ANM CFEM royalty collection (Brazil)

Quirks

• ·SOLD, not produced - the quantity is what was commercialised that month, so stockpiling and drawdown both distort it
• ·the CNPJ is the royalty PAYER for that title-month, which need not be the current title holder; attribution follows the CNPJ the source names, never our own title-to-holder mapping
• ·units never combine: t, m3, l, g, kg, m2 and ct all appear, so the unit is part of the aggregation key
• ·99.9% of the naive tonne total is corruption - summing the file gives 134 TRILLION tonnes against ~100 billion for world production of everything
• ·the corruption is a QUANTITY typo against a CORRECT payment, so royalty-per-tonne separates it: real rows run R$0.14-8.13/t, corrupt rows R$4e-11 to R$8e-9/t
• ·a tonnage ceiling alone cannot work - the same quantity is legitimate for iron ore (Brazil ~400 Mt/yr) and impossible for phosphate (~5 Mt/yr)
• ·13,198 rows are verbatim duplicates of another row, timestamp included; the first three data rows of the file are three copies of one
• ·emitted under measure `royalty_assessed_sales`, deliberately NOT `production` - only production and reserves are read by the dossier supply surfaces