Research
Published outputs from our systematic trading research.
Literate Exploratory Development (LED) — with ccxt-multibrokers as a worked example
Literate Exploratory Development (LED) is a way of building software that brings the rhythm of exploratory data analysis (iPython / Jupyter, the REPL) to system design, implementation, and documentation. Instead of writing a complete blueprint up front, you advance in small steps — and each step produces a written document, a small implementation, and an immediate verification against real data — then integrate the whole at the end.
It sits at the confluence of two established lineages: exploratory programming (try small, decide the next move from immediate feedback) and Knuth's literate programming (grow code and its explanatory prose together). The loop, in one line:
- Decide the next small step → write it up as a document → implement a small piece → run it and verify immediately → reflect the results back → repeat, and integrate at the end.
ccxt-multibrokers is a worked example built entirely in this style: a read-only cross-exchange cryptocurrency arbitrage monitor. It handles CCXT-supported and unsupported exchanges through one common interface, lets you select target exchanges and trading pairs, collects order books in parallel, and detects arbitrage opportunities net of fees and minimum order sizes — then notifies. It performs no order placement or trading. Its ten-part document series (01–09, 99) is itself the record of the LED process.
OANDA NY / Tokyo Server Rate Comparison Tool
Even within OANDA, the New York server (v20 REST API) and the Tokyo server (MetaTrader 5, Equinix TY3) can stream different rates. This tool fetches quotes for multiple currency pairs from both servers and compares them side by side with tables and charts. Documentation is in Japanese.
- Two modes: historical comparison (M1 candles) and live comparison (tick polling)
- Charts show both servers' mid prices on top, with the NY − Tokyo divergence in pips below
- On real data, divergence of roughly ±0.3 pips and cross-server spread differences are observable even in calm markets
- Results are saved as PNG / CSV per currency pair