Launch TapeRecorded since Oct 3 2026 JSONL · Parquet · CSV
Solana memecoin backtesting data, recorded trade by trade.
Test a pump fun bot on what actually happened: the recorded buys and sells after each followed graduation, each with its slot, its fees and the pool’s reserves after it. Fill a block late, charge fees both ways, size against the real pool, and keep the coins that crashed. Gaps are marked, never filled in.
Get every recorded day Try a rule free
The in-browser backtest needs no wallet, signup or API key.
- Recorded sinceOct 3 2026a new UTC day after each one closes
- Trade window30 minafter graduation, followed coins only
- Full 30-minute tapes~4 in 10graduates; the rest are partial or not followed
- Same rule, two methods+2.3% vs −0.9%a trade: naive vs honest replay, 67 coins
What a backtest needs, and where it is
The dataset name on the rightRight after graduation things move fast: when a big holder dumped in the first 10 minutes, the median time was 2.2 minutes after graduation (894 coins, Oct 10–12 2026). A test that checks prices every 30 seconds sees little of that window. These are the pieces an honest replay needs.
A fill after your signal, never at it. Every recorded trade carries its slot and block time, so you can fill each signal at the end of the first slot after it.Sort by slot: trades read back from the chain after a gap are marked
gradtradessrc: "backfill"and were received later than they happened.Your size against the real pool. The pool’s tokens and SOL after every trade, so you can price your own buy and sell with the pool’s math instead of the quoted price.A rule that works at a tiny size can fail at a bigger one on the same coins.
gradtradesFees on both sides. Each trade’s LP, protocol and creator fees in SOL.The trade rows carry no priority fees and failed transactions are not included: add your own priority fee and delay as assumptions.
gradtradesThe coins that went badly. One row per graduation the recorder saw that UTC day, followed or not, so the coin list is fixed before you look at outcomes. Followed coins keep their trades through a crash.The
graduatesfollowed,tape_secanddata_gapcolumns say how much of each coin’s window is on file.Who was behind each launch. The creator’s share at graduation, how fast the curve filled and who bought in the graduation slot, so a filter can be tested instead of trusted.Raw facts only: addresses, amounts, times. No wallet labels.
pregradBefore graduation, too. Recorded buys and sells on the launch curve for followed coins, from creation to graduation, for rules that enter early.Since Oct 13 2026.
curvetape
Field-by-field schemas: Datasets on the home page and the data documentation. How to check a popular filter: Filters people swear by.
How a backtest fills its trades can decide the answer on its own. We ran one simple rule on two hours of recorded trades (October 10 2026, 14:00 to 16:00 UTC): buy 15 seconds after a pump fun graduation, sell at 3 minutes or at −20%, 0.1 SOL a trade.
- The usual backtest+2.3%a trade, filled at the price that triggered each signal, no fees
- The honest replay−0.9%a trade, filled a block late, fees and slippage both ways, size moving the pool
Same 67 coins, same rule. The window had 105 graduations: 36 had no recorded trades (not followed) and 2 were left out because the file ended mid-trade. Stops set at −20% filled at −23.9% (median): a stop is an exit attempt, not a guaranteed price.
On the full day of October 12 2026 (775 coins traded) the same rule averaged −3.9% a trade. Two hours and one day are small samples: they show how much the method moves the result, not whether a rule works. Better rules may well exist; an honest replay is how to find them. The method is on Test your bot, and the five fixes are in Your backtest is lying.
What you can and can’t replay
Read this before you trust a resultFollowed coins only. The recorder follows coins whose cap at graduation is 300–600 SOL, at most 30 at a time, for 30 minutes each. Coins outside that rule get a graduation row and no trades, so they can’t be replayed. Full 30-minute tapes cover roughly 4 in 10 graduates.
Gaps are flagged coin by coin. In the graduates CSV, data_gap is 1 when a recorder gap touched that coin’s window (picked up late, a restart, a dropped feed): its trades may be incomplete, and nothing is filled in. A 0 means no gap was flagged for that coin, not proof that every transaction was captured. tape_sec says how many seconds of trades the day’s file holds after graduation, up to 1800; a coin that graduated in the day’s last 30 minutes continues in the next day’s file.
Filter, then count. For full-window tests keep followed = 1, data_gap = 0 and a tape_sec of 1800, and say how many coins each filter removed. A coverage filter shrinks the sample, so report it next to the result.
Known gaps. October 18 2026 has no graduates or post-graduation trades (a recorder outage), October 17, 19 and 22 are partial days, and November 1 was interrupted for part of the evening UTC. Failed transactions are not included. Details in the coverage notes and the data documentation.
Try it before you buy
Free · no signup- In-browser backtestTry a rule on five recorded graduatesChange the take-profit, stop, exit time and size; no wallet needed
- File previewReal rows from a recorded day
- Free coin lookupWhat happened to one coin
- Open source · GitHubThe backtest, with a public preview file
- Python · PyPIpip install launchtapeLoad the files and run the same replay in a terminal
- AI connectorAsk Claude, ChatGPT or Cursor about recorded days
Get the files
Every recorded day, on one link
Month · US$20 once: every day recorded since Oct 3 2026, plus every new day for 30 days, no automatic renewal. Or just the days you pick: One day · US$5 · 5 days · US$10, counting toward Month within 7 days when paid in SOL or USDC on Solana.
Daily spreadsheets and all the detailed file types, as JSON lines and, for almost every day, Parquet, on one download link that grows as each UTC day closes. Every current plan is in the buy section on the home page. Pay with SOL or USDC on Solana, other crypto through NOWPayments, or Interac e-Transfer in Canada.
If a file doesn’t open or isn’t as described, you get your money back. Historical observations for research, not signals; no profit is guaranteed.
Questions
Can I backtest a bot that buys right after graduation?
Yes, on followed coins. Each one’s recorded trades run for up to 30 minutes after graduation, with slot and block time, so an entry can fill at the first slot after its signal. Coins outside the follow rule have a graduation row and no trades, so they can’t be replayed.
Are coins that crashed included?
Yes. The graduates CSV has a row for every graduation the recorder saw that day, whatever happened next, and followed coins keep their trades through the drop. Fix your coin list before you look at outcomes, and count the coins that never reached your entry.
Are fees and price impact in the data?
Each recorded trade has its LP, protocol and creator fees in SOL and the pool’s reserves after it, so you can charge fees and price your own size against the pool. The trade rows carry no priority fees and failed transactions are not included: add your own priority fee and delay as assumptions.
How do I leave out coins with gaps?
Use the graduates CSV: for full-window tests keep followed = 1, data_gap = 0 and a tape_sec of 1800, and report how many coins each filter removed. A data_gap of 0 means no gap was flagged for that coin, not proof that every transaction was captured.
Can I try it before buying?
Yes. The in-browser backtest on five recorded graduates, the file preview, the coin lookup, the open-source backtest on GitHub and the AI connector are all free.