The anatomy of on‑chain reputation.

TL;DR. A lending protocol that doesn't ask for collateral has to ask a much harder question instead: can we trust this wallet? Reputation is our answer. We treat it not as a single number from a single data source, but as the convergence of eight distinct signals - identity, sanctions risk, peer trust, builder credibility, ecosystem engagement, multi-chain transaction history, and net worth - blended into one score on a 0-1000 scale. This article explains what we mean by reputation, how each of our partners contributes, and where we think the primitive is going - because if reputation is the input, credit stops being a function of capital and starts being a function of behavior.

What Reputation Actually Means

In traditional finance, "credit" is a one-line answer to a much larger question. A FICO score is the visible tip of a decades-old apparatus: bureaus aggregating data from lenders, lenders pricing risk against bureau data, courts and bailiffs standing behind the contracts that make the whole loop enforceable. The number is a shorthand for an institution.

On-chain, that institution doesn't exist. There is no bureau, no court, no central source of truth about who you are or whether you pay your debts. Earlier under-collateralized protocols tried to work around this by restricting lending to KYC'd institutions or pre-vetted DAO contributors, sensible for early traction, but structurally limited in addressable market. Reputation-backed credit that scales to pseudonymous borrowers needs a different primitive. What we have is something stranger and more powerful: a decade of public, immutable behavior recorded on networks that anyone can read.

Reputation is what you've actually done, signed and timestamped, on networks that don't forget.

That definition is intentionally narrow. Reputation, as we use the word, is not a vibe or a brand or what someone says about you on a podcast. It is the trace your wallet leaves across blockchains, identity primitives, and trust networks - read carefully, weighted honestly, and combined in a way that survives the failure of any single source.

Three properties matter to us:

• Multidimensional. A single score can be gamed; a profile across many orthogonal dimensions cannot. We read identity, risk exposure, peer endorsement, builder activity, ecosystem engagement, transactional behavior, and asset depth - separately, then combined.
• Adversarial. Anything we measure will eventually be optimized against. We assume sybil farms, wash trading, and laundered-on-chain identities exist - and we lean on partners whose business is detecting exactly that.
• Recoverable. A reputation system that punishes a single mistake forever is useless. Time-decay, repayment history, and new positive signals must be able to lift a wallet back into eligibility - otherwise we'd just be a permanent denylist with extra steps.

Eight Signals, One Score

Under the hood, an Unlloo reputation lookup fans out to eight separate data sources, each one a partner we've integrated deliberately. They run in parallel; the results are normalized to a common 0-1000 scale; the final score is a weighted blend. If a source is offline, the remaining weights re-balance automatically - the wallet still gets a score, just one based on slightly less evidence.

The Partners, and What Each One Tells Us

The choice of partners is not aesthetic. Each one answers a specific question that the others cannot. Below is the full roster, organized by the dimension of reputation each one covers.

Identity: Is there a real person behind this wallet?

Anonymous credit is a contradiction. Not because we need to know your name - we don't, and in most cases we never will - but because we need to know that there is exactly one of you. A reputation system without sybil resistance, protection against fake-identity abuse, where one actor creates many wallets to game the system - is a coupon to mint identities until one of them looks creditworthy. One partner answers this question for us.

Human Passport (formerly Gitcoin Passport) is the most widely adopted proof-of-personhood layer in Ethereum-aligned ecosystems. Its model is a collection of "stamps" - independently issued attestations that a wallet has passed some humanness check: a verified social account, a unique biometric, a government ID hash, a Web of Trust endorsement. No single stamp is sufficient; the score is the sum of how many distinct, hard-to-fake stamps a wallet has accumulated. We pull a wallet's Passport score and normalize it onto our 0-1000 scale. A wallet with no Passport activity isn't penalized - it just gets less identity weight in the final blend.

Risk: Has this wallet already done something disqualifying?

Identity and risk are different problems. A wallet can be perfectly human and still be a known drainer; a wallet can be a fresh address with no stamps and be totally benign. We need a partner whose job is specifically to flag the former.

Webacy is a risk-detection platform for blockchain addresses. Where most reputation services try to answer "how good is this wallet?", Webacy answers the inverse: "how bad is this wallet, and in what specific way?" Their threat feed flags a defined taxonomy of high-severity behaviors - draining contracts, known hacker addresses, OFAC- and DPRK-sanctioned wallets, phishing infrastructure, exploit-linked accounts, and active scam operations. We don't use Webacy as a positive signal; we use it as a hard filter and as an inverted contribution to the score. A wallet flagged with any critical risk is structurally ineligible regardless of how the other seven signals look. A clean Webacy result, combined with everything else, is a meaningful positive.

Social and builder reputation: What have peers said about this wallet?

The cleanest signal in any credit system is what other people who know the borrower already believe about them. On-chain, that signal has three flavors, and we integrate one partner for each.

Ethos Network is a peer-reputation protocol where one wallet can publicly vouch, review, or stake against another. The result is a graph-derived trust score where the weight of an endorsement depends on the reputation of the endorser. Ethos returns a score on a 0-2800 scale; we normalize it down to our 0-1000 and treat it as the closest on-chain analog to a personal reference letter. It's especially valuable for wallets that are otherwise reputation-light: a new wallet vouched for by ten high-reputation wallets has meaningfully different risk than a new wallet vouched for by nobody.

Talent Protocol publishes the Builder Score, a 0-1000 measure of a wallet's track record as a contributor in crypto - shipped projects, verified credentials, GitHub activity, on-chain work history, and human verification status. For Unlloo, this is the closest thing to "employment history" we can get without leaving the browser. A wallet with a strong Builder Score is, by definition, someone whose income is tied to long-term work in the ecosystem - the on-chain equivalent of payroll. We normalize it onto our 0-1000 scale; it's the single most direct measure of professional track record in our stack.

growthepie publishes ecosystem analytics across Ethereum L1 and Layer 2s, and we use their API to read a wallet's history of interaction with vetted projects across multiple chains. Importantly, growthepie's project list is curated and weighted - interactions with a major DeFi protocol count differently than interactions with an unaudited fork. We aggregate these interactions with logarithmic scaling and per-project caps, which significantly raises the cost of gaming by distributing the required effort across projects and chains rather than allowing concentration in a single venue. The signal we extract is, roughly, how genuinely engaged is this wallet with the productive parts of the ecosystem?

On-chain: What does this wallet's own history and cryptographic state actually say?

Three partners turn on-chain primitives into reputation signals. Two of them - Blockscout and Octav - give us the raw material for what we call our custom creditworthiness layer, the part of the score that isn't outsourced to any single reputation provider, but built directly from on-chain behavior and balances. The third, Zeru zPass, ingests an already-computed, AI-derived behavior score for the wallet, condensing cross-chain financial activity into a single portable signal.

Blockscout is an open-source block explorer with public APIs across the major EVM chains, and it's the workhorse of our on-chain analysis. We query Blockscout in parallel across Ethereum, Arbitrum, Base, Avalanche, and Optimism, building a unified view of a single wallet's activity over the last several years. From that data we extract four sub-dimensions: transaction history (volume, frequency, counterparty diversity, wallet tenure), financial behavior (inflow/outflow ratios, stable-vs-volatile exposure, realized P&L shape), reliability indicators (lending protocol interactions, repayment patterns, historical health factors), and DeFi engagement breadth (protocol diversity versus suspicious concentration). The result is a 0-1000 score we compute ourselves - and because it's our model, we can inspect, tune, and back-test it against real loan outcomes in a way we can never do with a partner's closed score.

Octav handles the orthogonal question of portfolio depth: across the chains we care about, what is this wallet actually worth right now? Octav aggregates native balances and token holdings across many chains into a single net-worth view, with price feeds attached. We don't treat net worth as creditworthiness - someone with a $5M wallet may still be a terrible borrower - but it's a useful sanity check on the rest of the picture. A wallet claiming high activity but holding nothing across any chain looks different from a wallet with consistent activity and a real, multi-chain balance sheet.

Zeru zPass is a behavior-intelligence primitive built around zScore, an AI-derived 0-1000 reputation number computed from cross-chain wallet activity by a behavior-centric, no off-chain identity required. The score is anchored in an updatable NFT (the zPass token, live on Ethereum) that makes a wallet's reputation portable across protocols. Where Blockscout reads what a wallet has done and Octav reads what it holds, Zeru zPass arrives as an already-computed score.

How the Eight Signals Combine

Once the eight sources have returned (or failed gracefully), the final score is built in two stages.

First, the six third-party reputation sources - Human Passport, Zeru zPass, Ethos Network, Talent Protocol, growthepie, and Webacy - are merged into a single third-party aggregate. Each one carries a configured weight; if any source is unavailable, the remaining weights are renormalized so the aggregate still sums to one. No single vendor can take the system down.

Second, that aggregate is blended with our Blockscout-derived custom creditworthiness layer using a fixed 60/40 split: 60% third-party, 40% custom on-chain. The split is deliberate. The third-party aggregate is broader and incorporates signals we couldn't reproduce ourselves (sanctions screening, peer vouching, ZK identity). The custom layer is narrower but fully under our control, fully auditable, and directly informed by the metrics that predict default. Neither layer is sufficient alone; together they cover each other's blind spots.

The blend is also a redundancy contract. If all third-party sources fail, the custom layer carries the full weight and the wallet still gets a score. If our Blockscout pipeline fails on every chain, the third-party aggregate carries it. The only scenario where a wallet gets no score at all is when literally nothing returns data - a wallet so new and so absent from every source that there is genuinely nothing to measure. We acknowledge this as a real constraint: the users who most need under-collateralized credit are sometimes the ones with the thinnest on-chain histories. Expanding score coverage to early-stage wallets is an active design priority for our next protocol version.

Why Eight, and Not One

The obvious question - "why not just use the best provider?" - is one we asked ourselves seriously early on. There are two answers, and both come from operating credit systems, not from architecture preference.

No single source is reliable enough alone. Every reputation provider in this space, including the very best, has been wrong, has been gamed, has had outages, or has shipped a model update that re-scored its entire user base overnight. Building a lending protocol on top of a single provider means inheriting all of that fragility. Eight separate providers, with renormalizing weights, means one bad update or one extended outage doesn't break us.

No single source covers every dimension. Human Passport is exceptional at sybil resistance but tells us nothing about lending discipline. Talent Protocol captures builder credibility but doesn't know if a wallet just received funds from a sanctioned mixer. Webacy detects threats but doesn't reward good behavior. Our Blockscout-derived layer measures on-chain history but can't see ZK-proofed off-chain identity. Each partner is a sensor pointed at one part of the elephant. The score is the assembled picture.

A reputation system built on one provider is a single point of failure. A reputation system built on eight, with graceful degradation, is a portfolio, with meaningful redundancy even when individual signals occasionally move together.

What Reputation Is Not

We are explicit with ourselves and with users about what this primitive does not do, because over-claiming on reputation is how protocols get themselves and their lenders hurt.

• Reputation is not a guarantee of repayment. It's a probability adjustment. A high-reputation wallet is meaningfully less likely to default than a low-reputation one, but it is never impossible. The protocol must, and does, price for that.
• Reputation is not identity. Most of our partners deliberately stop short of revealing who you are. A high score tells us a real, unique, productive on-chain actor is on the other side. It does not tell us their legal name or jurisdiction, and we don't want it to.
• Reputation is not static. A wallet that scored 750 last quarter and got liquidated three times since then is a different credit subject today. Our scores are cached for freshness, not for permanence - they refresh automatically at application time and can be triggered by lenders to re-evaluate a borrower's current standing. Reputation is a moving picture.

Where Reputation Goes Next

We think the next three years of on-chain reputation look very different from the last three, and we are building toward four specific shifts.

From snapshot to time series

Today, most reputation scores - including ours - return a single number for "right now." That number obscures everything interesting. A wallet that climbed from 300 to 800 over two years is fundamentally different from a wallet that crashed from 900 to 800 in the last month, even though they read identically as snapshots. The next version of the score should be a trajectory, not a point - and our caching layer is already structured to support exactly that.

From observation to obligation

Reputation today is almost entirely backward-looking: it measures what you've done. The interesting frontier is forward-looking: signed, verifiable commitments to future on-chain behavior, with reputation consequences attached. Once a wallet can stake its reputation on a promise - a repayment schedule, a behavior commitment, an attested off-chain income flow - the primitive starts to look less like a credit score and more like a contract.

From private models to verifiable models

Today, most reputation models are closed boxes. A wallet gets a number, but the function that produced it is opaque - operators ask to be trusted on faith. We hope the next few years bring something more open: score-generating logic that anyone can independently verify produced a given score correctly, without needing to take the operator's word for it. That's where we want this space to go.

From individual to portable

Reputation today is locked into the wallet that earned it. That's both right (you can't trade it) and wrong (you also can't gracefully migrate it across wallets you control, across upgraded smart-account standards, or across identity primitives you adopt later).

The Bigger Thesis

If reputation is the input, then credit stops being a function of capital and starts being a function of behavior. That is the shift we're underwriting.

None of this works without partners. The reason we can ship a serious reputation-backed lending protocol now, when it was impossible even three years ago, is that an entire layer of infrastructure has matured in parallel: identity primitives that respect privacy, risk feeds that detect adversaries, peer-trust networks that aggregate human judgment, builder-credentialing systems that capture professional history, ecosystem analytics that distinguish real engagement from noise, and multi-chain data infrastructure that makes a single wallet's full footprint legible.

We're still early. The score will get sharper, the partner list will evolve, the model will be stress-tested by real defaults. But the architecture is right - and the partners are the proof.