aere-contracts

Solidity smart contracts for Aere Network (chain ID 2800) and their Hardhat test suites.

This repository is part of the Aere Network "verify yourself" core. The deployed contract bytecode is already public on-chain (readable via eth_getCode and the block explorer), so publishing the source here lets anyone read, compile, and re-test the exact logic that runs on mainnet. Nothing in this repository can move funds or impersonate the network: there are no private keys, no mnemonics, and no deployment credentials. Every network account is supplied at runtime through environment variables (see hardhat.config.js).

Scope, stated up front

Aere Network runs post-quantum cryptography natively on mainnet: four signature verifiers, Falcon-512 (0x0AE1), Falcon-1024 (0x0AE2), ML-DSA-44 (0x0AE3) and SLH-DSA-128s (0x0AE4), and the SHAKE256 hash (0x0AE5) have been live as precompiles since block 9,189,161 (corrected 2026-10-28: this listed SHAKE256 among the signature verifiers). You can call them yourself against https://rpc.aere.network without asking us for anything.

Consensus on chain 2800 is QBFT with post-quantum layers on top of it, each dated (since block 25,080,000, AIP-22 stage 5, headers no longer carry ECDSA commit seals and are admitted on the Falcon-512 seal of their proposer alone; a commit still carries the ECDSA commit seal of its validator as an additional check). Since block 13,014,000 every anchor block carries a certificate of validator post-quantum seals whose 32-byte digest sits under the block hash (Falcon-512, with SLH-DSA-SHA2-128s as well since 2026-09-04, at every 128th block since 2026-09-29; every anchor was 128 blocks apart from block 17,225,968 to block 20,746,736). Since block 20,715,632 (2026-09-29) a node rejects an anchor block with fewer than seven valid seals per scheme, as many as the quorum of the ten validators; the minimum was three (f+1 of nine, not a quorum) from 2026-08-14 and six from 2026-08-21. Below block 24,148,752 (AIP-22 stage R, 2026-10-24) those seals proved that the validators signed the block, not that it was decided (AIP-22 erratum of 2026-10-01); from that block a certificate that carries the quorum of one round proves the decision. Every validator refuses a commit without a valid Falcon-512 seal of its author, and since block 24,300,000 (AIP-22 stage 4, 2026-10-25) the Besu-derived validators, nine of the ten, count a consensus vote only through the verified post-quantum seal of its author. (Corrected 2026-10-28: until that date this paragraph opened by calling consensus on chain 2800 classical secp256k1 ECDSA QBFT on every block, and said without a date that the seals prove signing, not decision.) That checkpoint lives in the aere-node repository (patch 0003 and anchor/), not in this one: nothing in this repository makes consensus post-quantum, and the chain is not "post-quantum consensus" without that qualification (the per-block quorum claim published on 2026-08-15 was withdrawn on 2026-08-19).

Two further limits worth knowing before you judge anything else here:

• 0x0AE6 (ML-KEM-768) and 0x0AE7 (Falcon HashToPoint) are testnet only. They are not active on mainnet.
• The on-chain zero-knowledge verifiers are classical BN254. They are broken by Shor's algorithm like any other elliptic-curve construction, and we do not describe them as quantum-safe.

The network is operated by ten Foundation-run validators since 2026-09-11 (f=3, commit quorum 7 of 10; nine, with quorum 6 of 9, from 2026-08-09, and seven before; corrected 2026-10-28: this said nine), nine of them on the Besu-derived client and the tenth, since 2026-09-11, on a second, Nethermind-derived client that proposes and seals blocks, so its Nakamoto coefficient is effectively one today. That is a real limitation, it is on the roadmap, and it is not fixed by any code in this repository.

Layout

• contracts/ the Solidity sources, grouped by domain (pqc, pqfinality, settlement, lending, governance, paymaster, oracle, mpc, zkverify, and more).
• test/ the Hardhat test suites, including unit tests, fuzz/invariant/property tests, and the cryptographic known-answer test (KAT) vectors under test/fixtures/ and test/*.json.
• hardhat.config.js the compiler and network configuration. Keys and RPC URLs are read from environment variables only; the sole literal mnemonic present is the well-known public Hardhat default (test test ... junk) used for local test signers.
• package.json the dependency manifest.

contracts/pqc-evidence/: on-chain equivocation evidence (AIP-22 section 6; added 2026-09-23)

PqEquivocationEvidence records, permanently and without any owner, that ONE validator row of a post-quantum registry epoch sealed two different consensus messages of the same type for the same height and round. It proves the row into the registry hash pinned on chain for the epoch in force (an 11-byte read-only getter, slot 0), rebuilds both signed pre-images exactly as the clients do, and verifies both Falcon-512 seals through the chain's precompile at 0x0AE1. contracts/pqc-evidence/test/MockFalcon512.sol is a stand-in for that precompile used ONLY by the local tests (test/evidence/); real seals are checked against the live precompile. test/evidence/ExercitiuEchivoc.java is the program that produced the drill seals below, with the clients' own signing code. Measured on the public testnet 28001 on 2026-09-23, not on chain 2800: a drill instance recorded a PREPARE and a COMMIT equivocation of drill keys (322,404 and 303,103 gas), and an instance for the testnet's own registry epochs proves the rows of the live epoch and verifies live anchor seals through the contract. The procedure and the addresses are in the AIP-22 stage records (aere-research).

Verify

npm ci            # installs the exact locked dependency tree
npm run compile
npm test

npm run lists every entry point this repository offers. Until 2026-08-03 that list was empty, because package.json carried no scripts key at all: npm run answered with zero lines and exit code 0, which reads as "this project has nothing to run" beside 155 test files. The available entries are now:

Command	What it runs
npm run compile	hardhat compile
npm test	hardhat test, the whole suite
npm run test:pqc	the Falcon, SPHINCS+ and ML-DSA known-answer tests only
npm run coverage	coverage over the economic-core cluster, the one scope we measure
npm run clean	hardhat clean

These are entry points, not claims of a green result. Read the next paragraphs before quoting anything from them.

MEASURED 2026-08-03 against this published tree, on Node v24.14.1 and npm 11.11.0: npm test aborts before a single test runs. test/shutter-mempool-v2.test.js and test/shutter-mempool-v3.test.js both require('../scripts/shutter-crypto'), and scripts/ is one of the directories deliberately excluded from publication, so Mocha fails to load and the run ends with MODULE_NOT_FOUND. Placing that one helper back into scripts/ makes the suite load and start executing normally, which is how the cause was confirmed rather than guessed. npm run test:pqc is unaffected and passed 101 tests in 32 seconds on the same tree. Until the helper is either published or the two suites stop depending on it, treat the full-suite result from this repository as NOT MEASURED, not as passing and not as failing.

Use npm ci, not npm install. ci installs the exact tree recorded in package-lock.json and fails if the lockfile and package.json disagree, so you are running against the same dependency versions we measured. install is free to resolve newer versions and will quietly give you a different tree.

Step-by-step instructions with the expected output at each step, the measured pass and coverage numbers, and the required tool versions are in REPRODUCE.md in the aere-docs repository, not in this one. An earlier version of this paragraph referred to REPRODUCE.md as though it sat beside this README, which would have sent a first-time reader looking for a file that is not here.

Read that document before quoting any number from this repository, because two of the four CI jobs are currently RED and it says so explicitly: the full test suite does not complete on a 16 GB machine, and the whole-repository coverage figure is NOT MEASURED. Note that the first of those two reasons is the one observed on our internal tree, which still has scripts/. On this published tree the full suite fails earlier and for the different reason given above, so memory is not the explanation a reader here should accept. Any "99.8% coverage" figure ever associated with this project was never produced by a coverage tool.

The same three commands run on every push through the CI pipeline, which is a pair of files:

• .gitea/workflows/ci.yml runs on git.aere.network, the canonical host.
• .github/workflows/ci.yml runs on GitHub mirrors.

Gitea Actions scans both directories and prefers .gitea/; GitHub scans only .github/. The two files are identical except for the upload-artifact version, and each file's header explains why they are split and records which jobs are currently passing and failing.

To reproduce against the live chain, point AERE_RPC_URL at the public RPC (https://rpc.aere.network, chain ID 2800) and compare compiled bytecode to the on-chain code. The full reproducibility manifest lives in the aere-docs repository (REPRODUCE.md).

Following citations in this repository

Paths cited in this repository are written relative to the document that cites them, or relative to a side-by-side checkout of the published Aere repositories, in which case they begin with a repository name such as aere-docs/REPRODUCE.md. MEASURED on 2026-08-03, every cited path in this repository resolves to a published file: CITATIONS-UNRESOLVED.md is the enforced list of the ones that do not, and it currently has zero entries. It is generated by a check, not maintained by hand, so it cannot silently fall behind.

What is deliberately not here

Deployment scripts, deploy keys, .env files, infrastructure configuration, and business or operations material are intentionally excluded. This repository is source plus tests only.

License

MIT. See LICENSE.