Cognitive Causal Fingerprinting of Sedimentary-Basin Brines: A Calibrated, Refusal-Capable Swarm Intelligence Applied to 113,135 US Produced-Water Analyses

Produced water — the brine co-produced with oil and gas — is the most abundant waste stream in the petroleum industry and, increasingly, a resource: a host for lithium and other critical elements, a forensic tracer of subsurface fluids, and a diagnostic of reservoir condition. Its chemistry is high-dimensional, sparsely sampled, and strongly basin-dependent, which frustrates conventional predictive modelling. We introduce the Neuro-CogniSwarm (NCS) Cognitive Engine, a machine-intelligence system architected within the AGI Monolith Brain framework, built around three properties that standard regressors and classifiers lack: it learns the causal structure among dissolved species rather than mere correlations, it reports calibrated confidence whose magnitude tracks its true error, and it refuses to answer for samples that lie outside the chemistry it was trained on — the “knows what it doesn’t know” property that makes it deployable where confident errors are dangerous. Applied to the US Geological Survey National Produced Waters Geochemical Database (113,135 samples, 71 basins), the NCS system yields one principal scientific result and three applied capabilities. The principal result is that each sedimentary basin carries a distinctive causal fingerprint — the graph of conditional dependencies among major ions — with two links (Mg–Ca, Cl–Na) common to all basins and the remainder basin-specific; fingerprints are stable within a basin (Jaccard 0.80–1.00) and distinct between basins (0.20–0.58), and a basin can be identified from its fingerprint alone with 100% accuracy (chance 16%). The applied capabilities are calibrated basin provenance (78% accuracy, expected calibration error 0.061, rising to 99% on high-confidence cases), lithium prospectivity screening from common ions (area under the ROC curve 0.93), and chemical geothermometry (reservoir-temperature error ±15 °C).We argue that for high-stakes subsurface decisions the system’s willingness to decline — its awareness of the limits of its own competence — is as valuable as its accuracy.