ABAlgobot

Project overview · FX and crypto

Algorithmic trading where evidence comes before execution.

Algobot is a trading system built around one rule: no strategy reaches the broker until it passes the same statistical bar as every other strategy. Ideas go through a standard lab, a forward paper trial, and hard risk gates before any money is at stake.

Forex engine live on OANDA · crypto in paper trading

1 labevery strategy, market and direction scored by the same engine
40+forex strategy–market cells tested
Pre-registeredhypotheses fixed before results are seen
Paper firstforward shadows before real orders

How it works

From idea to live order.

  1. 01

    Market data

    Broker and exchange candles are downloaded, checked for gaps and alignment, and stored as a reproducible dataset.

  2. 02

    Strategy Lab

    One backtest engine replays each strategy with realistic spreads, fees, slippage and fill rules, using the live bot's own signal code.

  3. 03

    Verdict gates

    Results must hold up statistically: confidence intervals, multiple-testing correction, stability across periods and survival at double costs.

  4. 04

    Paper shadow

    Survivors trade forward on live data with no orders, under a decision rule committed before the trial starts.

  5. 05

    Live engine

    Only then does a strategy trade, with position limits, loss limits, broker reconciliation and an emergency halt.

Markets

Two projects, one standard.

Live

Forexbot

Trades major currency pairs through the OANDA v20 API on four-hour bars. Every poll reconciles the bot's ledger with the broker, and every decision is journaled.

  • Risk gates on size, daily loss, spread, trade count and news calendar
  • Paper shadow runs beside the live engine
  • Operator dashboard with open trades, alerts and research archive

Paper trading

Cryptobot

Researches large-cap crypto on Coinbase spot and perpetual-style futures, where fees are roughly ten times forex costs, so strategies must earn more to survive.

  • Pre-registered test battery with fees, funding and contract sizes modeled
  • Overfitting diagnostics and look-ahead checks on every variant
  • Paper shadow verified trade-for-trade against the backtest

Engineering

Built to be operated, not just backtested.

Both projects share a repository but not state. Each owns its data, ledger, credentials, services and dashboard; shared code is extracted only once both projects use it.

Language
Python, pandas, NumPy, SQLite
Brokers
OANDA v20 REST and streaming, Coinbase Advanced Trade
Research
Bootstrap confidence intervals, Holm correction, Monte Carlo, deflated Sharpe, PBO
Hosting
AWS EC2 (ARM64, Amazon Linux), systemd services, nginx with TLS
Operations
Terraform, deploy scripts with backups and config checks, emergency halt switch
Interface
Static, script-light dashboards under a strict Content Security Policy

Principles

What the process protects against.

Fooling ourselves

Tests are registered before they run, and failures are published alongside wins.

Unfair comparisons

Every idea is judged in the same units, under the same costs and fill rules.

Backtest drift

The lab calls the live signal code, and shadows must match the backtest trade for trade.

Silent failure

Reconciliation, alerts and a halt switch keep the live engine observable and stoppable.

Private

Operator dashboards

Live dashboards and research results are password protected.