MELT: A Behavioral Trace Dataset for High-Risk Memecoin Launch Detection

MELT (MEmecoin Launch Trace) is the first behavioral-trace dataset for analyzing and detecting high-risk memecoin launches on Solana. It covers 41,470 memecoin launches issued through Pump.fun, with 200M+ on-chain transactions parsed into typed behavioral records, coordinated-account (bundle) traces, 122 engineered features, and configurable risk-level annotations.

Dataset structure

The dataset is organized into six loadable configs:

Config	Rows	Description
memecoin	41,470	One row per memecoin launch (metadata + timing).
pre_migration	~30.8M	Pre-migration (bonding-curve) transactions, typed.
post_migration	~187.7M	Post-migration (DEX) transactions, first hour after migration.
bundle	~3.33M	Coordinated-behavior traces (Co-purchase / Fund-flow / Jito Bundle) linking an entity to a shared identifier.
label	41,470	Risk-level annotation per memecoin.
feat	41,470	122 behavioral features per memecoin (five groups).

memecoin, label, and feat share the same 41,470-memecoin population and can be joined on the mint address.

memecoin

mint_addr, creation_time, migrate_time, name, symbol, description, image_url, twitter, website, telegram, signature, creator. Times are UNIX seconds.

pre_migration / post_migration (transactions)

mint, type (mint&swap / swap / transfer / zero), timestamp, signature, trader_map (JSON string: {account: token_change}), token_amount, sol_amount, fee_amount, price, block_slot, block_index.

bundle

entity, entity_type, identifier, source. A single table of coordinated-behavior traces across three sources; entities sharing an identifier within the same source are treated as one coordinated group. The entity_type field distinguishes what entity holds per source:

source	entity_type	entity	identifier
Co-purchase	account	trader account	co-purchase transaction signature
Fund-flow	account	trader account	common funder account
Jito Bundle	signature	transaction signature	Jito bundle id

For the Jito Bundle source, transactions sharing a bundle_id (identifier) were submitted in the same Jito bundle.

label

mint_address, min_ratio, manipulated (yes / no / empty), label (high / medium / low).

feat

mint_address, mint_ts, and 122 features across five groups: group1_* (contextual), group2_* (holding concentration), group3_* (market activity), group4_* (bundle statistics), plus time-series fields ts / ts_len. All features are computed strictly from pre-migration data to prevent label leakage (no label or return fields are included).

Usage

from datasets import load_dataset

memecoin = load_dataset("Zinteck/MELT", "memecoin")
pre      = load_dataset("Zinteck/MELT", "pre_migration")
post     = load_dataset("Zinteck/MELT", "post_migration")
bundle   = load_dataset("Zinteck/MELT", "bundle")
label    = load_dataset("Zinteck/MELT", "label")
feat     = load_dataset("Zinteck/MELT", "feat")

# trader_map is a JSON string
import json
row = pre["train"][0]
traders = json.loads(row["trader_map"])

License

This dataset is released under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license. It is derived from public Solana on-chain data and is intended for research and non-commercial use.

Citation

This dataset accompanies a paper under review. Citation details will be added upon publication; please refer to the anonymized submission for now.