Abstrakt

Trading

Where we prove our own methods

Trading is our internal proving ground. The automation discipline we sell to clients is the same machinery that researches, executes and monitors our own capital — end to end, unattended.

Flagship program

Abstrakt Macro

A systematic program that follows major market trends across global assets — equities, rates, commodities and crypto. Signals, execution and risk management are fully automated. The methodology stays private by design; the numbers speak for it.

Abstrakt Macro vs S&P 500

Growth of $1, log scale. The live account continues the backtest curve; live returns are time-weighted with deposits excluded.

CAGR

Max drawdown

Sharpe ratio

Live return

Architecture

Code executes. Agents investigate.

The familiar demo is an AI agent handed a broker account and told to trade. We build the opposite, because it is the version that survives contact with real money: deterministic code owns every live action, while the agents sit outside the trade path — measuring what the system actually did, running the investigations, and proposing changes that are reviewed before they ship.

Execution Code only

The live path is deterministic

Market data, signals, position sizing, risk limits and order placement are plain code on a schedule. Same inputs, same orders, every time — testable, reproducible, and reviewable line by line. Nothing in this path is asked to improvise.

  • Market data
  • Signals
  • Risk & sizing
  • Broker orders
Fills, positions, job runs
Observation Economic dashboard

Everything it does is measured

Every fill, position, scheduled job and failure lands in one dashboard: capital, time-weighted returns, drawdown, execution lag, pipeline health — and an alert when a number leaves the range it should be in.

  • Capital & returns
  • Drawdown
  • Execution lag
  • Job health
Anomalies worth explaining
Research AI agents

Agents study the record and propose fixes

A scientist agent picks up the anomalies the dashboard surfaces and does what a researcher would: form a hypothesis, rebuild the test independently, check for lookahead and survivorship bias, price in real costs, and report what it found — including, often, that the edge was never there. That last answer is the valuable one, and it is the one a trading agent has no incentive to give you.

  • Hypothesis
  • Independent re-test
  • Bias & bug hunt
  • Proposed change

Reviewed changes are deployed back into the execution code. That is the only route by which the live path ever changes — no agent edits a running strategy on its own.

The agents change the system. They never place the trade.

Prediction markets

Event-market bots

We also design and run automated strategies on prediction markets such as Kalshi and Polymarket: bots that read market data, price event probabilities and manage orders end to end — profitably. They double as a public demonstration of our automation applied to a hard, adversarial domain.

Verification

An audited track record, in progress

Backtests are a starting point, not proof. We are building a publicly audited track record on Darwinex so performance can be verified by an independent third party rather than taken on faith. The link will appear here when it goes live.

This page is for information only and is not investment advice or an offer to buy or sell any security. Backtested results are hypothetical and do not represent live trading. Live results are from a real account over a short period. Past performance does not guarantee future results.