How ARC AI Calculates Scam Probability: The Technical Details

The scam probability score in every ARC AI report is not a single metric — it is the output of a multi-factor analysis that examines the token contract, liquidity setup, holder distribution, social signals, and deployer history simultaneously. Here is how it works.

Contract analysis

Sentinel decompiles and analyses the token's smart contract bytecode looking for patterns associated with known scam mechanisms:

• Hidden mint functions: Can the deployer create new tokens at will, diluting your holding to near zero?
• Ownership not renounced: Does the deployer retain the ability to modify the contract after deployment?
• Blacklist functions: Can specific wallets be blocked from selling — the classic honeypot setup?
• Tax manipulation: Can buy/sell taxes be modified post-deployment to trap traders with 99% sell taxes?
• Proxy patterns: Is the contract upgradeable in ways that could change its behaviour after audit?

Each of these patterns contributes to the scam probability score. A contract with a hidden mint function and owner-controlled blacklist scores very high regardless of other signals.

Liquidity analysis

Legitimate projects lock liquidity. Scam projects keep it accessible so they can drain the pool and run. Sentinel checks:

• Lock duration: Is liquidity locked? For how long? Under 6 months on a new launch is a red flag. No lock at all is nearly always a scam indicator.
• Lock provider: Which platform is the lock held on? Known, audited lock providers score better than obscure or homemade locking contracts.
• Liquidity ratio: Is the liquidity pool large enough relative to market cap to support real trading? A $10M market cap token with $5K liquidity has serious problems.

Holder distribution

Concentrated token supply is a structural risk factor. If 3 wallets control 80% of the supply, a coordinated dump can destroy the price in seconds. Sentinel checks:

• Top 10 wallet concentration: What percentage of supply do the 10 largest non-DEX wallets hold?
• Deployer wallet balance: Does the token creator still hold a large position? When did they last transact?
• Distribution over time: Is the holder count growing (healthy) or shrinking (suspicious)?

Deployer history

ARC AI maintains a database of known scam deployer addresses. When a new token is deployed from an address that previously ran a rugpull, the scam probability score gets a significant automatic penalty regardless of the new token's other characteristics.

This database is updated continuously as Sentinel processes scam alerts and post-mortem analysis of confirmed rugpulls.

Social signal analysis

Coordinated social activity is a hallmark of pump-and-dump schemes. Sentinel analyses the pattern — not just the volume — of social mentions:

• Account age distribution: Are a large fraction of accounts mentioning the token recently created? Bot farms typically use fresh accounts.
• Mention velocity vs. organic growth: Did the project go from zero to viral in 24 hours with no prior community history?
• Identical or near-identical posts: Coordinated shill campaigns leave statistical fingerprints.

How the final score is calculated

Each factor above contributes a weighted component to the final scam probability score. Contract red flags carry the highest weight, followed by liquidity setup, deployer history, holder concentration, and social signals.

Critically, the weighting is non-linear: a single hard red flag (like no liquidity lock on a brand new token) can push the score above 60% on its own, even if all other factors are clean. This reflects the real-world reality that a single critical vulnerability is enough to get rugged.

Edge cases and false positives

No scoring system is perfect. The most common sources of false positives:

• Very new tokens: Insufficient data can cause Sentinel to assign higher scam probability as a precautionary measure. Scores typically stabilise after 24–48 hours of on-chain activity.
• Unusual but legitimate contract patterns: Some DeFi protocols use non-standard contract patterns that can superficially resemble scam mechanisms. The AI report narrative section will usually flag this explicitly.
• Low liquidity on small chains: The liquidity ratio check can flag legitimate small projects as risky purely due to size, not intent.

When in doubt, read the full narrative section of the AI report — Sentinel provides a plain-English explanation of every significant risk factor it detected.

Run a scam probability check on any token at arcai.io/analyze.