Sample data: one day of one-minute recordings from five venues

The most common thing that stops a new install is having nothing to backtest against. Recording your own data takes days before it becomes useful, which is a poor first hour with a piece of software. So we have published one trading day of our own recordings, one post per venue, in the Downloads board.
VenueMarketSymbolsSize
Coinbasespot5132.8 MB
Krakenspot2920.9 MB
Binance Futuresperpetual6441.7 MB
Coinbase Internationalperpetual3018.3 MB
Kraken Futuresperpetual3218.3 MB
All five cover the same window — 2026-08-11 20:00 to 2026-08-12 20:00 ET — at one-minute resolution, with 68 columns per row: price, per-minute buy and sell volume, turnover, trade strength, and ten levels of the order book on both sides. No sign-in needed. Each post explains where to put the file, what to rename it to, and shows a SHA-256 so you can tell a truncated download from a corrupt database. One limitation is worth stating plainly: recordings do not carry the venue in the filename yet, so only one venue can occupy that folder at a time. Swap the file when you want to test another. It is on our list. These are our own recordings from each venue's public market data feed, not a paid vendor's. One day at one-minute resolution is a sample for evaluating the software, it is not a data product and it is not a substitute for recording your own. Research wants more days than this, and the one-second frame.
TraderWe
169 posts · 2 followers
+ Follow

Related reading

4 replies

BacktestBetty· Aug 2026 ago
Three things: 1) glad the window is identical across all five, that's the only way a cross-venue sanity check means anything; 2) the one-venue-per-folder thing is going to bite people who forget to swap and then think they found an edge on the wrong feed, so please name your renamed copies something obvious before you move them; 3) a single day is enough to check that your loader works and your fills look sane, and nothing else - if anyone tunes parameters until a strategy looks good on this sample they've just memorised one day of noise, not found anything. I'll use it to verify my order book columns line up and then go back to recording my own.
H
HalfKelly· Aug 2026 ago· edited Aug 2026 ago
Useful for kicking the tyres, and I'll add the obvious caution for anyone tempted to size positions off it: one day of minutes gives you a per-minute vol number whose own standard error is roughly 1/sqrt(2N) of itself, and with N=1440 that looks tight until you remember it's one regime, one day, one draw. Edge estimates are far worse, the noise in a mean shrinks like 1/sqrt(N) while the thing you're measuring doesn't grow, so a day of data will happily show you a Sharpe that isn't there. Treat this as "does my code run end to end" rather than "what fraction do I bet". Question though: with ten book levels on both sides, is there enough there to build a rough fill-cost curve, or does the one-minute frame smear it too much to be worth it?
N
NightOwl_Yuki· Aug 2026 ago
perp venues plus spot in the same window is nice, I mostly trade the overnight stretch so I'm curious whether the depth columns look thin in the hours I actually click buttons. downloading tonight, will report back if the checksum matches on a phone tether lol
TraderWeTraderWe Team· Aug 2026 ago
Yuki - the window does include the overnight stretch, so you should be able to see for yourself how depth behaves there. Betty's point 2 is fair and it's on us: until the venue is in the filename, renaming is manual and easy to get wrong. If you do swap files, check the symbol list matches what you expect before trusting any result. And to be clear on the limits, one day at minute resolution is for checking the software loads and behaves, not for drawing conclusions about any strategy.
Sign in to reply →
← All announcements