Chart 1 from the Nyxus Protocol Risk Assessment: DeFi fraud and exploit risk analysisMOLUSCCARisk Assessment · Product Demo
PRODUCT DEMONSTRATION · This is a sample of a Moluscca risk assessment deliverable. The subject is synthetic; the methods, models, and output format are exactly what a real engagement produces.
Sample Deliverable · DeFi Fraud & Exploit Risk

Nyxus Protocol:
can it survive its own fraud and exploit risk?

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.

SubjectNyxus Protocol (synthetic)
SectorDeFi lending / on-chain
Window540 days observed
Overall risk68.5 / 100 · High
This is a product demo. Nyxus is fictional and all data is synthetic, generated by simulation. What is real is the method: every model below is a standard technique, and this is the exact format and depth a Moluscca engagement delivers.
What this demo shows

The product, in one glance

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.

01

Ingest

On-chain transactions, wallet behavior, TVL, and fee flows become a structured dataset.

02

Detect

Machine-learning fraud models score every wallet and flag anomalies before loss.

03

Simulate

100,000 Monte Carlo futures test solvency against fraud and exploit shocks.

04

Score

Ten risk dimensions collapse into one weighted, explainable number.

05

Deliver

This report: findings, recommendations, and an insurance-readiness verdict.

01 · Executive Summary

The verdict, on one screen

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.

Overall risk
68.5
High, 10-dimension weighted
P(solvency · 18mo)
38%
reserve survives to horizon
P(insolvency)
62%
reserve wiped by shocks
Fraud model AUC
0.80
wallet-level detection
VaR (95%)
$19M
peak drawdown vs $9M reserve
CVaR (95%)
$22M
avg loss in worst 5%
Exploit tail ξ
0.37
heavy-tailed (EVT)
1-in-1000 exploit
$23M
modeled extreme single loss
◆ AI-generated insight

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.

02 · In Plain Language

The whole thing, explained simply

No jargon. If you read one section, read this.

What is Nyxus?

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.

What's the danger?

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.

What did we find?

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.

What's the fix?

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.

◆ One sentence

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.

03 · Protocol Activity

What the data shows

Total value locked over the observation window, with daily fraud losses overlaid. Notice how fraud clusters around confidence shocks, not steady state.

Chart 2 from the Nyxus Protocol Risk Assessment: DeFi fraud and exploit risk analysis
Simulated TVL path with daily fraud loss. Peak TVL $92M, 3 exploit events over the window. Fraud clusters at volatility spikes, not uniformly.
04 · Fraud Detection

Scoring every wallet before the loss

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.

Chart 3 from the Nyxus Protocol Risk Assessment: DeFi fraud and exploit risk analysis
Detection ROC curve. AUC 0.76: the model separates fraud from legitimate wallets well above chance.
Chart 4 from the Nyxus Protocol Risk Assessment: DeFi fraud and exploit risk analysis
What drives a fraud flag. Mixer usage, burstiness, and odd-hour activity dominate, matching known illicit-flow patterns.

Logistic regression + random forest

Two models: logistic for a transparent, explainable baseline; random forest to capture nonlinear interactions.
logit(fraud) = β₀ + β·[age, tx, value, night ratio, mixer, burst]
Base fraud rate 6% of wallets. Logistic AUC 0.795, random forest AUC 0.760. Isolation Forest, an unsupervised backstop, independently recalls 27% of fraud with no labels, useful for novel patterns the supervised model has not seen.
05 · Loss Modeling & Tail Risk

The everyday, and the catastrophic

Two separate engines: a Loss Distribution Approach for frequent fraud, and Extreme Value Theory for the rare exploit that actually threatens solvency.

Chart 5 from the Nyxus Protocol Risk Assessment: DeFi fraud and exploit risk analysis
Annual aggregate fraud loss (Poisson frequency, Lognormal severity). P99 $8.89M. Frequent, bounded, survivable.

Extreme Value Theory · exploit tail

Peaks-over-threshold: a Generalized Pareto fit to exploit severities above the 90th percentile.
P(X−u > y | X > u) = (1 + ξy/σ)^(−1/ξ)
Shape ξ = 0.373 confirms a heavy tail. Modeled 1-in-100 exploit $9.7M, 1-in-1000 $23.2M, both large versus the $9M reserve.

Loss Distribution Approach

S = Σ Xᵢ, N ~ Poisson, Xᵢ ~ Lognormal
Mean annual fraud $7.03M, tail to $9.68M at P99.9.
06 · Solvency Simulation

100,000 futures for the reserve

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.

Chart 6 from the Nyxus Protocol Risk Assessment: DeFi fraud and exploit risk analysis
Peak reserve drawdown across 100k simulations. Reserve $9M, VaR95 $18.9M, CVaR95 $22.4M. Mass beyond the reserve line is where the protocol becomes insolvent.

Why Monte Carlo

P(solvent) = (1/N) Σ 1[ reserveₜ > threshold ∀ t ]
Reserve evolves with fee income minus daily fraud and a 0.6%-per-day exploit probability. Result: 38% solvency, expected peak drawdown $12.4M.
07 · Contagion & Correlation

Nyxus does not fail alone

DeFi protocols are composable, which means they are entangled. A shock to one propagates. We model that spread across a synthetic protocol network.

Chart 7 from the Nyxus Protocol Risk Assessment: DeFi fraud and exploit risk analysis
Contagion spread from a Nyxus shock across linked protocols. Red is origin; warmer nodes absorb more of the shock through direct exposure.
Chart 8 from the Nyxus Protocol Risk Assessment: DeFi fraud and exploit risk analysis
Correlation of on-chain drivers. Fraud loss rises with volume and transaction count, confirming fraud scales with activity, not calendar time.
08 · Risk Register & Scoring

Ten dimensions, one weighted score

Each risk scored from likelihood, severity, velocity, and detectability, then weighted into 68.5 (High).

Likelihood →
SMA
FRA
LIQ
ORA
MAR
REG
CON
GOV
OPE
MOD
low↑ Severityhigh
Risk dimensionLikelihoodSeverityExp. lossScoreBand
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

Scoring methodology

scoreᵢ = 100·(0.40·Lᵢ + 0.40·Sᵢ + 0.12·Vᵢ + 0.08·(1−Dᵢ))^0.9
L likelihood, S severity, V velocity, D detectability. Smart-contract exploit and liquidity risk carry the highest weights, since they are the fastest and most severe.
09 · Stress Testing

Eight ways it gets attacked

Each scenario re-runs the full solvency model under a specific shock. Survival probability is a fresh 20,000-path simulation per scenario.

ScenarioReserve troughFinal reserveP(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
◆ Reading the table

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.

10 · Recommendations

What we would tell the team

01

Resize the reserve to the tail

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.

02

Insure the exploit, not the fraud

Everyday fraud is self-fundable. The tail exploit is exactly the low-frequency, high-severity shape insurance exists for.

03

Deploy the fraud score live

The wallet model (AUC 0.76) should gate high-risk wallets in real time, with the isolation forest catching novel patterns.

04

Monitor composability exposure

Contagion analysis shows Nyxus inherits peer risk. Track exposure to linked protocols as a live risk metric.

05

Add an oracle circuit breaker

Oracle manipulation scores high on velocity. A price-deviation halt caps the damage window before a drain completes.

06

Stage a bank-run buffer

The bank-run scenario is severe. A withdrawal-rate limit during confidence shocks buys time for the reserve to hold.

◆ Final classification

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.