Research · 6 min read
Do they fail together?
A portfolio of seven engines is only a portfolio if the seven do not lose on the same days. That is a testable claim, not a slogan, so here is the test — every pair, every trading day, from the same daily record published on the calendar.
Every pair, measured
Each cell is the correlation of two engines’ daily results, calculated only on days when both were running. 1.00 would mean they move as one. 0.00 means knowing what one did tells you nothing about the other. Below zero — green — means one tends to be up when the other is down.
| Coil | Iris | Aperture | Rift | Surge | Ripple | |
|---|---|---|---|---|---|---|
| Vault | 0.13 | 0.15 | 0.16 | 0.08 | 0.30 | 0.09 |
| Coil | 0.04 | 0.00 | 0.10 | 0.01 | 0.12 | |
| Iris | 0.57 | 0.15 | -0.08 | 0.22 | ||
| Aperture | 0.06 | -0.21 | 0.25 | |||
| Rift | 0.14 | 0.13 | ||||
| Surge | — |
Hover a cell for the number of shared days and how often the pair was down together. Pairs with fewer than 30 shared days are left blank rather than shown as a number that cannot carry weight.
An average of 0.12 is the number that matters. These engines are not seven versions of the same idea wearing different names — on most days, what one does tells you very little about what the others did. That is the entire mechanical argument for running a book instead of a single strategy, and it is measurable rather than asserted.
One pair stands out and it would be dishonest to bury it. Iris and Aperture move together at 0.57 and were down together on 34% of their 155 shared days. That is not a coincidence: they read the same kind of setup, in the same session, on two closely related index markets. Running both is closer to one position than two, and anyone building a book from these engines should size that pair as a single bet rather than a pair of independent ones.
The same thing, day by day
Numbers hide the texture. Below, every trading day in the published record is a row and every engine is a column: green for a winning day, red for a losing one, brighter with size. The faint columns at the top are days before that engine started.
What to look for: a horizontal band of red running clean across the row is the book failing together. Scattered red, with green beside it on the same row, is the book doing its job.
How often does the whole book go red?
During the published backtest period, simultaneous losing days occurred on 19 of 255 days when five or more engines were running (7%). The largest combined loss among those simultaneous losing days was $1,187 across the seven-account hypothetical portfolio (2026-07-28).
The same test the other way: every running engine finished the day up on 29 of those 255 days (11%). So on 81% of days the book was split — some engines up, others down on the same session. That split is what a portfolio of independent edges is supposed to look like, and it is the reason the whole book rarely moves as one.
What this does not prove
Correlation measured over one published window is a description of that window. In a genuine market-wide shock, things that normally have nothing to do with each other can move together, and a book of seven engines is not immune to that. The figures above are backtested, they carry the limitations set out at the foot of this page, and the honest reading is “these engines were largely independent across this record” — not “these engines cannot lose together.”
It is also why the pair flagged above matters. Independence is a property to be checked and re-checked, not a badge earned once.