.jpeg)
Causal Intelligence · Near-Horizon Programme
Prediction says what may happen.
Causality asks what changes it.
Pythology is developing a causal reasoning layer for systems where decisions depend on mechanisms, downstream consequences and competing interventions—not correlation alone.
Core distinction
Correlation, prediction and causation are not interchangeable.
A model may forecast an outcome accurately while remaining unable to explain whether changing one variable would alter that outcome.
What varies together?
Useful for finding relationships, but vulnerable to confounding, selection effects and hidden common causes.
What is likely next?
Optimises forecast performance, but does not automatically identify a safe or effective intervention.
What would change the outcome?
Requires explicit assumptions, mechanisms, interventions, counterfactuals and tests capable of proving the explanation wrong.
MODEL
Causal reasoning architecture
From observed state to accountable intervention.
The planned system carries evidence, proposed mechanisms, uncertainty and competing hypotheses through every stage.
Observe
Collect events, measurements, spatial context and system state.
Represent
Construct variables, relationships, time order and known constraints.
Compare explanations
Evaluate multiple plausible causal structures rather than committing to the first correlation.
Simulate interventions
Estimate outcomes under alternative actions, delays or unavailable interventions.
Expose limitations
Show assumptions, uncertainty, missing variables and evidence that would falsify the model.
EarthNet integration
The layer that turns monitoring into decision intelligence.
EarthNet currently assembles events across environmental and infrastructure domains. The causal layer is intended to connect those events into mechanisms and consequences.
Emerging risk
Identify changes across atmosphere, ocean, land, infrastructure and communities.
Cross-domain pathways
Represent how one environmental or infrastructure condition may influence another.
Consequences
Estimate exposure, delay, cascade and which uncertainties matter most.
Alternative actions
Compare warning, closure, evacuation, barrier, resource and timing strategies.
Human accountability
Keep the evidence path visible to operators, scientists and decision-makers.
Post-event revision
Update assumptions when reality disagrees with the model.
First proving ground
Project Hyggja / Awa.
Flood response provides measurable hazards, agents, interventions and consequences. That makes it a controlled environment for testing whether causal and counterfactual reasoning improves decisions.
Required output
Every inference must be inspectable.
A causal score without its assumptions is not sufficient for high-consequence use.
Observed support
Measurements, events and known relationships supporting the pathway.
Proposed mechanism
The causal chain the system believes may connect state to outcome.
Competing explanations
Other plausible structures and the evidence distinguishing them.
Limits and falsification
Missing variables, extrapolation, sensitivity and observations that would overturn the conclusion.
Research foundations
Causal inference and representation learning.
These sources establish research foundations only and do not imply endorsement of Pythology.
Causal Inference: What If
Open textbook on causal questions, interventions and observational evidence.
Open source →Toward Causal Representation Learning
Research agenda connecting representation learning with causal structure.
Open paper →Elements of Causal Inference
Foundations and algorithms for causal discovery and inference.
Open source →Research and deployment collaboration
Causal systems require domain evidence.
Pythology is interested in discussions with organisations holding well-defined intervention histories, environmental observations, infrastructure data or simulation environments suitable for causal evaluation.