To make better exploration tools, we first need to better understand ore deposits

In an era of industry-funded research, there is an imperative that research yields outcomes that can be applied directly, at the expense of improving fundamental understanding. Reality often proves more complex than our simplistic experiments anticipated. A summary of common experimental behaviour might be framed thus:

1. observe something of possible use
2. collect limited orientation samples
3. become enthused by early results
4. roll out the tool to genuine unknowns
5. generate abundant data with ambiguous results
6. analyse using a machine learning algorithm hoping for a miracle.

This paper documents some of our recent trials and errors, mostly through application of SEM-based Tescan Integrated Mineral Analyser (TIMA) quantitative mineralogy and rock mapping, and attempts to apply mineral chemistry and chronology analysis to detrital sediment samples. We've learned that to improve our new tools we must:

1. expand or customise our characterisation of natural background
2. observe mineral appearance, assemblage and morphology before microanalysis
3. accept that detailed observation may demand revisions to existing models of hydrothermal mineral stabilities, and hence provoke more fundamental research
4. accept that machine learning classifications tend to be successful only when trained against local or specifically relevant orientation data.