Early token scoring
The dedicated early-token score components, the deterministic final formula, and the blocking rules that override it.
Early tokens need a dedicated scoring model. The components reward launch quality and momentum, and penalize security, sellability and data-uncertainty risk.
Score components
| Score | Measures |
|---|---|
| Launch viability | Is the launch becoming a real market? (liquidity, buyer/tx growth, migration, unique wallets) |
| Early momentum | Short-term acceleration (price/volume velocity & acceleration, buy pressure, holder & social velocity) |
| Security | Technical/behavioral risk (permissions, creator behavior, removal risk, concentration, malicious patterns) |
| Sellability | Realistic exit feasibility (simulated sell, exit slippage, liquidity, restrictions) |
| Memory | Similarity to historical launches (winners, rugs, false pumps, sustainable, bot-dominated, migrations) |
| AI interpretation | Qualitative context (narrative, identity, social authenticity, event significance, contrary evidence) |
| Execution | Can an order be submitted and filled under acceptable conditions? |
The deterministic final score
final early opportunity score
Final Early Opportunity Score =
Launch Viability + Early Momentum + Memory Similarity
+ AI Interpretation + Execution Feasibility
- Security Risk - Sellability Risk - Data UncertaintyThe formula is versioned and tested. The LLM supplies structured evidence and classifications; it never writes the number — that is deterministic code.
Blocking rules override the score
An excellent momentum score must never override a hard safety block:
- confirmed honeypot
- failed sell simulation
- prohibited contract function
- creator removing liquidity
- liquidity below minimum
- sell slippage above maximum
- unsupported chain state
- stale market data
- execution engine unavailable
- duplicate order
- kill switch active
- risk state halted / emergency
Safety beats score
Blocking rules are evaluated independently of the score and recorded as blocking_reasons on the score snapshot. No quantitative signal can buy past a confirmed safety failure.