Why one hand on the rack and not ten

A desk that sells access to trading systems has every commercial reason to list many of them. A long rack looks like depth, every member finds something that suits their temperament, and when one hand has a bad quarter another can be pointed at. This desk listed eight. On October 1, 2026 the founder stood down seven of them and left one, The Double Down, dealing on paper. This page is the reasoning, written so that a member can check it rather than take it on faith — and so that the next time a hand is added, the bar it has to clear is already public.

Ten hands on one index are not ten bets

Every hand on the rack trades the same contract, the Micro E-mini S&P 500. When the index has a bad week, every long-side hand on the index has a bad week together; the losses do not average out, they add. Diversification is a property of things that move differently, and ten rules for entering the same instrument move, mostly, with that instrument. The spread of outcomes across the rack was never the spread of a portfolio; it was the spread of a single bet seen from ten angles. A member who dealt five hands "to diversify" had, in the only sense that matters for a drawdown, taken five times the position.

A netted account turns two hands into one

There is a second, more mechanical reason, and it lives in the broker's ledger rather than in statistics. A futures account nets by symbol: it holds one position in MES, not one per strategy. The desk's own rules (the ones every routed order obeys) make every exit a close, which flattens the symbol on the account — the only safe way to guarantee a strategy's exit never leaves a stray position behind. Put two hands on one account and one symbol and the arithmetic is immediate: one hand's exit closes the other hand's trade. The ledger would record two independent round trips; the account would hold one netted position that whichever hand exited first had flattened. The desk forbids that combination outright — no two enabled slots may share a broker account and a routed symbol — and the rule is enforced in the database and re-checked on every fan-out. It is not a rule that limits members; it is a rule that stops the rack from quietly contradicting itself.

What the records actually supported

The rack's first hands were chosen on short windows of data, in a vocabulary of classic intraday entries — opening ranges, breakouts, pullbacks, inside bars — with the day's position flattened at the bell. The desk's engine, run across the full corpus of CME hourly bars from 2010, did not support them: the 9,363-configuration sweep on the win-rate page found that a configuration's win rate explained almost none of whether it made money, and the larger search that followed (52,380 configurations of the classic vocabulary, at every timeframe the desk trades, with the strict split and gates described on the backtest page) produced zero holdout survivors. The hands that remained in the rack by August were the ones the engine had measured, and of those the only one that cleared every gate — a fit period, a holdout the search had never seen, an edge over simply holding the index, and survival at doubled costs — was a patient double-bottom hand on the hourly chart that holds its trades overnight.

That last clause matters. The desk measured its day-only twin: the same entries, flattened at 15:55 ET, take the hand's profit factor from 1.59 to 0.84 over the full span. The overnight hold is the hand's edge, and the desk says so on every page that mentions it, because a hand that holds overnight needs the overnight margin, which is about fifty-seven times the day rate (the margin page has the figures with their date), and a member has to know that before dealing it.

The search for a flat-at-the-bell hand

A fair question — the founder's, in October 2026 — is whether a hand exists that is both profitable and flat at the close, so that nobody is exposed overnight. The desk ran the search rather than guess at it. The chart-pattern vocabulary (engulfing bars, pin bars, double tops and bottoms, head-and-shoulders, flag breaks, pullbacks to moving averages, each across stops, targets, trend gates and both directions) was tested on the hourly, thirty-minute and fifteen-minute frames under a hard flatten at 15:55 ET and, separately, inside the New York session only. Same split as always: fit on 2013–2022, holdout from the split date to August 2026, with the holdout never opened until a configuration had cleared the fit gate.

On the hourly frame: 3,072 configurations, none cleared even the fit gate; the best any pattern family managed was a fit-period profit factor of 1.09 (flag breaks), with most families below 1.0 before the holdout was ever opened. On the thirty-minute frame: 4,608 configurations, again none through the fit gate, the best family at 0.97. The Double Down, run through the same code on the same hourly bars for comparison, cleared both gates (fit 1.53, holdout 1.81); its flattened twin failed both (fit 0.95, holdout 0.73) — and the same hand re-read on thirty-minute bars barely broke even (profit factor 1.05 over the full span), which is its own lesson: the edge lives at one frame, and a hand is a frame as much as a pattern. On the fifteen-minute frame: 4,608 more configurations, none through the fit gate, best family 1.09 — and The Double Down itself re-read on fifteen-minute bars loses outright (profit factor 0.82). Twelve thousand two hundred and eighty-eight configurations under the day constraint, zero through the first gate; a null is a result the desk publishes.

The desk also measured the question in between — may the hand hold overnight, but only for a bounded number of sessions? The Double Down's average trade lasts about 112 hours, five calendar days, and 87% of its trades cross at least one night. Cap the hold at one day and the profit factor over the full span is 0.91; at two days 0.96; at three days 1.07; at five days 1.34; at ten days 1.46 against 1.64 uncapped. Every cap short of a week costs more than it saves, and no cap removes the overnight exposure, because the trades that pay are the ones that take several sessions to reach their target.

Why does the flatten cost so much? Because a meaningful share of an index's move happens between sessions, and a swing hand that enters on a pattern and exits on a target or a stop is paid, in part, by the overnight drift that a day hand hands back every afternoon. A flat-at-the-bell rule is a choice to buy safety from overnight gaps with the part of the return that lived in them. It can be the right choice — it is the choice the founder made with real money by standing everything down — but it is a cost, not a free improvement, and the search shows the size of it.

The number of hands that have earned it

Bailey and López de Prado's point about trials applies to racks as much as to configurations: a desk that tries many hands and lists the ones that look best has selected on noise unless each one cleared gates set before the search, and the right correction for the number tried makes most of them disappear. One hand that cleared the gates and reads the same on a holdout it never saw is worth more than ten that were each the best of a sweep. The honest length of a rack is the number of hands that have earned a place on it. The desk's answer in October 2026 is one, on paper, under a live lock that keeps real money out until the founder lifts it. A second hand joins when it has cleared the same gates on data it has not seen and the account it would share has room for it under the netting rule — not when the rack looks thin.

• How a futures contract works, in money terms
• Margin — day, overnight, and why the number moves
• What a backtest can and cannot tell you
• How to read a win rate
• What holding overnight really takes

← Futures — what the desk does · All of Learn