Early tokens

Early token features

The eight feature groups the quant engine computes for a freshly launched token.

For every early token the quant engine computes eight feature groups. They are measurable, registered and versioned — the raw material for the early-token scores.

The eight groups

Group 1

Launch

Contract / launch / tradable / pool age, platform, type, migration status, bonding-curve progress, time to first liquidity & trade, discovery latency.

Group 2

Liquidity

Initial & current liquidity, growth/reduction, concentration, ownership, locked/burned, removal permissions, reserve imbalance, buy/sell slippage, price impact.

Group 3

Transaction

Tx per second/minute, buy/sell count & volume, net buy pressure, unique/repeat buyers & sellers, avg/median size, size distribution, acceleration.

Group 4

Wallet

Creator balance/sales/transfers, deployer-linked wallets, top-10/50 concentration, whale accumulation/distribution, sniper & bot wallets, coordination, new-wallet ratio.

Group 5

Price

Velocity, acceleration, realized volatility, drawdown, distance from launch & peak, recovery speed, microstructure imbalance, spread, cross-market divergence.

Group 6

Security

Honeypot probability, mint/freeze authority, blacklist, transfer restrictions, buy/sell tax, hidden privileges, upgrade authority, proxy, pausable, max-wallet/tx, malicious-code similarity.

Group 7

Social

Mention velocity, unique-author growth, engagement acceleration, bot probability, influencer involvement, creator-channel activity, narrative, sentiment, organic-vs-coordinated.

Group 8

Market context

BTC/ETH direction, chain activity & congestion, gas/priority fees, market volatility, meme regime, sector momentum, exchange conditions, time of day, day of week.

Features are evidence, not verdicts

Indicators are treated as features, never as standalone proof. The agents interpret them and the deterministic scoring engine combines them — see early-token scoring.

The earliest snapshots of these features are archived in the early_token_snapshots Parquet dataset, so the first seconds of a launch stay replayable for backtesting and memory.