Bot Detection Algorithms
Bot activity is one of the most influential and distortive forces in Solana token markets. Automated trading systems—MEV bots, snipers, latency arbitrage bots and volume inflators—create patterns that can mislead traders and distort perceptions of liquidity, momentum and demand.
Raphael Sentinel includes a dedicated Bot Detection subsystem designed to identify, classify and quantify bot-driven behavior across both SPL and bonding-curve tokens. This system operates alongside the Wallet Intelligence Model (WIM) and uses temporal, structural and behavioral signals to distinguish bots from organic market participants.
The result is a cleaner, more reliable intelligence layer that removes artificial noise from the analysis.
1. Purpose of Bot Detection
Bots introduce several risks:
artificial volume creation
fake liquidity signals
misleading momentum
inorganic pump behavior
precision sniping that disadvantages retail
accelerated exit cascades
Traditional scanners fail to detect bots because they rely on surface-level metrics (volume, price, holder count), not underlying behavioral patterns.
Sentinel’s detection algorithm monitors deeper signals.
2. Categories of Bots on Solana
Sentinel organizes bots into four operational categories:
1. Snipe Bots
Act within microseconds at launch. Patterns include:
extremely tight timing
predictable transaction sequence
routing through specialized MEV pathways
identical BUY instruction patterns
These bots create misleading early demand.
2. MEV Arbitrage Bots
Bots exploiting price discrepancies across DEXs or between bonding curves and DEX pricing. Signals include:
constant micro-trades
minimal slippage
rapid identical-frequency activity
consistent profit-taking intervals
3. Wash Trading Bots
Bots designed to inflate volume for marketing or manipulation. Detected by:
repeated back-and-forth trades
self-trading patterns
artificial liquidity creation
circular flows through the same wallets
4. Liquidity Attack Bots
Bots that:
snipe LP creation
sandbag liquidity transitions
exploit early volatility to trigger cascading exits
These bots are especially relevant for SPL tokens.
3. Detection Methodology
Sentinel uses a multi-factor detection system combining:
A. Timing Signatures
Bots operate with machine-level consistency.
Key timing indicators:
exact millisecond alignment
repetitive timing intervals
simultaneous entries across wallets
microsecond clustering after LP creation or bonding-curve fill
If 10 wallets buy within the same 50–150ms window → swarm bots are likely.
B. Routing Pattern Similarity
Bot-controlled wallets often use identical:
token program accounts
transaction structures
DEX routing paths
fee payer patterns
compute unit configurations
Structural similarity = high bot probability.
C. Transaction Shape Analysis
Sentinel analyzes each transaction's internal structure:
number of instructions
identical instruction sequences
repeating input-output mapping
signature entropy (bot signatures differ from humans)
Bots produce highly consistent transaction shapes.
D. Behavioral Fingerprinting
Long-term activity patterns per wallet:
extremely high-frequency micro-trades
exact buy → sell time ratios
predictable loop behavior
absence of human variance
This links unknown wallets to known bot clusters.
E. Cross-Wallet Correlation
If multiple wallets:
share funding
share timing
share routing
share profit-taking patterns
…they are flagged as a coordinated bot cluster.
This is especially important for Pump.fun analysis where bots often simulate community interest.
4. Bot Impact Scoring
Bots do not automatically mean high risk — some are harmless. Sentinel calculates the Bot Impact Score for each token:
Low impact
scatter bots
low trade volume
arbitrage bots smoothing price gaps
Medium impact
snipers causing early volatility
artificial volume inflation
High impact
wash trading to fake market activity
coordinated bot swarms around exit patterns
bots manipulating price to trap retail
The Bot Impact Score contributes directly to the SPL Risk Engine and indirectly to the Bonding Curve Risk Engine.
5. Integration With Other Systems
With WIM (Wallet Intelligence Model)
Bot detections update each wallet’s long-term behavior profile.
With Cluster Engine
Bots are treated as automated actors affecting momentum, not dev/human clusters.
With Risk Engine
Bots increase or decrease risk depending on:
density
intent
cluster overlap
effect on market stability
With Analysis Pipeline
Bots are removed from certain calculations (e.g., organic holder count, natural volume).
6. Continuous Learning
Sentinel continuously improves bot detection by:
observing new bot patterns
updating MEV routing signatures
adding new cluster fingerprints
tracking emerging sniper algorithms
comparing against historical patterns
Over time, the system becomes harder to fool and more accurate at distinguishing bot behavior from real traders.
Last updated