Validation purpose
Validation is central to platform development. It compares earlier classifications with independent evidence produced later and tests whether prioritization improves investigation efficiency.
A target discarded later is not necessarily a failure. Mineral exploration progressively reduces uncertainty; rejecting a weak hypothesis early also preserves resources for more promising investigations.
Planned methods
- Backtesting on known regions, deposits, and occurrences.
- Benchmarking against reference methodologies and models.
- Holdout and blind testing with independent datasets.
- Analysis of how known deposits and occurrences are classified by the system.
- Field validation comparing model outputs with evidence obtained later.
Performance metrics
Once the historical base is sufficient, performance can be tracked with technical and operational metrics.
- Precision, recall, false-positive rate, and false-negative rate.
- Concentration of relevant occurrences among the highest-scoring targets.
- Performance changes across model versions.
- Time to qualified target, cost per target, and cost per structured project.
- Progression rate between pipeline stages.
Learning loop
Proprietary data, field results, diligence, evaluated projects, discarded areas, and decision history form a verifiable base. Each new analysis should help make models more precise, consistent, and efficient.
The GEOAI Validation Report should evolve with the platform, recording versions, assumptions, limitations, tests, errors, and material model changes.
Evolution and intellectual property
Some information may come from public, scientific, licensed, or commercial sources. The proprietary asset grows in how those inputs are selected, treated, related, and enriched by BTR's own results and decisions.
The system should evolve from a base supported by external information toward infrastructure progressively enriched by proprietary data, decision history, and validation outcomes.