A quantitative risk assessment of a synthetic decentralized lending protocol. Fraud detection, exploit tail modeling, contagion, and a full solvency simulation, delivered the way Moluscca delivers to a real client.
A client hands us their protocol. We return a defensible number on every risk it carries, each one explained in plain language, with the math shown. Here is that output.
On-chain transactions, wallet behavior, TVL, and fee flows become a structured dataset.
Machine-learning fraud models score every wallet and flag anomalies before loss.
100,000 Monte Carlo futures test solvency against fraud and exploit shocks.
Ten risk dimensions collapse into one weighted, explainable number.
This report: findings, recommendations, and an insurance-readiness verdict.
Nyxus is a high-risk protocol. Its fraud losses are frequent but survivable. Its real threat is a single large exploit hitting a reserve that is too thin to absorb it. The math is fine; the capital buffer is the problem.
Frequency is handled; severity is not. The protocol comfortably absorbs day-to-day fraud, but a 62% insolvency probability is driven almost entirely by rare exploit events whose severity (EVT shape 0.37) is heavy-tailed. The single highest-leverage fix is not better fraud detection, which already works, but a larger reserve or exploit cover sized to the tail, not the average.
No jargon. If you read one section, read this.
A made-up DeFi lending protocol, a place where people deposit crypto and others borrow against it. We invented it to show how we assess this kind of risk. None of the data is real.
Two kinds. Small everyday fraud (stolen funds, fake accounts, laundering), which is frequent but individually small. And rare catastrophic exploits, where an attacker drains a huge amount at once through a bug or a manipulation. The first is a nuisance. The second can kill the protocol.
The protocol easily handles the everyday fraud. But it only stays fully solvent in about 38% of our simulated futures, because its safety reserve of $9M is too small to survive a big exploit. In a bad-luck 1-in-1000 case, a single exploit could reach $23M.
Not more fraud tools, those already catch most bad wallets. The fix is holding a bigger reserve, or buying exploit cover sized to the worst case, not the average. Most of this danger is insurable once it is measured.
Nyxus survives the small stuff easily but is under-insured against the one big exploit that actually threatens it, and that gap is measurable and fixable.
Total value locked over the observation window, with daily fraud losses overlaid. Notice how fraud clusters around confidence shocks, not steady state.
A supervised model scores each wallet's fraud probability from behavior: account age, transaction burstiness, odd-hour activity, and mixer usage. This is the "detect" layer of the product.
Two separate engines: a Loss Distribution Approach for frequent fraud, and Extreme Value Theory for the rare exploit that actually threatens solvency.
Each path runs the protocol day by day, drawing fraud and exploit losses against fee income and the reserve. The distribution of the worst drawdown is the answer.
DeFi protocols are composable, which means they are entangled. A shock to one propagates. We model that spread across a synthetic protocol network.
Each risk scored from likelihood, severity, velocity, and detectability, then weighted into 68.5 (High).
| Risk dimension | Likelihood | Severity | Exp. loss | Score | Band |
|---|---|---|---|---|---|
| Smart-Contract / Exploit | 0.55 |
0.92 |
$9.1M | 76 | Critical |
| Liquidity / Bank Run | 0.50 |
0.85 |
$7.7M | 71 | Critical |
| Fraud / Illicit Flow | 0.78 |
0.60 |
$8.4M | 70 | Critical |
| Market / Collateral Vol | 0.70 |
0.65 |
$8.2M | 69 | High |
| Oracle / Price Manipulation | 0.48 |
0.80 |
$6.9M | 69 | High |
| Contagion / Composability | 0.45 |
0.82 |
$6.6M | 68 | High |
| Regulatory / Compliance | 0.62 |
0.78 |
$8.7M | 67 | High |
| Governance / Key Control | 0.38 |
0.75 |
$5.1M | 59 | High |
| Operational / Infra | 0.42 |
0.55 |
$4.2M | 53 | Moderate |
| Model / Parameter Risk | 0.40 |
0.50 |
$3.6M | 48 | Moderate |
Each scenario re-runs the full solvency model under a specific shock. Survival probability is a fresh 20,000-path simulation per scenario.
| Scenario | Reserve trough | Final reserve | P(solvency) | Outcome |
|---|---|---|---|---|
| Base Case | $5.8M | $5.8M | 74% |
Holds |
| Major Exploit ($8M) | $-2.2M | $-2.2M | 27% |
Insolvent |
| Bank Run (-70% TVL) | $2.2M | $2.2M | 31% |
Holds |
| Stablecoin Depeg | $2.0M | $2.0M | 33% |
Holds |
| Oracle Manipulation | $1.4M | $1.4M | 33% |
Holds |
| Regulatory Freeze | $4.6M | $4.6M | 52% |
Holds |
| Contagion (peer collapse) | $-0.2M | $-0.2M | 18% |
Holds |
| Gas Spike + Congestion | $4.5M | $4.5M | 61% |
Holds |
Exploits and contagion are the killers, not everyday fraud. A major exploit or a peer-protocol collapse pushes solvency sharply down, while routine congestion and even a regulatory freeze are survivable. The protocol should size its defenses to the exploit and contagion cases specifically.
The $9M reserve covers the average, not the exploit. Size it to the 1-in-100 loss ($10M), or buy exploit cover for the gap.
Everyday fraud is self-fundable. The tail exploit is exactly the low-frequency, high-severity shape insurance exists for.
The wallet model (AUC 0.76) should gate high-risk wallets in real time, with the isolation forest catching novel patterns.
Contagion analysis shows Nyxus inherits peer risk. Track exposure to linked protocols as a live risk metric.
Oracle manipulation scores high on velocity. A price-deviation halt caps the damage window before a drain completes.
The bank-run scenario is severe. A withdrawal-rate limit during confidence shocks buys time for the reserve to hold.
Nyxus: High risk (68.5/100). Viable, but currently under-insured against its own tail. Fraud detection works. Solvency does not, because the reserve is sized to the average loss rather than the extreme. Priced and reserved correctly, this is an insurable, survivable protocol. Insurance readiness 55.8, resilience 50.7.