Regime-gated overnight long, AI infra 8

The rules

At 15:46 ET, buy any of eight AI-infrastructure stocks above today's open but down over the prior five sessions, and sell at 09:35 the next session. The check runs on each stock's 15:45 one-minute bar: it buys 12.5% of equity when that bar closes above the open of the stock's first bar that day and the previous session's close is below the close six sessions before today. A stock with no trade in the 15:45 minute is skipped that day, and a stock can be bought only after six sessions of its own in the test. The order fills at the next minute's open, 15:46. On a half day the check runs on the 12:45 bar and fills at 12:46. The sale is decided on the first bar at or after 09:34 in the next session and fills at 09:35. Version 6 holds only long positions; it never sells short. It opens no new trades from March 7 to May 12, 2025 or from March 23 to April 8, 2026, both ends included, though it still sells what it holds. Each range starts the session after QQQ, the Nasdaq-100 fund, first closed below its 200-day moving average and ends on the session it first closed back above. They are fixed dates because a program reads only its own stock's bars and cannot check QQQ, so trading it live would need a daily task to update them from QQQ's close; as written, nothing stops it buying after April 8, 2026.

Universe

Eight AI-infrastructure stocks: APLD, NBIS, OKLO, WULF, GLXY, CIFR, IREN and CORZ

Bars

1-minute

Position size

12.5% of equity per stock; all eight at once use the whole account

Rebalance

Every session: buys at 15:46, sells at 09:35 the next session

Results

Hypothetical: a backtest with modeled costs, not traded with real money.

Forward testing since October 29, 2026, with no live market gate: version 6's date ranges end in April 2026. Backtest numbers are from version 6, one of seven versions tried, in the strategy's 11th run.

Hypothetical growth of $10,000

Rebuild it in Orbital

Paste this into Orbital Chat. The Strategy Developer builds the workflow and asks before it runs any backtest.

N01's own run used 0.03% slippage a side. A backtest from Chat assumes Orbital's default of 0.05%, so it comes out a little lower.

Trade APLD, NBIS, OKLO, WULF, GLXY, CIFR, IREN and CORZ on 1-minute bars during regular hours. Build it as a program that uses only 1-minute data, decides on every bar and has no warm-up. Each session's open is the open of the stock's first 1-minute bar that session, and its close is the close of its last one, whatever minute that is. On each stock's 15:45 bar (12:45 on a half day), buy with 12.5% of equity if it holds no shares and has no order pending, the bar closes above that session's open, and the previous session's close is below the close six sessions before today. Check only that bar: a stock with no 15:45 bar is not bought that day. Buy a stock only after six sessions of its own inside the test. Sell on the first bar at or after 09:34 in any later session. Orders fill at the next bar's open. Open no new positions from March 7 to May 12, 2025 or from March 23 to April 8, 2026, both ends included, but keep selling inside them. These ranges are the sessions right after a QQQ close below its 200-day moving average; write them in as fixed dates, since a program reads only its own symbol. Long only, on a margin account so all eight buys can fill on the same day. Backtest from January 2, 2024 to October 27, 2026 with $100,000.

Is it real?

Versions tried

7 versions; this is the strategy's 11th run

Discounted Sharpe

0.20: the raw 2.45 less the 2.25 the best of 7 tries would reach by luck

Chance it beats luck

64.5% that its true Sharpe tops 2.25

Market gate

Two date ranges typed in after the fact; live, they need updating daily

Listings

4 of the 8 traded under these tickers at the start; all 8 from May 16, 2025

Beta to SPY

0.19

Alpha vs SPY

+40.1% a year

Data

1-minute bars, 708 sessions

Costs

Alpaca's $0 commission, 0.03% slippage a side, next-minute fills

Read this before you trust it

Seven versions were tried, four of them built to sell short. Their Sharpes ran from −1.58 to 2.45, and the 2.45 is this run's. With that much spread, the best of seven tries with no real edge would be expected to reach about 2.25 by luck alone. This run clears that bar by only 0.20, and the chance that its true Sharpe is above it is 64.5%. Several of the seven ran over shorter stretches, which tends to widen the spread and raise the bar, so the discount errs on the strict side. Orbital's report shows only the raw Sharpe; its Backtesting strategy list and Chat show the discounted one. The two date ranges were typed in from QQQ's history after both had passed, and the strategy sat out both. They do not move in the forward test: QQQ closed below its 200-day moving average on November 5, 2026, and version 6 can still buy. The eight stocks were picked in October 2026, with hindsight of how this group had done. Four of them were not trading under these tickers on a US exchange when the test began: CORZ relisted on January 24, 2024, OKLO first traded on May 10, 2024, NBIS resumed trading on October 21, 2024, and GLXY listed on May 16, 2025. Each needed six sessions of its own before it could be bought, so on those listing dates the test could hold at most six of the eight until late October 2024, and all eight only from late May 2025. Its deepest fall from a high, about 14%, began on October 2, 2026 and had not recovered by the end of the test. The run was set to end on October 28, 2026, but its data ends on October 27, the last settled session when it ran, so the prompt ends there. Slippage of 0.03% a side is a thin allowance for volatile stocks bought near the close and sold five minutes after the open.