build_stockpiles.py → stockpiles.json.Who quantifies a buffer, and who doesn’t
The 32 materials, by published cover
Which of the atlas’s materials have a published per-material days-of-cover (Japan/JOGMEC), and which don’t — either held on an undisclosed basis (US NDS) or with no reserve at all.
Why this is the quiet decision
Every other lever the atlas measures — build an alternative supplier, substitute the material, recycle it — takes years. A stockpile is the only one that works on the timescale of an actual shock: it buys the months or years the others need. That is exactly why its absence matters. For the geological chokepoints (phosphate, the platinum-group metals, niobium) there is no faster option — you cannot out-build the rock — so the stockpile is not a backstop but the front line, and for most of them no country publishes how large a buffer it holds. Japan drew the right lesson from its 2010 rare-earth shock and built — and publishes — a 60-to-180-day buffer; the US holds ~28 commodities but its dollar inventory is small (~6% of its own identified shortfall) and it doesn’t disclose per-material cover; the EU has a plan and no reserve; and China, which holds the chokepoints, also holds an undisclosed reserve behind them. One honest qualifier: government reserves aren’t the only cushion — commercial inventories, scrap mobilisation and rationing all absorb part of a shock — but those are private and unplanned, and for the geological chokepoints there is no substitute for a deliberate public buffer. The cheapest insurance in the whole toolkit is the one almost nobody has bought. What else you can do →