The Verifiable Performance Layer
for Finance

Performance should be provable, not claimed.

Every day, capital moves on performance claims that nobody can check: backtests on private data, track records in PDFs, cropped screenshots. Aevum Protocol is the verification layer that fixes this.

The core of the protocol is the Verifiable Backtest Oracle (VBO), a set of smart contracts on Ethereum mainnet. A strategy’s rules are fingerprinted and committed on-chain before it runs. It is then measured only on market data that arrives afterward, priced by an independent oracle, and the result is recorded as a permanent certificate that anyone can check on Etherscan.

This paper describes what is live today, what is still target design, how the system resists the most common ways performance is faked, and where it goes next: from crypto trading strategies to forecasts, bots and autonomous AI agents.

Finance already runs on machines.

Global foreign-exchange markets turn over about $9.6 trillion a day (BIS Triennial Survey, April 2025). Industry estimates put the majority of US equity volume in the hands of algorithms, and in DeFi, market makers, liquidation bots and arbitrage agents generate much of on-chain activity. Most financial activity is already executed by software.

The next step is already under way: AI moving from tool to actor.

• Autonomous trading agents managing capital without per-trade human sign-off.
• Agents negotiating and executing smart contracts on behalf of their principals.
• Data agents selling insights through on-chain micropayments.
• Cross-protocol arbitrage agents that move faster than any human approval loop.
• AI-managed yield strategies rebalancing across DeFi around the clock.

The infrastructure these systems run on was designed for human-initiated transactions. The most important missing piece is not speed or cost. It is trust: a way for people, and other agents, to know whether a strategy or agent actually performs the way it claims.

Anyone can produce an impressive backtest.

A trader shows you a strategy that returned 80% last year. You have no way to know whether it is real. Three failure modes make most performance claims unreliable, even when nobody is lying on purpose:

• Curve fitting. Parameters are tuned until years of history look perfect. The strategy was designed around the data, and it breaks the moment it trades live.
• Cherry picking. Fifty backtests are run and the three that worked are shown. The real hit rate stays hidden.
• Look-ahead bias. The backtest quietly used information that didn’t exist at trade time.

The tools we use as “proof” don’t solve this. Screenshots can be faked in seconds. Spreadsheets have no audit trail. Broker statements are only as good as the broker. Platform dashboards mean nothing off-platform. Audited accounts didn’t stop Madoff. None of them prove the strategy existed before the results were known.

Five problems for autonomous finance

The cost

Capital flows to the best storytellers, not the best performers. That is the inefficiency Aevum addresses.

From “trust a person” to “verify a system.”

A few innovations permanently reduced how much finance depends on trust. Each one replaced a person you had to believe with a system you could check.

On-chain, you can already trade, hold assets, lend and borrow without trusting an intermediary. What you cannot do is prove that a strategy performed the way its owner says it did. That missing piece matters more every year, as more capital is allocated by algorithms and, increasingly, by AI agents that other people and other agents need to evaluate.

Commit first. Results second.

The VBO enforces one ordering rule: a strategy is judged only on data that did not exist when its rules were locked. Ethereum’s block timestamps enforce that order, not Aevum.

The output is a certificate on Ethereum mainnet that anyone can verify in seconds, without trusting Aevum or the strategy owner.

Two independent proofs in every record.

A certificate combines two layers of evidence that come from different parties.

Layer 1 — Process integrity (Aevum)

• The strategy’s fingerprint existed on-chain before the window opened.
• The result covers only the sealed forward window.
• The outcome is attested on-chain and can’t be edited afterward.

Layer 2 — Data integrity (Atlas Oracle)

• Prices come from Atlas Oracle, CoinMarketCap’s oracle, aggregated across 905+ exchange sources.
• Signed price payloads are verified on-chain at commit and attestation.
• The strategy owner never supplies the prices used to judge them.

What a certificate records

Time-weighted return is used so that deposits and withdrawals can’t inflate performance. When this rule was introduced, Aevum restated its own first testnet certificate from +6.93% to +3.82%.

A certificate is not a promise of future returns. It is proof that a specific claim wasn’t manufactured after the fact.

Honest commitments, dishonest selection.

Pre-commitment stops anyone from editing a strategy after seeing results. It does not, on its own, stop someone from committing many strategies and marketing only the winner. This is the main attack on any verification system, and Aevum addresses it directly.

Live today

• Public lifetime history. Every identity has a page listing every commitment it has made, attested or pending.
• Rate limits. Each identity can make at most three commitments per 30 days.
• Bonds. Every commitment posts a bond that becomes slashable if the commitment is abandoned.
• Disclosure. Certificates show how many commitments the identity made in the same period.

The open problem: many wallets

No contract can prove that two wallets belong to the same person. Someone willing to spread strategies across many identities can still hide failures. Aevum’s position is to make that expensive and visible rather than to claim it is impossible. A single certificate from a new identity is weak evidence. A long, unbroken record is strong evidence.

How a certificate should be earned.

Pre-commitment proves when a strategy was locked. Methodology standards define how much evidence a certificate represents. The standards below are the target framework for certificate tiers; today’s certificates record the commitment, window, return, benchmark and regime.

Three layers of trust

• Cryptographic pre-commitment live: the strategy is hashed and committed before the scoring window opens, so it can’t be retrofitted to the window.
• Held-out forward window live: results come only from data that arrives after the commit.
• Determinism target: logic re-run in a sandboxed, reproducible environment across independent validators, with a certificate issued only on 2/3+ agreement.

Minimum forward windows by strategy type

Too short a window rewards luck; too long makes verification impractical. Target minimums follow the roughly 4:1 in-sample to out-of-sample ratio used by systematic funds, weighted by trade count and regime coverage.

Regime coverage

Calendar days and trade counts are proxies for what matters: did the strategy survive different market conditions? Certificates are designed to report the regimes crossed on two axes: volatility (low, medium, high, extreme) and trend (strong trend, weak trend, mean-reverting, choppy, crisis). Today’s contract records a single regime label per window (Bull, Bear, Chop, Neutral or Unknown).

Labels committed before the outcome

A regime label assigned after the fact can be chosen to flatter a result. The target design commits the label at window open, computed only from prior data, and requires agreement between independent signals:

• Realized volatility over the prior 30 days.
• Trend strength (ADX) over the prior 14 days.
• Momentum (rate of change) over the prior 10 days.

A label is committed when at least two of three signals agree. During transitions, when signals conflict, the prior label holds for a short grace period (3–5 days) until consensus returns. Some boundary trades are scored under the prior label; the guarantee that the label precedes the outcome is preserved.

Certificate tiers

Full certificate contents (target)

• Strategy hash committed before the window, with block reference and timestamp.
• Forward-window performance on data the strategy never saw.
• Historical backtest results across non-overlapping periods and a walk-forward efficiency ratio.
• Regime coverage signal and certificate tier.
• Validator signatures and a methodology transparency score.

A track record that follows the trader.

On most chains, an agent is indistinguishable from any other wallet. A bot with 10,000 verified trades looks the same as one deployed five minutes ago. AgentIdentity live on mainnet gives every trader and agent a persistent on-chain identity that certificates attach to.

Identity record

Reputation mechanics target

R(t+1) = R(t) × decay + w1·PnL_score + w2·Sharpe_score + w3·Consistency_score

• Decay prevents inactive agents from coasting on old results.
• Risk adjustment: raw returns without Sharpe consideration are penalized.
• Consistency rewards stable performance over lucky outliers.
• Scores are public, and slashing applies to confirmed policy violations.

Reputation updates pass through ReputationController, which applies a 24-hour timelock. In v1, scores are advisory (see Security & trust model).

Safe autonomy, and a market for proven strategies.

Verification is the foundation. Three further components, currently deployed on the Sepolia testnet, extend it toward a full stack for autonomous agents. All three are target for mainnet and will not launch before an audit.

Permissioned execution

Running an autonomous agent today usually means handing a bot a private key with no native limits. Aevum’s execution framework (AgentVault) enforces limits at the contract level rather than in bypassable application code.

Gas abstraction

True autonomy requires an agent to pay for its own execution without manual top-ups. The design includes agent gas accounts funded from earnings, automatic refills at configurable thresholds, batched micro-operations, predictable fees for strategy cost modeling, and sponsor accounts through which a principal pre-funds an agent under its execution policy.

Strategy marketplace

A marketplace where only VBO-certified strategies can be listed, so every listing comes with a verifiable record instead of a PDF.

Aligning excellence with influence.

Today a permissioned attestor signs certificates. The target design replaces that with an independent validator set with stake at risk.

• Validators stake $AEV and earn fees for honest attestation; false attestation is slashed.
• Certificates require an N-of-M quorum of independent validators.
• The set moves from permissioned to permissionless as it grows, following the same hardening path price oracles took.

Performance-weighted selection (Proof of Performance)

Earlier drafts proposed Proof of Performance as the consensus mechanism of a dedicated chain. On Ethereum, the same idea applies to how attestation weight is assigned: validators with a verified record of honest, reliable operation earn more influence.

V_weight = Stake_AEV × (1 + α × PerformanceMultiplier)

The multiplier is derived from a rolling verified record: risk-adjusted performance where relevant, uptime and liveness, and a slash-free history. To resist sybil attacks, the multiplier requires a long verified history (for example 12 months) before it applies.

Everything is public.

Four contracts have been live on Ethereum mainnet since September 3, 2026. They are written in Solidity 0.8.28 on OpenZeppelin v5.

A further set of contracts (AEVToken, TokenVesting, AevumDAO, AgentVault and AgentMarketplace) is deployed on the Sepolia testnet only. They will not be deployed to mainnet before a professional audit.

Appendix A — testnet contracts (Sepolia)

Testnet addresses are for development only. The same address can hold a different contract on Sepolia and on mainnet, so always use the mainnet table above for production. Live app: app.aevumprotocol.io.

How a commitment works on-chain

// 1. in the browser
strategyHash = keccak256(strategyText)

// 2. on Ethereum (payable: bond; signed Atlas price payload appended to calldata)
commitStrategy(strategyHash, forwardWindowDays)
  → commitmentId, priceAtCommit, block.timestamp

// 3. after windowEnd (attestor)
revealAndAttest(commitmentId, resultsHash, returnBps, regime, metadataURI, note)
  → permanent certificate

Honest about who you’re trusting.

A verification layer is only as trustworthy as the keys behind it. Aevum publishes its trust assumptions rather than hiding them.

• Ownership. The VBO is owned by a 2-of-3 Safe multisig with a hardware-wallet signer. No single key can change its settings. The remaining contracts are being moved to the same multisig.
• Attestation. Today, certificates are attested by a permissioned attestor operated by Aevum. Pre-commitment and independent pricing limit what an attestor can misrepresent, but this is still a trust assumption, and it is disclosed as one.
• Reputation scores are advisory in v1. The ReputationController supports multiple oracles and a 24-hour timelock, but until independent operators join, scores should be read as advisory rather than decentralized consensus.
• Testing. The mainnet contracts pass their full Foundry test suite and have been through internal review passes. Hexens has been selected for a professional audit, which begins at the first funding close.

Attack vectors

Audit commitments

• Professional audit (Hexens) before any further mainnet deployment; token contracts audited before any launch.
• A dedicated review of the oracle and attestation path.
• A public bug bounty program, sized at token launch.
• Formal verification of core identity and attestation logic as the protocol matures.

Verified bots for everyday investors.

AVA is Aevum’s consumer product: done-for-you crypto trading bots that trade Bitcoin, Ethereum and Solana, hosted 24/7 and controlled from a phone. Users connect Coinbase with one tap. Funds never leave their account, and AVA never has withdrawal access.

Every Bitcoin bot ships with a VBO certificate, so a customer can check that the bot’s performance is real rather than taking a screenshot on faith. Ethereum and Solana certificates follow VBO v3. AVA launches in Q4 2026 in three tiers: Basic ($199 + $19/month), Pro ($499 + $39/month) and AI ($999 + $79/month).

AVA serves two purposes: it brings verification to people who would never read a smart contract, and it creates steady demand for certificates.

An economic layer, not a fundraising tool.

Decentralized attestation needs incentives. Validators need a reason to do the work and a penalty for lying. $AEV is the planned protocol token for exactly that role.

• Validator staking. Validators stake $AEV as collateral. Honest attestation earns fees. False attestation is slashed.
• Fees. Certificate fees can be paid in $AEV or stablecoins. 50% of protocol fees are designed to be burned.
• Governance. Holders govern validator admission, certification standards and protocol upgrades.
• Supply. Hard cap of 1,000,000,000 tokens.

Proposed parameters

Proposed allocation

Burns apply to certificate (oracle) fees and marketplace fees. The token contract launches with transfers disabled until the network is live. All parameters are proposals and may change after audit and legal review.

Progressive decentralization.

Governance is designed to move from the founding team to token holders over time, through the AevumDAO contract target, currently on Sepolia with proposals limited to pre-approved contracts.

Path

• Today: the core team operates the protocol; contracts are owned by a 2-of-3 multisig.
• After token launch: DAO governance activates for treasury and parameter decisions.
• Mature network: full DAO governance with no special team veto.
• Long term: constitution-locked core parameters require an 80% supermajority to change.

Where verification sits.

Aevum sits where algorithmic trading, decentralized finance and AI agents meet. Each is large on its own, and each has the same unsolved problem: performance claims that can’t be checked.

Landscape

Starting with crypto trading. Built for all of finance.

The long-term goal is for verification to be the default: any strategy, forecast, bot or AI agent that asks for trust or capital should be able to point to a record anyone can check.

Built by people who lived the problem.

Jonathan Quintero, Founder & CEO. Self-taught engineer who started coding in March 2026. He built and ran live algorithmic trading bots on Coinbase, Kraken and Binance.US, saw firsthand how backtests fall apart in live markets, and took Aevum from idea to four mainnet contracts and a CoinMarketCap data partnership in six months.

• Built and ran live BTC trading bots (DCA, intraday and swing) on Kraken, Coinbase and Binance.US.
• Built an 18-signal intelligence stack combining on-chain data, macro signals and sentiment, with LLM-based confidence scoring in production.
• Built the backtesting frameworks behind those bots, and learned firsthand how far backtests can drift from live results.
• Completed Harvard CS50x. Founded Aevum at 19.

Alexis Alvarez, Marketing & Community. Leads content, social and community.

Team build plan

Aevum Protocol Inc. is a Delaware C-corporation based in Colorado. Contact: jonathan@aevumprotocol.io.

What we changed, and why.

The Aevum whitepaper was first published in June 2026; v0.5 was the last version before this one. Every topic it covered is carried forward in this version. Where plans changed or claims no longer held, we updated them openly rather than silently. The original remains available: read v0.5 (archived).