Skip to main content

Scoring Model

This page describes how Rug Radar calculates token risk today.

Core Principle​

Higher score means more visible risk.

The score is deterministic. Given the same upstream inputs, Rug Radar will produce the same risk score and label.

Major Inputs​

The risk model in src/utils/riskScore.ts weighs:

  • holder concentration
  • liquidity depth
  • honeypot status
  • mint authority
  • freeze authority
  • verification status
  • token age
  • volume-consistency risk
  • heuristic dev-wallet concentration

Score Bands​

  • 0-25: Low Risk
  • 26-50: Medium Risk
  • 51-75: High Risk
  • 76-100: Extreme Risk

Scoring Logic In Practice​

Holder concentration​

Large top-10 concentration is one of the strongest risk drivers.

  • above 80% adds the largest penalty
  • above 60% is still a major problem
  • healthier distribution produces a bullish signal instead

Liquidity​

Thin liquidity is heavily penalized.

  • under $10k is treated as very thin
  • under $50k is still risky
  • deeper liquidity produces a bullish signal

Contract controls​

The following can materially increase risk:

  • honeypot detected
  • active mint authority
  • active freeze authority
  • contract not verified

Token age​

Very new tokens receive extra risk points because many early-stage failures happen before the token has survived any real time.

Volume consistency​

If 24 hour volume is very high relative to liquidity and price action is extreme, the model can add a smaller penalty for suspicious short-lived spikes.

Dev wallet concentration​

If the largest visible holder looks outsized, the model applies an additional bump to reflect concentration risk.

Confidence​

Confidence is not part of the raw score. It is reported alongside the score and reflects upstream source availability.

The implementation uses:

available_sources / total_sources

So a run with only half the expected data sources available will naturally have lower confidence even if the score itself appears moderate.

Bags Adjustments​

After the base score is computed, analyze_token can apply small Bags-driven adjustments:

  • strong community score can reduce risk slightly
  • no Bags pool can increase risk slightly
  • high creator claim share can increase risk slightly

These adjustments are deliberately small compared with structural signals like concentration or honeypot status.

What The Model Is Not​

  • It is not a machine-learned probability of a rug
  • it is not a price forecast
  • it is not a substitute for manual inspection

It is a compact ranking and triage mechanism.