Evidence
A measurement layer is only as good as its documented track record. This page sets out how we define proof, what the readings have shown — including on past crises — and what they cannot show.
How we define proof
Claims about early detection are cheap. So we hold our readings to definitions stated in advance:
- Lead time — the interval between a documented regime change in the reading and the corresponding move in a stated benchmark (aggregate volatility, index drawdown). Counted only when the reading was published or timestamped before the event.
- False positive — a stress or rupture label that is not followed by a qualifying market event within the stated window. We count these and say so.
- False negative — a qualifying market event the reading did not flag. Counted with the same discipline.
- Out of sample — historical replays use only data available at each point in time. The computation is deterministic: anyone can re-run any reading and obtain the same result.
Replayed on past crises
Because the computation is deterministic, the engine can be replayed on historical data as if running at the time. Three forensic replays are maintained in the platform’s library, with the underlying methodology documented in our published research:
Dot-com — 2000
The full unwind replayed day by day — what the structure showed while price markets still looked unremarkable.
GFC — 2008
The long build-up of structural stress ahead of the crisis, replayed on the data available at each point in time.
COVID — 2020
The fastest crash in modern market history, replayed against what the geometry read in the weeks before it.
The detailed charts, thresholds and lead times are not published on this page — they are shown live, point by point, in a discovery session, where you can question every reading.
Published every week, in public
Since April 2026, our weekly analysis of the manifolds — The Three Manifolds — has been published openly, with dates, values and thresholds on record, and recent Issues pre-register falsifiable scenarios that are audited in the following Issue. We let the archive speak for itself: read it, check the dates, and form your own view.
The limits, stated plainly
Structural readings measure the geometry of price relationships. They do not predict prices, name dates, or quantify magnitudes. A tension label can persist without a qualifying event — that is a false positive, and it happens. Idiosyncratic shocks with no structural precursor — a fraud revealed overnight, an exogenous headline — are outside what any structural measure can flag. And a reading on a broad universe does not automatically transfer to a narrower one: validity is defined per universe, which is precisely what a scoped audit establishes.
If a claim on this page ever appears to outrun these limits, hold us to it.
The mathematics behind the readings
The methodology is not proprietary folklore: it is documented in five research papers available on SSRN, covering the geometric framework, the stress measures and their statistical behaviour. The Research section links each paper and maintains a glossary of the terms used across the platform.
Judge it on this week’s market.
In a 30-minute discovery session, we demonstrate the platform on current market data and identify where structural risk hurts in your process. You decide the next step.