A systematic trading lab. Every strategy is validated on data it has never seen, then paper-traded against the S&P 500. Most ideas do not get that far.
Live7 of 10 sleeves ahead of the S&P 500 since start
The hypothesis and the full parameter grid are fixed before any number is seen.
Every cell is written up, not just the winner. The losers stay on the record.
A slice of data stays untouched until a candidate is adopted — then it is the only test that counts.
Survivors face permutation nulls and cost stress. What fails goes to the rejection ledger.
Runs several sleeves in one shared account: an ETH trend sleeve, six smaller crypto sleeves at 10% each, a 28-ETF trend sleeve, and an adaptive ETH forward test. The numbers shown are the ETF sleeve only, not the whole account.
Backtest +4.2% ETF leg (EW) · Sharpe 0.66 · Max DD −12.3%
Ranks 59 large-cap stocks with a machine-learning model that keeps retraining on newer data, then holds the top eight. Live filters keep the inputs clean and cut exposure in bad markets.
Backtest +28.1% (2021–26 OOS, a_k8) · Sharpe 1.26 · Max DD −19.4%
A long-or-cash trend rule on Bitcoin: hold while price is above its 200-day average and momentum is positive, otherwise sit in cash. Built to sidestep bear markets.
Backtest +143.3% total (2021–26) · Sharpe — · Max DD −21.1%
Buys stocks after the largest positive earnings surprises and holds them about three months, but only while market volatility is below its one-year median. Sits in cash when volatility spikes.
Backtest +42.0% cum (2023+ OOS) · Sharpe 1.01 · Max DD −11.5%
Trades a daily shortlist from an LLM research model: the top eight names, equal-weighted, long only, filled at the next open. A kill-switch flattens the sleeve at a 5% drawdown. Live since 2026-09-19. The account is shared with the RL allocator test, so the figures shown are account-level.
No backtest · forward paper track
A forward paper test of an hourly momentum book on US stocks: each hour it goes long the top 20 and short the bottom 20, equal-weight and dollar-neutral. Unproven and expected to lose — the gross signal is weak and the trading costs are heavy. It runs on a reused account to record real fills; the monthly Conviction Stack is the validated cadence.
Backtest REJECT — net −97% OOS @10bps · Sharpe 1.61 gross (net negative) · Max DD −97%
Buys the strongest large-caps by six-month momentum, rebalanced about monthly, and moves to cash when credit markets get stressed. Long only. Live since 2026-09-08.
Backtest +1065% base (OOS) · Sharpe 1.25 · Max DD −41%
Intraday, Monday only. Before the open the book ranks the S&P 500 by its weekend gap, buys the 20 biggest gap-down names at the open, and sells them at the close. Costs matter here: only commission-free equities make it viable.
Backtest +57% (OOS, 2bps) · Sharpe 0.50 · Max DD −26.7%
A residual-momentum base book (top-50 large-caps, held about a month) whose weights are tilted by a per-name reinforcement-learning conviction score instead of equal weight. Live since 2026-09-28.
Backtest +4345% (OOS 2016+) · Sharpe 1.68 · Max DD −33%
A real brokerage account, mirrored here read-only. Positions are picked by hand with no systematic rule, and deposits or withdrawals never distort the return.
No backtest · forward paper track
Every idea gets written up, pass or fail. The papers cover the ones worth reading; the lab notes are the running log of everything else.