Binance perpetual futures — one day of one-minute data (2026-08-12)

Contents
One trading day of Binance Futures perpetual data at one-minute resolution, so you can run a backtest the day you install TraderWe instead of waiting for your own recording to build up.
VenueBinance Futures
Marketperpetual
Symbols64
Rows72,074
Window2026-08-11 20:00 to 2026-08-12 20:00 ET
Symbol formatBTCUSDT
Symbols in this file include VELVETUSDT, DRAMUSDT, XAUUSDT, LITEUSDT, CAPUSDT, AAOIUSDT and 58 more. Each row carries 68 columns: price, per-minute buy and sell volume, turnover, trade strength, and ten levels of the order book on both sides.

1. Loading it

Because the venue is not in the filename, only one venue can occupy that folder at a time. Swap the file when you want to test a different one. We know this is awkward and it is on our list to fix. 1) Put the file in your data folder under record/futures/1m/ — create the folder if it is not there. 2) Rename it to futures_1m_20260812.db. The app reads the asset class from the folder and the date from the filename; the venue is not part of either. 3) Open Backtest. The date shows up in the available-data list and you can run against it.

2. What this is

Our own recording, captured from Binance Futures's public market data feed by our recorder in Tokyo. Nothing here comes from a paid data vendor. 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. Serious research wants more days than this and the one-second frame. The widest symbol list of the five and the most active tape, which makes it the easiest file to test a long/short strategy against. Everything is quoted in USDT, so a strategy comparing across venues has to account for the quote asset rather than assume dollars. If a download looks short, check the SHA-256 shown next to the file. A truncated database opens as a corrupt file rather than failing loudly.
TraderWe
169 posts · 2 followers
+ Follow

Files

binance-futures_1m_20260812.db41.7 MB
SHA-256 601976751f29666aec313247f04090fdbe65d86c605fbdeb293f8a826a98b55c

Related reading

3 replies

MLcurious· Aug 2026 ago· edited Aug 2026 ago
ok so dumb question, 68 columns with 10 book levels each side, is that already the aggregated depth or raw top-of-book snapshots at the minute close? i want to try a tiny model on the imbalance but if it's just a snapshot at :00 i'm probably learning noise. also 72k rows over 64 symbols means some symbols barely have any minutes, right? do gaps show up as missing rows or filled forward? want to make sure i don't feed my thing garbage on day one lol
DataDrivenDee· Aug 2026 ago
MLcurious raises the question I'd want answered too, and there's a way to check it yourself rather than wait: if the ten levels are a snapshot at the minute close, the book totals should be jumpy from bar to bar on the thin symbols and basically uncorrelated with the volume in that same minute. If it's aggregated over the minute it'll look much smoother. Take one quiet ticker and one busy one, plot depth against the minute's traded volume, and the shape should tell you. Also, honest question for the imbalance idea — with one day you get maybe 1,400 bars per symbol, and the 64 symbols are all crypto so they're not independent samples. How would you tell a real edge from a day-shaped one?
TraderWeTraderWe Team· Aug 2026 ago
Not a dumb question, and the answer matters for what you were about to build. The ten levels are a snapshot at the close of the minute, not aggregated depth. The row carries the book as it stood in the last second of that minute. Volume and turnover are summed across the minute; price, trade strength and the ladder are the closing second. Gaps are missing rows, not forward fill. A minute row is only written when a new snapshot arrives that crosses the boundary, so a symbol that went quiet for six minutes has no rows for those six. And yes on the thin ones: 72,074 rows over 64 symbols against 1,440 minutes in the window averages about 1,126, so the average symbol is missing a fifth of the day and the quiet ones are far worse than that. Count rows per symbol before you trust anything per-symbol. DataDrivenDee, your test would have given you the right answer. Depth uncorrelated with the same minute's volume is exactly what a closing snapshot looks like.
Sign in to reply →
← All downloads