RED-PUMP-2026: pump.fun token launches with collector-recorded labels (May to June 2026)

Canonical source and DOI: this is a mirror of the data files of RED-PUMP-2026-v1 on Zenodo. Please cite the concept DOI 10.5281/zenodo.20633486, which always resolves to the latest version. The full deposit (code, pre-registration, manuscript and documentation) is on Zenodo; the data files here are byte-identical to the published version.

Author: Arati Uday Kamat (ORCID 0009-0000-4781-312X) · License: CC BY 4.0

What it is

A public record of 860,213 token-launch records (860,194 unique mints) on the Solana launch platform pump.fun, created between 2026-05-08 17:14 UTC and 2026-06-10 18:11 UTC, together with a per-mint terminal label recorded by the author's data collector within 24 hours of launch.

It is useful for studying launch activity, the metadata of new tokens (social links, description length, initial market capitalisation), and, above all, how outcome labels produced by a data collector differ from platform-side events.

Files

File	Rows	Contents	SHA-256
red_pump_2026_v1_launches.jsonl.gz	860,213	One JSON object per launch record: seenAt, t, mint, symbol, name, created_timestamp (ms), initial_market_cap_sol, has_twitter, has_website, has_telegram, description_length	042940379e8c897ac97403e6b25a5b302fb32b6902a8fc0cef4ab70ac11e8f84
red_pump_2026_v1_outcomes.csv.gz	833,192 data lines	Collector-recorded terminal label per mint: outcome (GRADUATED or TIMEOUT), note (bonding_curve_filled or no_graduation_within_24h), timing and market-cap fields	c0a327ea442d91c6f970b2bad9a2a9b778e163d8c3eb38f71eccd3e92209a974
reconstructed_mint_observation_windows_v1.parquet	860,194	Per-mint reconstructed observation windows and analysis flags used in the accompanying paper	f785a812a38445fedd222e357eede246132ae69dffc4ed62131246614d39bfd5

Dataset viewer and loading

The dataset viewer shows the launches table. The Hub viewer reads one file format per repository, so the outcomes file (.csv.gz) and the observation-window file (.parquet) are not in the viewer. They are kept here unchanged, byte-identical to Zenodo, and load directly:

from datasets import load_dataset
import pandas as pd
from huggingface_hub import hf_hub_download

repo = "AratiKamat/red-pump-2026"
launches = load_dataset(repo, "launches", split="train")

windows = pd.read_parquet(
    hf_hub_download(repo, "reconstructed_mint_observation_windows_v1.parquet", repo_type="dataset"))

# Tolerant read: the outcomes file mixes two schemas and has a few malformed lines (see below).
outcomes = pd.read_csv(
    hf_hub_download(repo, "red_pump_2026_v1_outcomes.csv.gz", repo_type="dataset"),
    on_bad_lines="skip", low_memory=False)

Important: what the label means

• The GRADUATED / TIMEOUT label is the collector's own classification, made by polling pump.fun's newest-coins feed (top 50), not the platform-side bonding-curve completion event.
• In the paper's primary window, GRADUATED labels were written about 1 to 6 minutes after on-chain creation. In practice, the label means "completed while still on the newest-50 feed"; it is not a record of every platform-side completion.
• Earlier survival-analysis claims made with this dataset (graduation-rate and hazard-ratio estimates) have been withdrawn. The current accompanying paper is a measurement audit: Auditing Collector-Generated Graduation Labels on Pump.fun: Measurement Error and Temporal Non-Generalization (arXiv:2607.02823).

Known data-quality notes

• The outcomes file mixes an early 8-field schema and a later 11-field schema. A small number of lines are malformed (for example, commas inside token symbols, and truncated or blank lines). Read it with a tolerant CSV parser. The full reconciliation (category counts) is documented in the Zenodo deposit.
• Token names and symbols are public on-chain metadata created by third parties, and are reproduced as recorded. Some contain offensive text.

How to cite

Kamat, A. U. (2026). RED-PUMP-2026-v1: A public benchmark of 860,213 pump.fun token launches with collector-generated terminal labels (May–June 2026 regime) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.20633486

@dataset{kamat_redpump_2026,
  author    = {Kamat, Arati Uday},
  title     = {{RED-PUMP-2026-v1}: A public benchmark of 860,213 pump.fun
               token launches with collector-generated terminal labels
               (May--June 2026 regime)},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.20633486},
  url       = {https://doi.org/10.5281/zenodo.20633486}
}