Each book asks a different question of the same market read, runs as its own capped sleeve, and keeps its own record. Every book here is live, and every figure is relative — percent up or down from each book's own starting line. Numbers are drawn from the ledgers as of 2026-09-23 and regenerate with the site build — nothing on this page is typed in by hand. Small samples are called small. What each book consumes from the engine is set out in the strategy outlines.
The flagship book. Multi-timeframe ICT chart reads on ETH and BTC every 30 minutes — market structure, sweep-failures, order blocks — executed live with a hard stop and staged take-profits on every position.
This is Eva's ICT desk. Every 30 minutes she assembles marked H4, H1 and M5 charts and reads them the way an ICT trader does: close-confirmed structure breaks, order blocks and breakers, swing-failure sweeps at prior extremes, mitigation and fibonacci entry bands — detected programmatically, then cross-checked by a vision model before anything trades.
The chart read never stands alone. A four-year cycle clock — every halving since 2012 and the tops and bottoms that followed — sets the posture she reads against, and a news pipeline ingests headlines as they break, assigning each a severity, direction and expiry. High-severity context can veto an entry or tighten a stop; it never initiates a trade on its own.
Structure, cycle and news are combined into high-quality ICT-based trade ideas: a complete plan with entry, stop, staged targets and a written thesis — or, far more often, no trade at all. Declining is the common answer, and every cited level is re-checked against raw candles before it can become a position.
The idea stream. Not an ICT desk — a programmatic scanner wired straight into Eva's brain that constantly mints structured intraday setups, every one published with entry, stop and targets, and every one scored in public.
The mill is deliberately not ICT-based. It consumes Eva's brain over the intelligence API — stance flips, breaking news, funding-regime switches, volatility spikes and session opens — and turns each trigger into a fully specified card with entry, stop and targets, around the clock. Roughly 25.3 ideas a day since 2026-08-10, 1138 published in total.
Where the ICT book waits for structure to line up and usually declines, the mill's job is constant throughput: every read the brain produces becomes a candidate idea, sized small and published with its full bracket before the outcome is known.
The live lane trades fixed clip sizes under hard halts — daily-loss, concurrency, and fills-per-day caps — and its return is shown above exactly as the ledger recorded it. Bracket geometry was rebuilt once, on 2026-09-12, and that change is disclosed rather than blended away.
The reversal lane. When three or more consecutive 15-minute candles run one way and the last one sweeps the prior extreme, the bot takes the other side of the event contract — a classic exhaustion read, settled within the quarter hour.
Kalshi's 15-minute BTC and ETH markets price a simple question — will this candle close up or down — and this lane bets that stretched runs snap back. The trigger is mechanical: three or more same-direction 15m candles where the final candle sweeps the prior extreme, entered near the window open at the mid, with take-profit and stop as fixed multiples and a circuit-breaker after consecutive stop-outs.
Directional posture comes from Eva's stance engine over the same intelligence API the other books use — the bot has never run a chart read of its own, and it refuses to trade on a stale read.
The rules were re-sized into their current form mid-September, and the record above starts there: the earlier, smaller-sized period is disclosed rather than blended in. Results settle every fifteen minutes, so this book compounds its sample faster than any other on the page.
The favourite lane. Mid-window, when one side of the 15-minute contract is priced 67–80¢ with four to ten minutes left, the bot buys it and holds to settlement — harvesting the gap between priced probability and realized outcome.
Late in a 15-minute window the market has usually made up its mind — but the priced favourite still trades below its realized win rate. This lane buys whichever side is quoted 67–80¢ with four to ten minutes to settle, pays the ask up to a hard cap, takes one entry per window, and holds to settlement. No exit management, no re-entry.
Like the reversal lane, it takes its market bias from Eva's stance API and stands down when the read is stale or the window is ambiguous.
The rule went live in its current form on 2026-09-17 after a sweep across the recorded book — the earlier wick-fade ruleset it replaced was retired with its losses kept on the all-time record, and this card starts its chart where the new rule starts trading.
The institutional sleeve. On-chain collateral and fixed-yield positions managed against hard health-factor floors — stable yield, marked to a daily NAV, and the most risk-free book Eva runs.
Built for allocators, not for punting: collateralized lending and credit-gated fixed-yield positions, managed against explicit health-factor floors with monitors and a planner that blocks any action that would breach them. Liquidation distance — not drawdown appetite — is the risk unit, which is what makes this the most risk-free sleeve on the page.
This is also the book where Eva's brain is most directly in the loop: the four-year cycle thesis and the stance read set the posture for collateral sizing — how much borrow the book carries against its collateral — while the yield legs themselves stay boring by design.
The record is a daily NAV mark published as recorded: +6.90% over 29 marks since 2026-08-26. Retail deployment is deliberately closed — join the waitlist and we will open allocations in order.
More strategies coming soon.
Every number regenerates from the system's SQLite ledgers — live fills, settled event contracts, daily NAV marks — with one script. If the ledger changes, the site changes.
Every book is shown as percent return from its own starting line, so books of different sizes read on equal terms. Live figures come only from exchange fills and on-chain marks, and no two books are ever averaged together.
Rule changes are disclosed on the chart they affect, and earlier losing periods stay in the all-time figures. A sample of seven is written as n=7, not annualized into a story.