avalanche's Interchain Messaging (ICM, formerly AWM). Under Avalanche consensus, a transaction finalizes in well under a second and stays final — there is no probabilistic reorg window and no optimistic-rollup dispute period to model. That single fact is why ART-492 flags FINALITY_RISK_MOVED_OFFCHAIN for Avalanche deployments and points here instead of at a challenge-window calculator.

Settlement risk on a permissioned Avalanche L1 shows up in three other places instead: cross-chain message trust (how many of a source chain's validators would have to collude to forge a message ICM would deliver), who holds protocol-level power on the chain itself (the five subnet-EVM precompiles that gate transacting, deploying, minting, and fee-setting), and whether the chain keeps running (ACP-77's continuous P-Chain fee draws down an L1 balance that can hit zero). None of those questions have a generic answer — each one needs the specific validator set, precompile configuration, or fee schedule an institution actually configured, which is exactly what the workflows below take as input.

Consensus finality
Sub-second, no reorg window — STEP-0 2026-07-30
Cross-chain trust
ICM/AWM · BLS aggregate signature over source validator set
Protocol-level access
5 subnet-EVM precompiles · genesis-configured
Validator economics
ACP-77 · continuous P-Chain fee, no stake bond
OpenChainGraph
Five workflows, one per Evergreen risk surface

Each node answers a distinct question. None of them read a chain, an RPC, or live data — every input is caller-supplied transcribed state, so the same inputs always reproduce the same receipt.

Cross-chain trust
ICM Quorum-Forgery Classifier
Given a source L1's validator stake distribution and the receiving L1's accepted quorum percentage, computes the smallest validator set that could jointly sign a message the receiver would accept — the minimum-colluding-set an operator has to trust when they configure ICM to accept messages from that source. This is the node FINALITY_RISK_MOVED_OFFCHAIN points at.
ART-494 · check_icm_quorum_forgery_risk
→ Open this workflow
Segregation of duties
Precompile Role & Tenant SoD Analyzer
The subnet-EVM precompiles use a role lattice (None < Enabled < Manager < Admin) across five gated functions — TX AllowList, Contract Deployer AllowList, native minter, fee manager, reward manager. Measured 2026-07-30 (AVAX-SOD-1): the shipped ART-459 SoD Matrix Check already expresses this correctly once the lattice is enumerated into the ruleset, so no new node was built. ART-459 ships an Avalanche preset that does the enumeration for you, plus a negative-control preset showing the false-green receipt a naively flattened ruleset would produce over the identical grants.
ART-459 · run_sod_matrix_check · Avalanche preset
→ Open the SoD analyzer
Control ownership
Evergreen Permissioning-Control Classifier
Takes your genesis precompile configuration and validator-manager mode (PoA / PoS / managed) and classifies each supervisory control as protocol-enforced, application-enforced, or absent, with a named gap register. Distinct from the SoD analyzer above: that one asks who holds conflicting power, this one asks where the control lives at all.
ART-495 · classify_avax_permissioning_controls
→ Open this workflow
The 3am question
L1 Continuous-Fee Runway Model
ACP-77 replaced the old 2,000 AVAX stake bond with a continuous, dynamic P-Chain fee per validator, drawn from an L1 balance that depletes and needs refilling. This model takes your validator count, fee-rate and growth assumptions, and infrastructure cost, and computes annual TCO plus the point at which your L1's balance runs out and validation stops. Fee rate is an input you supply, never a baked-in constant.
ART-496 · model_l1_fee_runway
→ Open this workflow
Change-control evidence
Validator Change-Control Receipt
Turns a validator add, remove, or weight-change event — authorizer, approval quorum, effective epoch, prior and posterior weight — into change-control evidence shaped the way the SOX/ICFR family already presents it, including the weight-delta and share-of-total effect and a quorum verdict against your own declared policy.
ART-497 · build_validator_change_control_receipt
→ Open this workflow
Quarter-end close
Consortium Reward-Flow Related-Party Classifier
Classifies reward-manager recipients against a caller-supplied entity-ownership map: same ultimate parent as the issuer, or a fellow consortium member. Emits an unconditional flag table and aggregate amount against materiality, a gap list naming every unresolved recipient, and clearly labelled draft ASC 850 / IAS 24 disclosure-note language — a convenience layer, never the only output, and never legal or accounting advice.
ART-498 · classify_reward_flow_related_party
→ Open this workflow

All four ART-494–498 nodes ship compute_proof_ready: deferred pending a batched GPU proving pass — the receipt and hash are live and reproducible today; the zero-knowledge compute proof binding follows in a later change.

Deployment model
Evergreen, AvaCloud, or self-run: a control-ownership choice, not a vendor comparison

The workflows above take your configuration as input regardless of who operates the underlying L1 — that choice is not a feature checklist, it is a decision about who holds operational control of the surfaces the nodes above measure.

  • Evergreen Avalanche's institutional permissioned-L1 suite. Genesis is expected to carry the TX AllowList and Contract Deployer AllowList precompiles by default — permissioning is the starting posture, not an opt-in. Your institution still chooses the validator-manager mode and every precompile's role assignments; ART-495 and ART-459 measure what you actually configured, not what Evergreen defaults to.
  • AvaCloud A managed L1 deployment service. It changes who operates the infrastructure the L1 runs on — it does not change who holds the precompile roles, the validator-manager keys, or the ICM quorum policy on that L1. Those remain your institution's controls to configure and, when there is a change, to evidence with ART-497.
  • Self-run Running your own subnet-EVM validator infrastructure end to end. Full operational control, and full ownership of the continuous P-Chain fee obligation ART-496 models — there is no managed party absorbing that runway question for you.
Verification state
What is docs-grade and what is reported-only

The mechanics above — sub-second finality, ICM's BLS aggregate quorum, the five subnet-EVM precompiles, ACP-77's continuous fee — are docs-grade facts, re-verified against primary Avalanche documentation 2026-07-30. Several widely cited institutional deployments on Avalanche are not: they are reported in press and vendor materials but not independently confirmed against a primary source, and no workflow on this page rests its output on them.

Reported only A Corda-to-Avalanche L1 migration reportedly carrying over $2B in tokenized real estate and corporate bonds, a reported ~$50M tokenized credit fund, a reported ~$75M tokenized loan, a reported permissioned-subnet pilot at a major bank, a reported tokenized-deposit settlement pilot, and a reported disaster-response platform pilot. None of these are primary-source verified as of 2026-07-30. They are named here only to be labelled — nothing above computes against them.
OpenChainGraph
How verification works on this page

Every tool on this page emits an OpenChainGraph artifact with the following properties, consistent across ART-459 and ART-494 through ART-498.

  • chaingraph_version All nodes emit chaingraph_version: "0.4.0" in their §4 artifact payload, so consuming systems can verify schema compatibility before ingesting the artifact.
  • execution_hash Every node emits a §4 execution_hash covering node inputs, processing parameters, and outputs, computed by the shared canonical hash path. Either party in a bilateral relationship — an operator and a counterparty, an institution and its auditor — can re-run verify_execution_hash against the hash to confirm the node was run as claimed, without sharing the underlying validator set or ownership map.
  • Caller-supplied state only No node here reads a chain, an RPC, or any live data. Validator weights, precompile grants, ownership maps, and fee assumptions are all inputs you transcribe and supply — the same inputs always reproduce the same receipt.
  • MCP access All five nodes are callable over the AINumbers MCP server for agent-driven workflows, using the mcp_name shown on each workflow card above.
Other OpenChainGraph hubs
OpenChainGraph Hub → Tempo Network Guide Settlement Finality Classifier All Tools →