offdiagonal / replay forecasts against reality

Relationships

Two networks, held strictly apart. The Earth network describes pathways through physical reality. The model network describes shared evidence and historical co-failure between predictive systems — it never claims one model causes another to fail.

Earth relationship network

Every edge carries its evidential class; nothing is drawn as causal unless it is physically known

DERIVED
heat → vpd — physically knownvpd → veg — physically knownsoil → veg — physically knownveg → fire — physically knownfire → smoke — physically knownsmoke → aq — physically knownheat → uhi — physically knownuhi → aq — statistical candidatesst → surge — statistical candidatesst → heat — statistical candidatesurge → human — context onlyquake → human — context onlyaq → human — context onlyheat → aq — persisted compound risksurface heat — URBANsurface heatURBANhigh VPD — AGRICULTUREhigh VPDAGRICULTUREvegetation stress — AGRICULTUREvegetation stressAGRICULTUREsoil moisture deficit — AGRICULTUREsoil moisture deficitAGRICULTUREfire susceptibility — AGRICULTUREfire susceptibilityAGRICULTUREsmoke transport — AIRsmoke transportAIRair-quality degradation — AIRair-quality degradationAIRurban heat island — URBANurban heat islandURBANSST anomaly — SEASST anomalySEAstorm surge — SEAstorm surgeSEAseismic sequence — EARTHseismic sequenceEARTHhumanitarian response — GLOBAL CONTEXThumanitarian responseGLOBAL CONTEXT
physically knownpersisted compound riskstatistical candidatecontext only

model families

33

dependency edges

6

shared evidence, not causality

independence groups

30

genuinely independent

0

no declared shared provider or group

Model dependency network

An edge means these families read the same evidence. Two agreeing models sharing an edge are not two confirmations.

DERIVED

4 families carry at least one dependency edge. Red edges are shared independence groups — the strongest form of non-independence: agreement between them adds far less information than it appears to.