.jpeg)
What moves together?
Useful for finding patterns and candidate relationships, but vulnerable to confounding, selection effects and common causes.
Deeper explainer · causal intelligence
A convincing story
is not the same as a cause.
Pythology keeps evidence, mechanism, prediction and counterfactual scenarios separate on purpose. The aim is not to make every relationship causal. It is to make important explanations inspectable enough to challenge and test.
If we think we know why something is happening, the explanation should survive competing hypotheses — and eventually make predictions that reality can judge.
Pythology-built infrastructure · Causal Rust Core
A deterministic engine
for reasoning that can be inspected.
Pythology's founder built the Causal Rust Core to turn declared causal mechanisms into versioned, reproducible computation. It keeps the engine beneath the intelligence products separate from the evidence and presentation layers, so its conclusions can be traced, challenged and rerun.
EARTHNET
Test mechanisms against a changing world.
EarthNet uses the core to evaluate explicit causal pathways and intervention branches while preserving the evidence attached to each step.
ATLAS
Carry causal structure into the evidence map.
Atlas integrates the same engine so relationships, assumptions and provenance can remain consistent as evidence is organised and validated.
PROMETHEUS
Turn mechanisms into forecasts that risk being wrong.
Prometheus uses traceable causal outputs to form prospective expectations, commit them before outcomes and learn when reality answers.
Why Rust: the core is compiled, deterministic and versioned. The same inputs and engine version produce the same trace, giving Pythology a dependable foundation for systems that must show their working.
Three different questions
Useful together.
Dangerous when blurred.
A relationship can be correlated without being causal. A forecast can be accurate without identifying the mechanism. Pythology tries to keep those claim types visible rather than letting one quietly become another.
What should happen next?
A prospective expectation that can later be scored. Predictive skill is valuable, but by itself it does not prove the proposed mechanism.
What could make it happen?
A mechanism with assumptions, time order, competing explanations, contradictory evidence and observations capable of weakening it.
How the reasoning moves
No silent hand-offs.
The useful discipline is not a magical causal model. It is the separation between each step, so a sensor reading does not quietly become a story, and a story does not quietly become a decision.
01 · EVIDENCE
What do we actually know?
Preserve source, timestamp, freshness, contradictions and missing information before interpretation begins.
02 · MECHANISMS
What could connect it?
Represent plausible mechanisms and keep credible alternatives alive long enough for evidence to distinguish between them.
03 · FALSIFICATION
What would make us less confident?
Important beliefs should include the observations or outcomes that would weaken, overturn or redirect the explanation.
04 · PROMETHEUS
If we are right, what happens next?
Commit a time-bounded forecast before the answer is known, with confidence and a competing outcome attached.
05 · OUTCOME
What did reality do?
Resolve against later evidence, preserve the original record and use accumulated outcomes to recalibrate future confidence.
The boundary matters: a causal explanation can remain provisional even when a forecast resolves correctly. Likewise, a forecast miss does not automatically prove every part of the underlying mechanism false. The record is there so those distinctions can be examined rather than rewritten after the fact.
Prometheus
The mechanism has to risk being wrong.
Prometheus is where reasoning becomes prospective. If the current evidence and mechanism imply something should happen, he writes that expectation down before the outcome arrives.
The ledger keeps confirmations and misses together. That makes forecasting more than a persuasive explanation after the event.
Commit. Resolve. Calibrate.
- Commit: freeze the forecast before the outcome is known.
- Resolve: use later independent evidence to determine what occurred.
- Calibrate: compare confidence with observed frequency across many resolved cases.
- Keep misses: the uncomfortable outcomes stay part of the public record.
A forecast ledger is evidence about forecasting performance. It does not, on its own, prove causation.
Decision Futures
What changes if
the choice changes?
Once a baseline is explicit, a genuine human decision can be compared against a small number of conditional alternatives. The unchosen branches remain counterfactuals — they do not become evidence just because the simulation looks detailed.
BASELINE
What if nothing changes?
Start from the same frozen evidence and assumptions used by the baseline forecast.
INTERVENTION
What are we changing?
Record the decision, assumptions and expected downstream effects before the outcome.
REALITY
What did the chosen branch produce?
Only the branch actually enacted can be directly compared with later observed outcomes.
Portable architecture · non-portable certainty
The discipline can travel.
Confidence has to start over.
Planetary, biological and physical systems have different evidence, mechanisms and validation rules. The architecture can be reused; the confidence cannot be inherited.
Planetary intelligence
EarthNet provides live, noisy, independently observable events where causal reasoning and prospective forecasts can be challenged by reality.
Explore EarthNet →Biological intelligence
Biological claims require their own reviewed evidence, mechanisms, experiments, domain experts and outcome validation.
Explore biological research →Physical intelligence
Engineering and physical systems add hard constraints where simulated plausibility must remain subordinate to physics and measured tests.
Explore physical research →Research foundations
We are not inventing causality from scratch.
Pythology's implementation is our own research direction, but the field sits on decades of causal inference work. These are useful foundations, not endorsements of Pythology.
HERNÁN & ROBINS
Causal Inference: What If
Open textbook on causal questions, interventions and observational evidence.
Open source →SCHÖLKOPF ET AL.
Toward Causal Representation Learning
Research agenda connecting representation learning with causal structure.
Open paper →PETERS · JANZING · SCHÖLKOPF
Elements of Causal Inference
Foundations and algorithms for causal discovery and inference.
Open source →Research or deployment collaboration
Bring us a problem
that can be tested.
The best fit is a problem with well-defined observations, outcomes, intervention history or a prospective environment where the reasoning can be challenged rather than simply admired.
No sensitive operational, patient or restricted research data should be submitted through the public site. Start with the shape of the problem and the kind of evidence available.
Pythology's causal intelligence work is research in live validation. Causal claims remain provisional until supported by appropriate domain evidence, assumptions and validation.