Executive summary (20-second highlights):

Copper's widest cash premium in five years collapsed by 89% in four sessions. We pulled the London Metal Exchange's daily series to work out how much metal that really took to move. The answer is not close to what the price action implied.
The deferred copper price rose in the same four sessions the spread collapsed. If the market had decided copper was suddenly more abundant, both prices should have moved the same way. They didn't.
China's July magnet tonnage barely moved, and one G7 destination's share fell by half. We pulled the customs series ourselves, one line beneath the one everyone reads, and it tells a very different story about who's actually being throttled.
The largest Japanese magnet maker signed off on a United States terbium oxide with no further validation required. It reads like exactly the diversification story everyone wants. Two details buried in the same week's filings complicate that considerably.
A county environmental bureau walked onto the site of the world's largest lepidolite mine and found no ore moving, seven weeks after the market booked the restart. The document that finally moved the file forward tells you exactly how much smaller this mine is coming back.
Kazakhstan's state uranium producer convened a shareholder vote on selling into a Russian enrichment plant and to a Chinese state buyer, terms withheld at the buyers' request. What happens to that uranium between the mine and the reactor is a question this filing never answers.
The United States government finished capitalising the vehicle that has locked up fifteen years of output from the only non-Chinese mine producing all four magnetic rare earths. Three days earlier, the contract's funding condition was quietly rewritten to match what the government had actually delivered.

Illustration of the London Metal Exchange headquarters

The spread that a warehouse decision erased

Let’s start with what the copper market told you, because most of it was true.

London Metal Exchange (LME) warehouse stock fell for 42 consecutive sessions into August 14th, the longest downward run since 2014, bottoming at 204,975 tonnes.
Chile cut its national output forecast for a second straight quarter.
Indonesia's Gresik smelter stayed offline after a boiler leak.
The Democratic Republic of the Congo had banned concentrate exports on August 6.
Cash copper set a record above US$14,900/t on August 19th, and the exchange introduced emergency measures on August 14th to contain the front end.

On our own calculation from the LME official daily series, the cash-to-three-month spread settled at US$535/t on Monday, August 17th, against US$34/t on July 31. Intraday, it ran wider still. Backwardation of that size is the market saying it will pay almost 4% of the value of a tonne of copper for the privilege of having it today rather than in ninety days.

Zambia, Africa's second-largest copper producer, declared Hakainde Hichilema president-elect on August 18th, removing an election discount from the one African jurisdiction with a stated path to 3 Mt/yr by 2031.

Every input to that story is a supply input. Mines, smelters, export bans, stock levels, and even an election. The reader who concluded that the world had run short of copper was reading the evidence in front of him, and he had plenty of company.

One thing about that setup before going further: every element of it describes the global stock of copper. The spread describes something narrower, though: the cost of getting a tonne on a specific day (or timeframe).

Those two are usually close enough that nobody separates them, and the whole of last week is an argument for separating them (this is why we are presenting specific dates and events).

What you’ll see on the table below is that suddenly, over four sessions, the spread fell to US$56/t. An 89% collapse.

The obvious explanation is that the metal came back, and the obvious explanation is where most coverage of the week stopped. The LME publishes enough detail to test it, and the test does not come out the way the price action suggests.

This matters beyond copper, which is why it opens the edition rather than sitting in a market section.

What is curious enough is that a similar thing was reported on CMJ’s last briefing: different critical minerals swinging in the same direction, almost as if they were performing a choreography.

Three other minerals moved along with copper, on the same mechanism, and in each case what changed was the administrative status of the material rather than its quantity:

A Chinese customs line whose total barely moved while one destination's heavy rare earth allocation sat at zero
An American oxide that became sellable because a Japanese customer signed a specification
And a Chinese lithium mine the market has counted as running since June, which a county inspector found idle in August

Copper is the cleanest case because the exchange publishes the evidence daily and nobody has to take our word for it.

We flagged the test in CW33. When the spread hit US$434/t on August 14th, we wrote that the August 19th prompt date would settle how much of the backwardation was positioning and how much was genuine scarcity of metal. The prompt date has passed, so the answer exists.

Getting at it required pulling the data rather than reading the coverage.

Wire reports give a spread print and a stock headline. What decides the question is the relationship between three published series across the same nine sessions: cash settlement, three-month settlement, and total warehouse stock. We pulled all three from the exchange's daily reports and calculated the spread ourselves for every session across the move.

Two figures settle it: how much copper it took to erase a five-year record spread, and where that copper came from, which turns out to be a question about paperwork rather than about freight.

There is also a third number that moved in the opposite direction to everything above, in the same four sessions, and almost nobody wrote it down: the deferred price went up while the spread collapsed.

Here is what that combination means, and here is the arithmetic on how little metal it took: