How We Build Predictions
Every prediction follows our D/E/M/A framework — a structured approach to uncertainty that forces us to show our work.
D
Data Quality
What do we actually know?
We assess source reliability, recency, and potential biases. Official statistics, academic research, verified reporting — each weighted by trustworthiness. If the data is weak, we say so.
Example: For economic predictions, we verify current rates via official sources — not training data that may be months old.
E
Epistemology
What does history tell us?
We anchor predictions in base rates from academic research. How often do sanctions cause regime change? What's the survival rate of authoritarian leaders during war? Numbers before narratives.
Example: "75–85% of authoritarian regimes survive sanctions at 5-year horizon" (Escribà-Folch & Wright, 2010)
M
Model
How do the pieces fit together?
We build explicit causal models: what drives each scenario, what blocks it, what are the dependencies. If our reasoning is wrong, you can see exactly where.
Example: Scenario A requires X AND Y. Blocker: Z would prevent it. Historical analogue: similar situation in 1998.
A
Aleatoric
What could flip everything?
We identify specific cascade scenarios — named triggers with chain reactions. Not "something unexpected" but "if X happens, then Y enables Z." Black swans we can name, even if we can't predict.
Example: "If Crimea land bridge cut → 'existential' framing → mobilization wave OR nuclear threat → escalation spiral"
Our Principles
◆
Show confidence honestly.Every prediction includes a confidence score. Low confidence (30%) means high uncertainty — we won't pretend to know what we don't.
◆
Publish calibrations.When we're wrong, we document why and update our methodology. Misses are published alongside hits in our
retrotrack.
◆
Academic sources.Base rates come from peer-reviewed research with DOI links. You can check our homework.
◆
Version tracking.Predictions evolve. We track every update so you can see how our thinking changed.
