There is a gap at the center of how organizations currently handle geopolitical and political risk. On one side are the information sources: news wire feeds, government document repositories, analyst briefings, field reports. On the other side are the decisions: whether to hedge a supply chain position, how to stage a market entry, how to brief a board on regulatory exposure. In the middle, where a structured probability estimate should be, there is usually a gap. Someone reads the sources, forms a judgment, and communicates it as a verbal assessment: "elevated," "uncertain," "watching closely."
We started Cade Market because we thought that gap was worth closing in a specific way. Not by adding more information sources or better visualization tools, but by building a system that produces a number: the probability that a specific event occurs within a defined timeframe, with the confidence interval and source attribution attached.
The Problem We Are Solving
The problem is not information scarcity. Most risk analysts who work on political and economic events have access to more sources than they can effectively use. The problem is that none of those sources, individually or collectively, produce the analytic output the analyst actually needs: a calibrated probability estimate.
A calibrated probability estimate has three properties. It is attached to a specific event with a binary resolution condition: either the event happens or it does not, within a defined timeframe. It comes with a confidence interval that represents genuine uncertainty rather than false precision. And it is documented: the sources that contributed to the score are named and weighted, so the analyst can understand why the number is what it is and evaluate whether to trust it.
Most existing tools produce something different. They produce information volume (more sources, more coverage), risk ratings (high/medium/low on a regional or sector basis), or narrative briefings (what happened and what it might mean). These outputs have real value for context and orientation. They do not, by themselves, produce the number the analyst needs when they are deciding whether to act on a specific risk exposure.
Why We Built a Research Instrument
The language we use internally is "research instrument" rather than "risk platform." The distinction matters to how we think about product decisions. A research instrument is defined by the quality of its measurements: whether the output is accurate, reproducible, and appropriately uncertain. A risk platform is often defined by coverage breadth and feature count. Those are different optimization targets.
Building a research instrument means being deliberate about which sources we include and why. A source that adds coverage volume without improving calibration makes the instrument less reliable, not more comprehensive. This means saying no to source categories that would be easy to add but have not demonstrated predictive accuracy on the event types we are scoring. It means defining events carefully so that resolution conditions are clear and outcomes can be tracked against predictions. And it means representing uncertainty honestly: a wide confidence interval on a genuinely uncertain event is a correct output, not a failure state.
These constraints are not comfortable. They mean the product covers a narrower range of events than an information aggregator would. They mean the output looks simpler than a dashboard with many data visualization layers. But the discipline is what allows us to claim that the probability estimates mean something: they are calibrated measurements, not summarized narratives.
Our Source Architecture
We aggregate across four source categories: news wires, government documents, academic preprints, and field intelligence. Each category has different latency, reliability, and calibration properties across event types.
News wires are the most time-sensitive and most noisy. They carry high recency weight for fast-moving events, but wire coverage amplification means that multiple signals can derive from a single underlying data point. The scoring system has to filter for genuine new information rather than recirculated coverage.
Government documents are the most structurally informative for regulatory and legislative events. A formal consultation period opening, a budget line item, a legislative committee schedule: these documents constrain event probability in ways that news coverage does not. They represent institutional commitments to process timelines that wire reports can describe but cannot replace.
Academic preprints have the longest lead time of any source category. A research paper on the policy case for a regulatory change, published months before formal action, represents a kind of directional intelligence that other categories do not provide. We weight this category conservatively, because the path from academic recommendation to regulatory action is long and uncertain, but the directional signal is real.
Field intelligence is the most expensive category to operate and the most valuable for events where public-source environments are thin. We are selective about what qualifies as field intelligence and how we incorporate it into the scoring model. The quality verification burden is higher than for other categories.
What Calibration Actually Means
Calibration is a technical term that is worth defining precisely, because it is the core claim of what we are building. A probability scoring system is calibrated if, across a large number of resolved events, the events it scored at 70% occurred approximately 70% of the time, the events it scored at 40% occurred approximately 40% of the time, and so on across the probability range. A well-calibrated system is not necessarily a system that correctly predicted every event; it is a system whose confidence levels accurately reflected actual outcome frequencies.
Calibration is measured by tracking resolution outcomes against prior probability estimates. This requires that events actually resolve: that the timeframe closes and the binary outcome is observed. This is one reason why event definition discipline matters so much for calibration tracking. Events with vague resolution conditions or indefinite timeframes cannot contribute to calibration history. They effectively hide information about system performance.
We are in early stages of building out our resolution history. We are being deliberate about this: the calibration claim requires data, and we will report calibration performance as our resolution dataset grows rather than make claims we cannot yet support. What we can say is that the scoring architecture is designed for calibration from the start: source weights update on resolution, event definitions are held to a standard that permits tracking, and the confidence intervals represent genuine uncertainty about source-environment quality.
What We Are Not Trying to Do
We are not building a geopolitical news service. We are not building a general-purpose risk monitoring platform that covers every industry, region, and event type. We are not trying to replace the judgment of the analyst who works with the output.
The Cade Market output is an analytic input: a structured probability estimate with uncertainty and source attribution. The analyst decides what to do with it. They bring contextual knowledge about their organization's specific exposure, their confidence in particular source categories, and their judgment about which events are relevant to their situation. The probability estimate makes that judgment more precise. It does not replace it.
If you are interested in what we are building, you can find more detail on the product page. We are working with a small set of analysts to develop the event coverage and calibrate the scoring architecture. This is the work we are doing, and this blog is where we will document the methodology behind it.
View all our analysis on the blog index.