The engine, exposed.

Deterministic multi-timeframe technical read (SMA trend + momentum confluence). The engine trades long-only in confirmed uptrends and holds cash otherwise.

Grouped by instrument because they are not comparable: a perp's P&L is on notional rather than the collateral posted, and staking is SOL-pegged so its USD move is SOL's, not a result — both are kept out of the headline. Cost is reconstructed from the wallet's own signed history wherever the chain tag appears; a holding whose buy we cannot reconstruct shows its value and says basis unknown rather than pretending it was free.

Realized P&L from every closed round-trip across the swarm, reconciled on-chain. ROI is shown as the median vs SOL — a near-zero-cost-basis remainder can read +15,000% and would poison a simple average.

Every close is attributed to the sleeve that opened it. Sharpe is the per-trade ROI mean ÷ stdev (a comparable risk-adjusted read, not annualized). This is how we prove which of the newly-enabled sleeves earn — and tune or cut the rest — before scaling. 742 attributed round-trips so far.

Every close is attributed to the lane that opened it, joined by the open signature — not by timing. Judge a lane on avg ROI, not on dollars: lanes deploy different position sizes, and comparing totals across them turns a sizing difference into a fake strategy result. Coverage 47% — 741 of 1580 round-trips carry a lane. The rest closed before lane stamping shipped, so the numbers above describe only the attributed portion (−$1,556.20 sits outside them).

Every level is a different game — the engine changes how it trades as the bag grows (risk scales down, capacity caps kick in, an income base builds). Checkpoints are enforced on-box: at the floor, new risk halts until the book recovers. The SCORE is SOL grown — every deposit (stables counted in SOL at deposit-time price) moves the start line, never the score; withdrawals lower the checkpoint floor instead of counting as a loss.

Every agent reports its closed trades to one shared brain, which learns which hold times and which tokens actually work and feeds that back to the swarm. Figures below are medians — memecoin returns are heavy-tailed, and a mean describes the outlier rather than the typical trade. Realized P&L is not restated here; the chain-derived figure at the top of this page is the single source for it.