The Pythology stack
The machinery underneath Pythology can get complicated. The job of each layer should not. Every part exists to answer one question, preserve one boundary, or stop the system from becoming too sure of itself.
How the pieces fit
A sensor reading should not quietly become a causal claim. A causal story should not quietly become a forecast. A forecast should not quietly become an instruction. Keeping those steps separate is the point.
EarthNet brings the outside world in: hazards, environmental conditions, infrastructure pressure, official updates and other authorised evidence. Its job is to notice meaningful change without pretending it already knows what the change means.
Observe broadly, preserve the source and keep repeated noise from looking like new evidence.
Explore EarthNet →Observations are timestamped, attributed and organised into a current state. Stale information, contradictions, missing evidence and modelled estimates stay visible rather than being flattened into one confident answer.
Before asking why, separate observation from inference and freshness from certainty.
The causal layer asks what mechanisms could connect the evidence, what else could explain it, which assumptions are doing the work and what evidence would weaken the preferred explanation.
Correlation is useful. We still want to know what could actually make one part of the system affect another.
Go deeper on causality →Prometheus turns the current belief into a test. He commits a time-bounded forecast before the outcome is known, records confidence and a competing outcome, then returns later and keeps the result in the ledger.
If we think we understand the system, say what reality should do next — and keep the receipts.
See the public forecast record →Decision Futures starts from the same frozen evidence and asks how downstream consequences could differ if an authorised human changes a decision, intervention or assumption. The branches remain conditional, not prophecy.
Compare the baseline with a small number of declared alternatives and keep the assumptions visible.
Sentinel is the human interrogation layer. It is where an authorised person can ask what the system sees, why it believes something, what is degraded, what it does not trust and what new evidence could change the conclusion.
Complex machinery should not earn the right to become a black box simply because it is complicated.
Later evidence closes the loop. Forecasts resolve, enacted decisions can be compared with measured outcomes, misses stay in the history and confidence can be adjusted without rewriting what the system originally believed.
Reality gets the final vote. The useful asset is the record that accumulates afterward.
The rules underneath it
Models will improve. Sources will change. Some layers will become more sophisticated. These are the boundaries we want to survive all of that.
Measured, reported, modelled and inferred information should never silently become the same thing.
A preferred explanation should have to outperform plausible competitors instead of merely sounding convincing.
Missing evidence, disagreement and weak assumptions are information, not blemishes to tidy away.
Forecasts and claims earn confidence through later outcomes, not through how persuasive the original explanation was.
Cross-domain proof
We are testing the same reasoning discipline in very different systems. That does not mean confidence transfers between domains. Every domain has to earn its own evidence model, mechanisms, validation and track record.
EarthNet is the most operational expression of the stack today: live global observation, New Zealand daily intelligence, guarded hazard-change signals and Prometheus forecasts that can later be scored.
See EarthNet → RESEARCH ARCHITECTUREBiological work applies the same insistence on provenance, mechanisms, competing explanations, experiments and outcome feedback across living systems. The evidence world is completely different, so validation has to be earned separately.
Explore biological work → RESEARCH ARCHITECTUREEngineered systems add a hard constraint: physics gets a veto. Geometry, materials, forces and measured field behaviour place boundaries around what the intelligence is allowed to claim.
Explore physical work →What transfers — and what does not
This distinction matters. The architecture can travel from one domain to another. The confidence, evidence and validation cannot hitch a ride with it.
The point of the stack
The goal is not to hide complexity. It is to organise it so a person can still ask: what did you observe, why do you believe that, what else could explain it, what do you expect next, and what happened when reality answered?
Research, investment or deployment
We are interested in real environments where the stack can be tested prospectively: clear evidence, meaningful decisions, measurable outcomes and people willing to challenge the conclusions.
.jpeg)