The Workshop · lesson 9
The index rose. The leaders were sold. What the rules saw.
From April to June 2026 the market rose 15% and the year's leaders — the top fifth of the desk's names by the previous six months' return, held at equal weight — rose 35%, and the bottom fifth went nowhere. A strength-run engine had the wind at its back. Then, in July, the index kept rising and the leaders were sold. Here it is, day by day. The leaders and laggards are re-picked on June 30 — the vertical line — the way a trend-following desk would have re-picked them.
Fact, from the desk's own price history. July 1 to August 31, 2026: the index +2.7%; the top fifth of the desk's names by prior six-month return, at equal weight, −18.6%; the bottom fifth +21.3%; half the names beat the index and half did not. The tape read Trending almost the whole time — because it was: the index was making highs. The rotation was inside the index, invisible to a chart of it.
Nothing, in advance. A rotation is the one weather a momentum-of-names engine cannot see coming: on June 30 every leader was a leader by every measure the engine keeps, and every laggard was a laggard. The rules did what rules do. Fresh runs kept qualifying in July, in the names being sold. The exits fired — Alpha back through its line, one bad close at a time — and turned what would have been a 19% hold into a run of small losses. The desk's own book lost money that summer while the index rose. Here is the whole year, month by month, for both generations' rules and the index:
Read the two generations together. Bernoulli rode the same leaders harder on the way up — three months of gains that ended with June, a month the index lost — and gave back about 5% of it in July when they were sold; Archimedes, with less of the leaders on, lost less in July and made far less before it. Across the year the current generation is ahead of the index with less than half its market exposure; the first generation is well behind both. That is the whole trade-off of a strength engine in one summer: a bruise in the rotation month, a wound avoided by the exits, and the year decided by everything around it. Held, the same names cost three times what the rules did.
The summer produced the questions Bernoulli answered. Whether a rotation can be read early was tested first and set aside: a rotation read is real pooled over years and would have said nothing useful in July. What did test out was structural. A cap on how many longs one sector may hold, then the gate that lists a run only when the name's longer trend agrees, and the sector rule that lists runs only where they have ever paid — all three keep the marginal runs of a crowded trade off the list. And a pair built on timing — the fresh riser against the stale leader in the same sector — because a rotation is exactly a hand-off, and a hand-off can be traded from both sides where a leader-vs-laggard pair could only be caught on the wrong one.
The engine page shows the year and the month the rules lost, in every generation, next to the index; the timeline shows what was tried afterwards and what was set aside. The tracker's closed view shows what your own summer looked like against the rules'. If a strategy you follow never has a summer like this on its record, it has not shown you its record.
Keep this: a rotation is invisible in the index and unavoidable in a strength engine. The exits decide whether it costs a month or a year; the record decides whether you believe the next one.
The Workshop is education, not advice. Replayed numbers are averages over many trades, after average trading costs (2 bps a side on a stock, 1 on a fund, included for fairness; yours vary by broker and liquidity) unless stated, and are not forecasts; the live record is on the engine page. Open the deskManualThe EngineChangelog