Inside the platform: what RegimeSignal actually publishes

The headline read for the S&P 500: where the market sits in its cycle, whether conditions are strengthening or weakening, and a year-end range with the probability attached to it — never a single number pretending to be a forecast.

Alongside it, two things most commentary never gives you: what the model is watching, and what would change the call.

Four models carry measured out-of-sample precision and are named wherever they appear: the Bear Regime Signal (BRS), two drawdown classifiers (MBS T1 and MBS T2), and the Recovery Signal (RRS). They are walk-forward validated — trained only on data available at the time of each call, never on the answer — and they are the only work on the platform covered by the independent validation and audit engagements.

They are held apart from everything else on the platform, deliberately. The line between a validated signal and a research finding is the one distinction the whole product rests on, and blurring it would cost more than it could ever sell.

What has actually followed a given setup — how often, against what baseline, and whether the sample is large enough to mean anything. Calendar and election-cycle effects tested rather than asserted. Index, sector, global, commodity and thematic readings, each measured against its own threshold.

These are not forecasts, and the platform says so on every one of them. They are the base rate you would be arguing against if you took the other side — which is the thing almost no market commentary gives you at all.

Every name in the screened universe is read through independent measures — earnings estimates, earnings growth, relative strength, technicals, accounting quality, analyst dispersion, share issuance, insider activity, valuation.

They are never averaged into a single score. That is a deliberate choice: averaging hides the disagreement, and the disagreement is the information. A name where every measure agrees is a different proposition from one where half of them are pulling against each other, and a single blended number makes those two look identical.

• How much of the evidence agrees — offense, neutral or defense, with the count printed beside it so you can see what it rests on.
• Whether it is improving or weakening — which measure changed, and when.
• Whether that change was its own, or the whole market's.

Index coverage across the Americas, Europe/Middle East/Africa and Asia-Pacific, plus credit, metals, energy, crypto and FX — each with day, year-to-date, three-month and one-year change and a one-year trend line.

This is the one call on the platform that does not name a direction. It says how big, never which way.

It measures how often the following quarter delivered a move of more than ten percent in either direction, grouped by how far the instrument sat below its own running peak, counted across its entire price record.

The platform states plainly that the finding is suggestive rather than established, that the observations are not independent because drawdowns cluster in crises, and that no signal on the platform reads it.