Cauldra

Validation on public datasets.

Cauldra has been trained and validated across three independent BC porphyry systems — different terranes, different operators, multiple drill seasons. Over 300 drillholes and 130,000+ metres of logged core. The blind validation below uses publicly disclosed BC ARIS drill data from a North American Cu-Au porphyry project across three consecutive programs (2021, 2022, 2023). Cauldra was trained on the 2021 and 2022 logging records and ran on the 2023 program blind. The model's outputs match the geological controls documented in the operator's published assessment reports — produced from logging alone, without access to the operator's interpretation.

The model recovers the porphyry zoning framework from logging alone — productive intrusive hosts, barren post-mineral dykes, sulphide paragenesis, and alteration zoning.

1. Porphyritic diorite phases as the primary mineralised hosts. Five intrusive lithology codes appear in the top seven predictive features — matching the productive hosts described in the operator's assessment reports.
2. Post-mineral monzonite dykes as negative grade contributors. The paragenetic distinction between mineralised pre-mineral diorite and barren late monzonite intrusions emerges from logging alone — the same separation documented in the published interpretation.
3. Sulphide paragenesis in the documented order. Chalcopyrite as the strongest positive sulphide signal, followed by bornite — matching the operator's stated abundance for grade-relevant sulphides.
4. Potassic and phyllic alteration as the productive assemblages. Both rank in the top fifteen predictive features. Propylitic alteration carries no predictive weight — as expected for the distal halo that surrounds but does not define the ore zone.

The geological controls on grade are learnable from standard core logging data — independently of the operator's interpretation. For an active program, that means a calibrated read on grade as logging comes in.