launchtape/solana-launch-tape

Solana Launch Tape — free 2-hour sample

Launch Tape records what actually happens on pump.fun, PumpSwap, two other Solana launchpads, and Meteora, 24/7, from public data only. This dataset is a free 2-hour sample of each of the seven files we sell by the day: 2026-10-10 14:00-16:00 UTC for the first four; curvetape and launchmeta 2026-10-13 14:00-16:00 UTC (their first day); gradholders 2026-10-14 04:00-06:00 UTC (on sale from that day). It is here so you can check the shape, the honesty of the recording, and whether the data answers your question before buying a full day.

• Website / full catalog / pricing: launchtape.io
• Contact: launchio48@gmail.com
• History: short and honest. The oldest full day for sale is 2026-10-03 (gradtrades, padfeed) / 2026-10-04 (pregrad) / 2026-10-06 (meteora-pools) / 2026-10-13 (curvetape, launchmeta) / 2026-10-14 (gradholders), and it grows by one UTC day, every day. Coverage gaps (a dropped connection, a coin skipped under our data budget) are written into the files as rows (down, up, resume, end with a why) — never filled in, never guessed.

Dataset structure

Seven splits, one per recorder, each covering 2 hours (windows above). MANIFEST.json and SHA256SUMS (in data/) carry exact row counts, byte sizes and checksums for every file.

Split	Rows	What it is
gradtrades	417,544	Every buy/sell on PumpSwap in the first 30 minutes after each pump.fun graduation in the window: wallet, amounts, price, pool reserves, fees
pregrad	14,417	Who is behind each pump.fun launch in the window: creator, funding, curve buyers, graduation-slot buyers, the cap path, plus a +10 min price outcome
padfeed	3,602	Graduations on Meteora DBC (Bags, Jupiter Studio, ...) and Raydium LaunchLab (bonk.fun) in the window: migration facts, pool balances, top holders, US$ quote prices
meteora_pools	23,794	Per-minute price, TVL, volume and fees for ~200 Meteora DLMM pools, for the window
curvetape	470,519	Every trade on every pump.fun bonding curve in the window, from creation to graduation: wallet, tokens, SOL, fees, cap and reserves after each trade; plus creations, completions and migrations
launchmeta	4,597	Every new pump.fun coin's metadata (name, ticker, description, website) and a SHA-256 of its picture, at creation and at graduation
gradholders	251	Graduates' 20 largest token accounts with owners and shares at +1, +10 and +60 minutes
graduates (a CSV, not a split: data/graduates-2026-10-10-1400-1600utc.csv)	105	The window on one sheet, one row per pump.fun graduate: what it was, who was behind it, and what its price did in its first 10 and 30 minutes (reference = the price 10 s after graduation). Opens on a phone or in the ChatGPT / Claude app; columns explained at https://launchtape.io/#graduates

Every file also comes as Parquet in data/, under the same name with .parquet (data/pregrad-2026-10-10-1400-1600utc.parquet and so on): the same rows in the same order, one typed column per field, checked against the .jsonl.gz before it ships; MANIFEST.json and SHA256SUMS cover both. One table per file: the columns are every field of every row type and a field a row does not have is null, so filter on type first. t stays integer milliseconds; nested objects are structs and arrays are lists; padfeed's usd is a map keyed by quote mint; pregrad's curve, src and cross (text in some row types, an object or a number in others) are text columns, an object or number as its JSON text. The paid day files come in both formats too, with these exact columns. (The splits above load the JSON lines: the datasets library has no type for padfeed's map column, so read the Parquet copies with pandas, Polars or DuckDB, below.)

How it's collected: only public data (the Solana chain through the public RPC, and a paid RPC for gap repair and holder snapshots; PumpPortal's free feed; public IPFS gateways for coin metadata; Meteora's public API; Jupiter's free price API). No private keys, no logins, no scraped social profiles. No wallet labels or opinions ("insider", "rugger") — we sell what happened on chain, not our guesses about who did it. Our own trading wallets are stripped from every file. Every line is parsed and checked before export; malformed rows are dropped, not patched.

Each row is a JSON object; every row carries type and t (ms since 1970 UTC, when our recorder saw it). Full field-by-field schemas (every row type, every field) are published at launchtape.io; condensed below.

gradtrades columns

type	key fields	meaning
grad	mint, pool, coinCreator, poolBase/poolQuote, price, capSol, curveSol, follow	the graduation: starting pool state, whether this coin is followed
trade	mint, side, user, baseAmount, quoteAmount, price, priceAfter, userTokensBefore, fees, creator	one PumpSwap buy or sell
end	mint, why, buys, sells	coin no longer followed, with totals
down/up/resume	t	feed gap marker — missing, not guessed

pregrad columns

type	key fields	meaning
create	mint, creator, name, sym, devTokens/devSol, capSol	the coin's creation and the creator's own buy
curvex/ctrade	mint, cap, sol, side, user, tokens, capAfter	curve cap crossings and trades, 200 SOL cap up to graduation
record	mint, ageS, instant, creatorPctAtGrad, creatorInfo, gradSlot, curve	full picture at graduation, incl. whether it filled the curve in <5s
outcome	mint, capAt, maxGain, minDrop, crashed, dump/firstBigSell	first 10 minutes after graduation: best/worst move vs the price 10 s after graduation (touched -80% = minDrop <= -80) and whether it fell 40%+ from its high (crashed; not an 80% drop)

padfeed columns

type	key fields	meaning
grad	pad (dbc/ray), mint, quote, b/q, px, supply, lagMs	migration to the destination pool, mint facts
px	pad, mint, b, q	pool vault balances whenever they change (~6s)
holders	pad, mint, top10, top20	20 largest token accounts at +60s and +5min
qpx	usd	US$ price of SOL and every quote mint, every 30s

meteora_pools columns

field	meaning
pool	the DLMM pool address
price	quote per base token
tvlUsd, vol24hUsd, fees24hUsd	TVL, rolling 24h volume and fees, US$
vol24hDeltaUsd, feesMinUsd	change since the pool's previous row

curvetape columns

type	key fields	meaning
create	mint, curve, creator, name, sym, uri, quote, vt, vs, rt, supply	a new coin and its starting curve reserves (raw units); quote null = priced in SOL
trade	slot, sig, i, bt, off, mint, side, wallet, tokens, sol, fee, cfee, capAfter, curveSol, vs, vt, ix	one curve trade: SOL without fees (a buyer paid sol + fee + cfee), cap in SOL and reserves after it; off = slots since creation. Curves priced in another token carry quote, qAmt, vq, rq, feeQ, cfeeQ instead
adopt	mint, why	a coin followed from a trade, not its creation (noCreate, revive)
complete / migrate	mint, pool, solAmount, mintAmount, poolMigrationFee	the curve completed; the coin moved to PumpSwap
end	mint, why, trades, buys, sells, vol, maxCap	coin no longer followed (graduated, idle, evicted) with its totals
tx	sig, role, txFee, prioFee, jitoTip, cuPrice, cuLimit, cuUsed	fee picture of a creation or early-buy transaction
gap	from, to, why, ms	the live feed was off: trades in the window are missing (from 2026-10-14 repaired from whole blocks: rows with src: "backfill" and a backfill row)

launchmeta columns

type	key fields	meaning
meta	phase (create / grad), src, mint, creator, name, sym, uri, uriKind	one row per coin at creation, another at graduation
meta	json (name, symbol, description, image, website, showName, createdOn, hasTwitter, hasTelegram), jsonKeys, metaSha, metaBytes	the metadata JSON's fields (social links reduced to true/false) and its sha256
meta	imgUrl, imgSha, imgBytes, imgType, cachedImg	sha256 of the picture's bytes: the same picture = the same imgSha (pictures never stored)
meta	err	why a fetch failed; never guessed
skip	mint, phase, why	not fetched (backlog on a very busy hour)

gradholders columns

type	key fields	meaning
holders	mint, mark (1 / 10 / 60), tGrad, lagS, slot, supply, pool	one snapshot per graduate and mark (minutes after graduation)
holders	top[] (acct, owner, amt, pct, role, pda)	the 20 largest token accounts, biggest first; role pool / curve, pda = program-owned
holders	top10Pct, top20Pct, top10PctAll, poolPct, curvePct	shares of supply, with and without the pool and curve
holders	err	the snapshot failed (429, late): no holder fields, never guessed

What you can do with it

From three days of the full feed (2,193 pump.fun graduates with prices in their 10-minute outcome, Oct 10-12 2026 UTC — bigger than this 2-hour sample, shown as a worked example of the kind of question pregrad answers). "Touched -80%" means minDrop <= -80 (vs the price 10 s after graduation). Of the 2,193, 38.1% touched -80%, 15.0% touched 2x and 5.3% touched 3x; 62.9% fell 40%+ from a high (crashed).

1. Fast fills were not the rug signal people think. Of 2,187 graduates that also have a record row, 1,384 filled their bonding curve within 5 seconds of creation ("instant"). They touched -80% about as often as the 803 slower fills (38.4% vs. 37.6%). The slower fills fell 40%+ from a high more often (85.6% vs. 50.0%), but also touched 2x three times as often (26.3% vs. 8.5%).
2. Creator exit timing matters. Of 1,913 graduates with creator-holding data, the 1,310 whose creator still held more than 5% of supply at graduation touched -80% 33.6% of the time; the 603 whose creator had already sold down to 5% or less, 47.4%.
3. Big sells come fast, rarely from the creator. Median time from graduation to the first big sell: 3.3 minutes (921 coins had one). The creator was that first big seller only 2.6% of the time.

(An earlier version of this card called crashed an 80%+ fall; it is a 40%+ fall from the high, corrected Oct 14.)

pregrad's record and outcome splits carry everything needed to reproduce all three (instant, creatorPctAtGrad, minDrop/maxGain, crashed, firstBigSell).

Load it

from datasets import load_dataset

data_files = {
    "gradtrades": "data/gradtrades-2026-10-10-1400-1600utc.jsonl.gz",
    "pregrad": "data/pregrad-2026-10-10-1400-1600utc.jsonl.gz",
    "padfeed": "data/padfeed-2026-10-10-1400-1600utc.jsonl.gz",
    "meteora_pools": "data/meteora-pools-2026-10-10-1400-1600utc.jsonl.gz",
    "curvetape": "data/curvetape-2026-10-13-1400-1600utc.jsonl.gz",
    "launchmeta": "data/launchmeta-2026-10-13-1400-1600utc.jsonl.gz",
    "gradholders": "data/gradholders-2026-10-14-0400-0600utc.jsonl.gz",
}
ds = load_dataset("launchtape/solana-launch-tape", data_files=data_files)
# or, from a local clone:
ds = load_dataset("json", data_files=data_files)

print(ds["pregrad"][0])

The Parquet copies, with pandas or DuckDB:

import pandas as pd
from huggingface_hub import hf_hub_download

path = hf_hub_download("launchtape/solana-launch-tape", "data/pregrad-2026-10-10-1400-1600utc.parquet", repo_type="dataset")
df = pd.read_parquet(path)
records = df[df["type"] == "record"]          # nested fields are dicts: records["creatorInfo"].str["funder"]

SELECT type, count(*) AS n
FROM read_parquet('hf://datasets/launchtape/solana-launch-tape/data/gradtrades-2026-10-10-1400-1600utc.parquet')
GROUP BY type ORDER BY n DESC;

Each split loads as a flat table of JSON objects; filter by type for the row kind you want (see the column tables above). Rows are plain JSON, so pandas.read_json(path, lines=True) and DuckDB's read_json_auto(path) work directly on the .gz files too.

Getting full days

This sample is one 2-hour window. Full UTC-day files (JSON lines and Parquet) are on launchtape.io, paid in USDC or SOL on Solana (other chains on request).

On sale now: every UTC day from 2026-10-03 to 2026-10-13 (each dataset from its first full day, listed above), and one more day is added each day as it closes.

Three ways to buy:

• One day, $10 once: any UTC day you pick, with every graduate's first 30 minutes (gradtrades) and who was behind each launch (pregrad), plus the test prompt with that day's files filled in. No subscription.
• Starter, $10 a month: all 7 datasets, each new day's files by email as the UTC day closes, plus the last 7 days on file
• Full history, $49 a month: everything in Starter, plus every day on file back to October 3 2026

Prices are in USDC; paying in SOL is the same amount at the day's SOL price.

Questions: launchio48@gmail.com.

License

Personal and research use only, inside your own team. No resale, no republishing, no redistribution of the files in whole or in part. Derived results (charts, papers, models, strategies) are yours to keep and use freely. Provided as is, no warranty — this is market data, not financial advice.

Solana Launch Tape - data licence

Copyright (c) 2026 Launch Tape.

You may use the data files (including this free sample) for personal and research use, inside your own team. You may not resell, republish or redistribute the files, in whole or in part, or share them outside your own team. Derived results (charts, papers, models, strategies) are yours.

The data is provided as is, with no warranty of any kind. It is market data, not financial advice.