Methodology
HOW OUR ANALYSES ARE GENERATED
Every assessment on this site is generated by AI following a structured intelligence analysis methodology. No human analysts are involved in the production process. The system combines large-language-model reasoning with tradecraft techniques used by professional intelligence organizations.
Analysis Pipeline
Every full assessment passes through six sequential phases. Collection must complete before analysis begins; the red team reviews only after domain analysis is finished.
Collection
Systematic gathering from multiple open sources. Facts, timelines, and source reliability are documented before any interpretation begins.
Hypotheses
Four or more competing explanations are generated to avoid anchoring on a single narrative.
Domain Analysis
Eight specialist analysts examine the evidence in parallel — each through their own lens, each unaware of the others' conclusions.
Structured Analysis
Key Assumptions Check and Indicators & Warnings are applied. Hypotheses are evaluated against evidence using structured techniques.
Red Team
An adversarial review challenges every conclusion. A pre-mortem asks: if this assessment is wrong in six months, why?
Synthesis
Findings are integrated with calibrated confidence levels and likelihood language. Dissenting views are preserved, not suppressed.
Specialist Agents
During Phase 3, eight AI-powered domain specialists analyze the evidence in parallel. Each operates independently to prevent groupthink.
Force postures, capabilities, and movements
Domestic politics, coalitions, and power dynamics
Sanctions, trade flows, and financial leverage
Decoding messages embedded in actions and statements
Diplomatic talks, venue choices, and format decisions
Leader decision-making patterns and risk tolerance
Historical precedents and long-term pattern recognition
Empathetic analysis from each actor's viewpoint
Analytical Techniques
Key Assumptions Check
Identifies and stress-tests the assumptions underpinning each assessment. If an assumption breaks, the conclusion may need revision.
Indicators & Warnings
Defines observable signposts that would confirm or disconfirm each hypothesis — creating a tripwire system for ongoing monitoring.
Analysis of Competing Hypotheses
Evaluates evidence against multiple hypotheses simultaneously, focusing on disconfirming rather than confirming evidence.
Devil's Advocacy
A dedicated adversarial role argues against the prevailing assessment to expose blind spots and groupthink.
Pre-Mortem Analysis
Assumes the assessment is wrong six months later, then works backward to identify what was missed or misjudged.
Estimative Standards
Confidence and likelihood are independent dimensions. You can have high confidence in an unlikely outcome — strong evidence that something probably won't happen.
Likelihood Language
| Term | Probability |
|---|---|
| Almost certainly | >95% |
| Highly likely | 80–95% |
| Likely / Probably | 55–80% |
| Roughly even chance | 45–55% |
| Unlikely | 20–45% |
| Highly unlikely | 5–20% |
| Remote | <5% |
Confidence Levels
Multiple independent sources, consistent patterns, strong analytical basis
Limited sources, some contradictions, plausible but not certain
Single source, significant unknowns, largely inferential
Source Transparency
OSINT Only
All assessments are based exclusively on open-source intelligence. No classified or privileged sources are used. This means our analyses may miss information available to government agencies or organizations with field presence.
Source Reliability Ratings
Sources are graded on a five-tier scale (A–E) assessing past reliability, editorial standards, and potential bias. These ratings inform how heavily each source weighs in the final assessment.
Dissent Preservation
When domain analysts disagree, both positions are presented rather than forcing artificial consensus. Minority views have historically proven correct — suppressing them degrades analytical quality.