25,484 tests, 60 discoveries, zero cascades: autopsy of a scan that passed everything — except reality

There is an idea everyone has had in front of a quote screen: "when A moves, B follows the next day." Oil leading currencies, copper leading equities, gold leading everything else. That is the snowball effect, the cross-asset cascade — and if it existed, it would be the simplest trading rule in the world.

In August 2026 we asked the question once and for all, at scale: 72 daily series, up to 40 years deep, every possible pair, every lag from 1 to 7 sessions — 25,484 exact tests. The protocol, thresholds and method were written and committed before the measurement ran, and the report is public, in its publishable version (the report, with its pre-registration).

Note, 16 October 2026. These links now lead to the publishable versions of both reports. Part of the series in this scan comes from our laboratory, under a strictly local-use licence: the full versions, which also published results computed on those series, are withheld from publication. The figures in this article that covered all the series, rather than the publishable share alone, have been removed, along with one finding established only on all of them; those that remain cover the publishable share, or are counts from the protocol.

Note, 17 October 2026. The "publishable share" also included the oil and gas spot prices distributed by the EIA: they are in fact Refinitiv data (LSEG group), relayed under a licence that does not allow redistribution, which we had taken for public data. The figures computed on that share — reproduction rate, number of exploitable tests, explained variance, bound — have therefore been removed as well. The verdict and the protocol counts remain.

What this scan returned is not the most interesting part. The most interesting part is how close it came to fooling us.