Market data, collected and checked against the exchange
I build collectors for exchange data and review backtests and trading libraries by comparing their numbers with what the venue itself reports. Crypto perpetuals, Python and ClickHouse.
What I build and check
Collectors
Websocket feeds into ClickHouse, with reconnects and batch inserts. The one behind my research has been recording Binance futures (order book, trades, mark price, liquidations, open interest) unattended since June 2026.
Backtest review
Lookahead, fill prices, fees, funding and liquidation. I rerun the engine on the same period and compare it with the exchange's own history.
Exchange adapters
Pagination limits, contract units, the side of a trade or a liquidation. Most of my merged patches are in this layer.
Research
Event studies and walk-forward tests, with the pass mark written down before the run.
Code and write-ups
strategy-graveyard: four months of tests of retail trading edges, each closed with a number, plus the collector and scripts that produced them.
Patches: 8 merged into ccxt and cryptofeed since October 2026. Others are open in nautilus_trader, vectorbt, zipline-reloaded and hummingbot.
dev.to/stooq1: extreme funding on 658 Binance perpetuals, and a test I failed after writing the pass mark down in advance.
Remote work on market-data pipelines, exchange integrations and backtest reviews. A one-off look at a single strategy or data feed works too. What to include.