How the regime gauge works.

Purpose and boundaries

The gauge summarises whether the prevailing backdrop appears more supportive or more pressured. It combines economic activity, inflation, labour, financial conditions and market confirmation. It is not a forecast of market returns, a recession probability, a valuation target or an instruction to buy, sell or switch an investment.

The model is deliberately defensive: unavailable or expired readings do not receive a neutral score and do not vote. Coverage and confidence must clear minimum standards before a result is considered publishable.

The five pillars

Weights apply at pillar and subfactor level. Individual readings are not simply averaged. Related readings are grouped into correlation families so that several versions of the same economic message cannot dominate the result.

Current V2 voting set

The controlled release candidate has 23 designed US-led voting slots. Its first governed live snapshot had 19 usable readings; four unavailable readings were excluded rather than filled with assumed neutral values. Each reading has a documented source, direction, transformation, cadence and freshness rule.

The two yield curves belong to one correlation family and therefore cast one capped family-level contribution rather than two independent full votes.

From a reading to the gauge

1. Validate the evidence. The source, observation date, history and transformation must satisfy the governed contract.
2. Assess freshness. Current readings can vote; warning-stage readings receive a confidence haircut; expired or invalid readings are excluded.
3. Place the reading in history. The current value is converted to a directional historical-percentile pressure score. The outer 5% tails are winsorised and ordinary readings cannot be presented as absolute 0 or 100 extremes.
4. Control correlation. Related indicators are combined within correlation families, then subfactors, before reaching their pillar.
5. Apply pillar weights. Available pillars are combined using the weights above. Missing evidence is not filled with an assumed neutral value.
6. Govern the headline. The score is smoothed to reduce reaction to one release. A regime change must clear a three-point buffer and be confirmed twice.

Publication and confidence safeguards

A result requires at least 60% weighted coverage, at least three available pillars, both Growth and Financial Conditions, and a confidence score of at least 55. Confidence reflects coverage (40%), freshness (25%), source quality (20%) and agreement across pillars (15%).

These controls reduce false precision. They cannot eliminate model risk or guarantee that a reading will match future economic or market outcomes.

Broader global evidence under evaluation

NorthStar is monitoring 52 additional regional readings across the Eurozone, United Kingdom, Japan, China and six emerging economies excluding China: India, Brazil, Mexico, South Korea, Indonesia and South Africa.

These readings are monitored evidence, not 52 equal gauge votes. A region must pass history, availability, confidence and breadth gates before it can enter the live model. Country readings will first be aggregated into regional and subfactor composites to prevent countries with more available series receiving disproportionate influence.

Long-term structural context

Twelve additional observations describe valuation and balance-sheet vulnerability: Shiller CAPE, excess CAPE yield versus real bonds, a transparent US corporate-equity-value-to-GDP proxy, and household, corporate and government leverage across advanced economies, emerging markets and the G20.

These readings may appear in the Valuation and Stress areas of the app as Long-term context. They do not vote in the regime gauge, radar or scenario scores. Valuation and leverage can remain elevated or depressed for years; they are included to encourage wide-angle reflection, not reactive trading.

Important limitations

• Historical relationships can change, and economic data can be revised.
• A broad aggregate can conceal important differences between countries, sectors and households.
• Market prices can move before, after or independently of the measured economic backdrop.
• Proxies are clearly identified and are not represented as the exact underlying concept.
• Scenario comparisons use observed history, published research or explicit modelled assumptions; they are not forecasts.
• No methodology can remove uncertainty, model risk or data-provider risk.

Methodology change log

28 July 2026

Published the V2 release-candidate framework, designed voting set, first-snapshot availability, weighting hierarchy, global shadow inventory and non-voting structural-context policy.