A simulated $10,000 account. Every position below is a real fill against fake money, mirrored into an Alpaca paper account. Two models, two different clocks: Haiku 4.5 reads the news on a fixed 30-minute timer while the market is open, and Opus 5 has no timer at all — it wakes only when Haiku or the chart screener finds something, plus fixed anchor runs at midday and after the close.
A second simulated $10,000 account, separate from the one above, on a venue with no bell, no weekend and no holiday. Eighteen tokens across six sectors — majors (BTC, ETH), L1 (SOL, SUI), memes (BONK, WIF, TRUMP, DOGE, PEPE), DeFi (AAVE, UNI, CRV, JUP), liquid staking (LDO, JTO) and DePIN (RENDER, FIL, PYTH). Eight are Solana-native, so a Solana story moves half the book at once and the harness treats it as one bet, not eight. It ticks hourly around the clock, but it only looks for new trades while the US market is open — outside those hours the run walks the stops, the trailing stops and the 24-hour scalp clock and calls no model at all. Its risk numbers are its own — a −7% day is noise here, so the stop is −15%. The equity book now sends its stop to the broker with the entry; Alpaca accepts no such order on any crypto pair, so every stop here is the harness polling the price, and the tokens Alpaca does not list are ledger-only. A ⚡ scalp row shows its remaining clock — the harness force-closes it at 24 hours, winner or loser.
Structural bets on a 1–3 year view: where demand is heading, and who gets paid for it. Re-scored twice a week — Sunday evening with the week-ahead digest, and Friday after the close — and never mid-session. Each one carries what would prove it wrong.
Positions the brain actually opened, added to or closed. These cleared every discipline gate. Open a row for the full technical and news analysis behind it.
Ideas the brain published but did not act on, and the exact reason why: a discipline rule blocked it, or it never rose to a proposed trade. Watchlist material, not positions.