Sybil Detection

Sybil Detection — Coordinated Wallet & Fake Activity Analysis

Sybil attacks in Solana token markets involve a single entity operating many wallets to create the illusion of organic activity. This skill covers detecting coordinated wallet clusters, wash trading, bundled transactions, and fake holder inflation — critical for evaluating whether a token's metrics reflect real demand or manufactured signals.

Why Sybil Detection Matters

Token markets on Solana are rife with manufactured signals:

• Inflated holder counts: 500 "holders" that are really 10 entities with 50 wallets each
• Fake volume: Wash trading between self-controlled wallets to simulate demand
• Artificial social proof: Many wallets holding small amounts to appear broadly distributed
• Rug preparation: Creator distributes supply across many wallets, then sells coordinated
• Bundled launches: PumpFun tokens where creator buys via Jito bundle in first slot

A token showing 1,000 holders with 80% funded from 3 wallets is fundamentally different from one with 1,000 independently-funded holders. Sybil detection separates real demand from theater.

Detection Categories

1. Funding Source Analysis

Trace each holder wallet back 1-2 hops to find who sent them SOL:

import httpx

def trace_funding_source(wallet: str, api_key: str, max_hops: int = 2) -> list[str]:
    """Trace SOL funding sources for a wallet via Helius parsed transactions."""
    url = f"https://api.helius.xyz/v0/addresses/{wallet}/transactions"
    resp = httpx.get(url, params={"api-key": api_key, "type": "TRANSFER", "limit": 50})
    transfers = resp.json()

    funders = []
    for tx in transfers:
        for transfer in tx.get("nativeTransfers", []):
            if transfer["toUserAccount"] == wallet and transfer["amount"] > 0.001 * 1e9:
                funders.append(transfer["fromUserAccount"])
    return funders

Key signals:

• 3+ holder wallets funded from the same source = cluster
• Funding within 24h of token creation = high suspicion
• Funding amounts are identical (e.g., 0.05 SOL to each) = automated distribution

2. Co-Trading Patterns

Wallets that buy the same token at nearly the same time are likely coordinated:

def detect_co_trades(buy_events: list[dict], slot_window: int = 3) -> list[list[str]]:
    """Group wallets that bought within the same slot window."""
    buy_events.sort(key=lambda x: x["slot"])
    clusters = []
    current_cluster = [buy_events[0]]

    for i in range(1, len(buy_events)):
        if buy_events[i]["slot"] - current_cluster[0]["slot"] <= slot_window:
            current_cluster.append(buy_events[i])
        else:
            if len(current_cluster) >= 3:
                clusters.append([b["wallet"] for b in current_cluster])
            current_cluster = [buy_events[i]]

    if len(current_cluster) >= 3:
        clusters.append([b["wallet"] for b in current_cluster])
    return clusters

Interpretation:

• Same slot, different transactions = coordinated (bot-driven)
• Same transaction = bundled (definite sybil)
• First 3 slots after token creation = launch sniping cluster

3. Bundled Transactions

Multiple buys packed into a single Solana transaction or Jito bundle:

def check_bundle_ratio(early_buys: list[dict], bundle_window_slots: int = 5) -> dict:
    """Calculate the ratio of bundled vs independent early buys."""
    bundled = [b for b in early_buys if b.get("is_bundled", False)]
    first_slot = min(b["slot"] for b in early_buys) if early_buys else 0
    early = [b for b in early_buys if b["slot"] - first_slot <= bundle_window_slots]

    return {
        "total_early_buys": len(early),
        "bundled_buys": len(bundled),
        "bundle_ratio": len(bundled) / max(len(early), 1),
        "bundled_supply_pct": sum(b["amount"] for b in bundled) / max(sum(b["amount"] for b in early), 1),
    }

See references/bundler_detection.md for PumpFun-specific patterns and Jito bundle mechanics.

4. Wash Trading Detection

Same entity buying and selling through multiple wallets to inflate volume:

Signals:

• Wallet A buys token, transfers to Wallet B, Wallet B sells — circular flow
• Multiple wallets trading back and forth with no net position change
• Volume concentrated in wallet pairs with funding links

def detect_wash_cycles(transfers: list[dict], holder_set: set[str]) -> list[tuple]:
    """Find circular transfer patterns among known holders."""
    # Build directed graph of transfers between holders
    edges: dict[tuple, float] = {}
    for t in transfers:
        if t["from"] in holder_set and t["to"] in holder_set:
            key = (t["from"], t["to"])
            edges[key] = edges.get(key, 0) + t["amount"]

    # Find reciprocal pairs (A->B and B->A both exist)
    wash_pairs = []
    for (a, b), vol_ab in edges.items():
        vol_ba = edges.get((b, a), 0)
        if vol_ba > 0:
            wash_pairs.append((a, b, vol_ab, vol_ba))
    return wash_pairs

5. Creator Network Analysis

Identify wallets controlled by the token creator:

• Creator wallet's funding history reveals other wallets it funded
• Those wallets holding token supply = insider distribution
• Creator selling from "different" wallets = disguised dump

Key Metrics

Composite Risk Score

Combine individual signals into a single sybil risk score (0-100):

def compute_sybil_score(metrics: dict) -> dict:
    """Compute composite sybil risk score from individual metrics."""
    weights = {
        "funding_cluster": 25,    # Wallets from same funder
        "co_trade": 20,           # Coordinated buy timing
        "bundle_ratio": 20,       # Bundled early transactions
        "unique_funder": 15,      # Diversity of funding sources
        "transfer_density": 10,   # Internal transfers between holders
        "wash_trade": 10,         # Circular trading patterns
    }

    scores = {}
    # Each sub-score normalized to 0-1, then weighted
    scores["funding_cluster"] = min(metrics.get("max_cluster_size", 0) / 20, 1.0)
    scores["co_trade"] = min(metrics.get("co_trade_pct", 0) / 0.3, 1.0)
    scores["bundle_ratio"] = min(metrics.get("bundle_ratio", 0) / 0.5, 1.0)
    scores["unique_funder"] = 1.0 - min(metrics.get("unique_funder_ratio", 1.0), 1.0)
    scores["transfer_density"] = min(metrics.get("transfer_density", 0) / 0.3, 1.0)
    scores["wash_trade"] = min(metrics.get("wash_pairs", 0) / 5, 1.0)

    composite = sum(scores[k] * weights[k] for k in weights)
    risk_level = "LOW" if composite < 30 else "MEDIUM" if composite < 60 else "HIGH"

    return {"score": round(composite, 1), "risk_level": risk_level, "components": scores}

Data Sources

Workflow: Evaluate a Token

# 1. Get top holders
holders = get_top_holders(token_mint, rpc_url)

# 2. Trace funding sources for each holder
funding_map = {}
for wallet in holders[:30]:  # Top 30 is usually sufficient
    funding_map[wallet] = trace_funding_source(wallet, helius_key)

# 3. Cluster by common funder
clusters = cluster_by_funder(funding_map)

# 4. Check co-trade timing
early_buys = get_early_buy_events(token_mint, helius_key)
co_trade_groups = detect_co_trades(early_buys)

# 5. Check for bundles
bundle_stats = check_bundle_ratio(early_buys)

# 6. Check wash trading
transfers = get_token_transfers(token_mint, helius_key)
wash_pairs = detect_wash_cycles(transfers, set(holders))

# 7. Compute composite score
metrics = {
    "max_cluster_size": max(len(c) for c in clusters) if clusters else 0,
    "co_trade_pct": sum(len(g) for g in co_trade_groups) / len(holders),
    "bundle_ratio": bundle_stats["bundle_ratio"],
    "unique_funder_ratio": len(set(f for fs in funding_map.values() for f in fs)) / len(holders),
    "transfer_density": len(wash_pairs) / max(len(holders), 1),
    "wash_pairs": len(wash_pairs),
}
result = compute_sybil_score(metrics)
print(f"Sybil Risk: {result['risk_level']} ({result['score']}/100)")

Integration with Other Skills

• token-holder-analysis: Use holder list as input; sybil detection adds cluster context
• helius-api: Primary data source for parsed transaction history
• jito-bundles: Detailed bundle detection and MEV context
• liquidity-analysis: Combine with sybil score — low liquidity + high sybil = extreme risk
• whale-tracking: Distinguish real whales from sybil cluster aggregates

Files