Upstream · a question answered by arithmetic, not regression

Explosives don’t travel

Blasting sits further upstream than anything else this atlas tracks, so a bottleneck there would sit above every mined material at once. Whether trade data can see it turns out to be settled by one number. World trade in prepared explosives was 445 kilotonnes in 2024 — less than the consumption ceiling of Chile’s copper mines alone. Explosives are mixed on site from ammonium nitrate and diesel, or made at the mine gate; what crosses a border is the feedstock, not the explosive. So the customs data cannot measure blasting, and the correlation this page originally reported was built on 0.4% of the relevant flow.

Could imported explosives even be the input?

This is the test that should come before any regression, and it is arithmetic. The intensity is calibrated on the USGS survey rather than an assumed engineering constant: US metal mining of every kind consumed 155 kt of explosives in 2019, and US copper mine output was 1,260 kt contained. Attributing every kilogram to copper — generous, since the same explosives blasted gold and iron — gives at most 0.123 kt of explosive per kt of contained copper. That makes the resulting figure a ceiling, so a country falling far below it fails conclusively rather than arguably.

Country	copper 2024	explosives ceiling	prepared explosives imported	share of ceiling
Peru	2,736 kt	337 kt	117.3 kt	34.8%
Philippines	64 kt	8 kt	1.4 kt	17.3%
Kazakhstan	567 kt	70 kt	3.2 kt	4.6%
Zambia	821 kt	101 kt	3.7 kt	3.6%
Indonesia	1,029 kt	127 kt	3.0 kt	2.4%
Mexico	717 kt	88 kt	1.6 kt	1.9%
DR Congo	3,158 kt	388 kt	4.6 kt	1.2%
Chile	5,506 kt	677 kt	2.9 kt	0.4%
Mongolia	352 kt	43 kt	0.1 kt	0.2%
Brazil	379 kt	47 kt	0.0 kt	0.0%

Eight of ten import under 5% of what their mines must consume. The explosives are made where they are used. That is not a data problem to work around; it is the physical fact that ANFO is two commodity inputs mixed in a truck at the pit.

What that does to the correlation

The original version of this page reported that detonator imports track copper output in Chile, and that Chile was the one country whose confidence interval excluded zero. Both statements are still arithmetically true. Set beside the coverage test they invert.

Country	imports as share of ceiling	year-on-year r	95% interval
Peru	34.8%	+0.14	[-0.30, +0.53]
Philippines	17.3%	-0.01	[-0.43, +0.41]
Kazakhstan	4.6%	+0.32	[-0.12, +0.65]
Zambia	3.6%	+0.24	[-0.21, +0.60]
Indonesia	2.4%	-0.25	[-0.61, +0.19]
Mexico	1.9%	-0.12	[-0.51, +0.32]
DR Congo	1.2%	+0.01	[-0.41, +0.43]
Chile	0.4%	+0.71	[+0.42, +0.87]
Mongolia	0.2%	-0.14	[-0.53, +0.30]
Brazil	0.0%	-0.05	[-0.46, +0.38]

The country with the signal has almost no imports; the country with the imports has no signal. Chile buys 0.4% of its prepared explosives abroad and shows r = +0.71. Peru buys about a third of its requirement abroad — the one case where the proxy is defensible — and shows +0.14, an interval straddling zero. A correlation computed on a 0.4% residual is measuring import substitution, not blasting.

What explosives are actually used for

The one piece of this that stands on its own. United States, 2019, industrial explosives and blasting agents sold for consumption, thousand metric tons. The United States is the only country that surveys the end-use split, which is also why it cannot be turned into a global statement.

Use	kt	share
Coal mining	961	55.5%
Construction work	306	17.7%
Quarrying and nonmetal mining	254	14.7%
Metal mining	155	9.0%
All other purposes	53	3.1%

Mining of all kinds is 79.2%, and within it coal is the bulk. That supports the first half of the original argument. It does not license extending it: the United States is an unusually coal-heavy mining economy, and nothing here measures the mix anywhere else. The table also cannot settle the civil-military split, because military explosives are a separate survey and are not in it. Detonators sold for mining and quarrying fell from 37.9 to 28.9 million units between 2015 and 2019 while oil-and-gas detonators rose from 1.7 to 4.8 million.

What was withdrawn, and why

This page was published on 12 September 2026, reviewed adversarially by two independent language models run separately on the same brief, and rewritten twice. The first review found the claims below unsupported. The second was asked to improve the design rather than dismiss it, and its first instruction — test whether imports could be the input before estimating anything — produced the arithmetic at the top of this page, which is a better answer than the regression it replaced.

Withdrawn: “Explosives are a coal business (as a general claim)” — the end-use split is measured only in the United States, which is an unusually coal-heavy mining economy; nothing was shown for any other country.
Withdrawn: “correlations are on tonnage, so price inflation cannot create the result” — HS 3603 is "detonators, safety fuses and detonating fuses" - detonating cord is sold by the metre and dominates the traded mass, so tonnes track product mix rather than blast count.
Withdrawn: “explosives coincide with mining rather than lead it” — the Fisher intervals at lag 0 and lag -1 overlap, and a lead of three to nine months would appear at lag 0 in annual data anyway, so annual resolution cannot distinguish the two.
Withdrawn: “the post's central claim is not supported” — the claim is about explosives PRICES; this page measures import tonnage of one HS code against copper output, and never tested a price at all.

The deepest error was never a statistical one. The original argument concerns explosives prices leading a supply surge; this page measured annual import tonnage of one customs code. It was not a test of that claim, and it should not have been presented as one.

What a real test would need

Consumption, not trade. Domestic production plus imports minus exports. Without it the only observable countries are the ones that barely import, and they are not representative.
Ammonium nitrate split by grade. Blasting-grade and fertiliser-grade share one customs code, and 8.5 Mt of it trades annually against 0.44 Mt of prepared explosives.
Unit counts, not tonnes. Customs data reports mass, and in the detonator code the mass is dominated by detonating cord rather than by the caps that correspond to blast holes.
Prices. Nothing here touches one. Bulk blasting agents are ammonium nitrate, so their price is mostly an ammonia and gas price — a strong argument in itself against reading explosives prices as a clean mining signal.

Sources. End use, detonator counts and the intensity calibration: USGS Minerals Yearbook 2019, Explosives, by Lori E. Apodaca, published March 2024, tables 2 and 3; survey collected by the Institute of Makers of Explosives; PDF archived in the repository. Trade: CEPII BACI (Etalab Open Licence 2.0). Copper mine production: BGS World Mineral Statistics, mine stage only — the cube holds three organisations' estimates of the same quantity and summing them overstates Chile by half. Built by build_explosives.py; every figure, including the withdrawn claims, is in out/explosives.json.