Is the cut leaking? Cobalt after the DRC's 2025 export ban

Trade-flow companion to the price wedge (R72). Every prior case in this corpus tracks a consumer-side control — China, the dominant refiner, weaponising an export licence or ban against a buyer (the US/EU), and the question is whether banned material re-enters the buyer through a laundered origin. Cobalt is the first *producer-side* case, and it inverts the whole geometry. Here the dominant producer — the DRC, ~73% of world mined cobalt — restrains its own exports to rescue a collapsed price. The circumvention question flips with it:

> consumer-side: did a laundered origin appear at the buyer to defeat the ban? > producer-side: did the restrained producer's material leak past the export cut under a neighbour's flag?

This is a DUAL-SCORE / alternative-track signal — never folded into any Tier-1 exposure score. Research, not investment advice; any origin-relabelling is INFERRED from statistical implausibility plus public records, never asserted as smuggling on any single shipment.

Verdict

On 22 February 2025 the DRC's market regulator ARECOMS (Decision 001/ARECOMS/2025) suspended all cobalt exports — industrial and artisanal — to arrest a multi-year price collapse driven by CMOC's Tenke Fungurume/Kisanfu oversupply. The ban was extended in June, then on 16 October 2025 replaced (Decision 005/ARECOMS/2025) by a hard annual quota: 18,125 t for the Nov–Dec 2025 transition, then 96,600 t/yr for 2026–27 — less than half 2024 output.

The control is unambiguously binding — the price channel proves it: in 2025 cobalt metal more than doubled and cobalt-hydroxide feedstock more than quadrupled (Fastmarkets; Investing News Q3-2025). So unlike a consumer-side ban, the supply-security question is not "is my dependency secretly intact behind a fake origin." It is the producer-cartel's question: will withheld tonnes leak enough to break the DRC's own price target?

Our answer on the flow channel: we cannot see it — and *why* we cannot see it is the finding. The detector needs the buyer's granular import-by-origin series. For cobalt the buyer/refiner is China (~76% of world refined cobalt), and China's monthly country-level Comtrade reporting for the cobalt-intermediate line stops at December 2024 — every ban-window period (Feb 2025 onward) returns empty. DRC and Zambia are chronic non-reporters, so mirror data offers no rescue. The origin channel is dark for exactly the window the detector needs. GATE 0.

What the pre-ban baseline shows (the only granular window China reports)

Before the wall, China's dependence on the DRC is near-total. HS 810520 (cobalt mattes / intermediate products / unwrought cobalt / powders) is the line that carries DRC cobalt hydroxide to Chinese refiners:

Source: UN Comtrade free public preview API (reporter China 156, flow imports M, monthly, HS 810520), committed artifact `data/intelligence/cobalt-circumvention.json`. netWgt is gross hydroxide weight (~30% contained Co), not contained cobalt.

This baseline is what a laundering surge would have to displace: a channel that is ~98% one country. When that country's exports are banned, the DRC line must fall — and if the cut is leaking, something has to appear in its place at the buyer.

The would-be fingerprint — pre-computed, but unobservable

If the origin channel were visible, this is the exact test we would run, and the one origin whose appearance would be a textbook tell:

• Volume-implausibility. Zambia mines only ~96 t of cobalt in 2024 and

1,401 t in *total* across 2019–2024 (Zambia Ministry of Mines & Minerals Development) — roughly 0.04% of world, against the DRC's ~230,000 t. A material "Zambian-origin" cobalt line to China would be volume-implausible by ~1,000×. Rwanda and Tanzania (the other flagged routes) have effectively zero genuine cobalt.

• …but Zambia is not a ~0%-capacity *stranger*. This is what makes cobalt

different from the antimony/Thailand shape. The DRC is landlocked; its cobalt hydroxide already trucks Kolwezi → Kasumbalesa → Ndola (Zambia) → Livingstone → Durban under DRC origin papers (ProjectBlue; New America). The corridor already carries the molecules. So the laundering tell here would be origin-document re-papering of a genuine transit flow, not a phantom exporter materialising from nowhere — harder to detect, because the trucks, the border posts and the port throughput all look normal; only the paperwork changes.

• Common-ownership tell: not traced (N). The quota holders are named

industrials (CMOC, Glencore, ERG); any leakage is expected to run through artisanal / informal channels and corrupt checkpoints (African Business, Nov 2025), not a namesake-shell corporate pipe like antimony's Youngsun → Thai Unipet → Youngsun & Essen chain. No corporate hop to lay out.

Smuggling via Zambia/Rwanda is flagged as an anticipated risk by analysts ("I'm almost sure smuggling will rise") but is not yet quantified in any free dataset — a labelled gap, not a claimed surge.

Why it matters for the buyer (and the risk model)

1. The detector presupposes an observing buyer. The two-layer method quietly assumes the buyer publishes granular import-by-origin data. When the dominant refiner (China) goes dark on Comtrade at the exact moment the control begins, the origin channel is unobservable from every reporter side — a reporter wall, structurally different from the HS-basket pollution wall that blinds gallium/indium (there the code exists but pools unrelated molecules; here the code is clean but nobody reports it). A risk model must then fall back on the price channel (hydroxide 4×) plus producer-country export figures and on-the-ground reporting — the flow-origin cross-check is simply off the table. 2. Producer-side controls change *who owns the risk*. A consumer-side ban threatens the buyer with a hidden intact dependency. A producer-side cut is binding by construction (the price proves it); the residual risk is that leakage undermines the producer's own price floor — so the exposed party is the producer cartel and cobalt-long investors, not the Tier-1 buyer whose feedstock simply got more expensive. 3. When the tell is paperwork on a real corridor, absence of a phantom exporter is not absence of leakage. Because Zambia legitimately transits DRC cobalt, the usual "implausible new origin appears" signature would be muted even if the buyer's data were visible — the surge would hide inside a corridor that already runs near capacity. The correct watch-item is Zambian *origin* attestation vs Zambian *mine* output, not Zambian trade volume per se.

Method & honesty rails

• Trade data: UN Comtrade free public preview, China imports, monthly, HS

810520 (cobalt intermediate), 282200 (oxides/hydroxides), 260500 (ores), 810530 (waste/scrap). China's granular series ends Dec 2024; the artifact records the empty ban-window periods explicitly as the reporter wall.

• Production shares: USGS MCS 2026 cobalt (DRC ~73% / 230,000 t; Indonesia

~14%); Zambia mine output from the Zambia Ministry of Mines (96 t 2024; 1,401 t 2019–2024).

• Alternative-track only: never touches buyerRelativeScore or the base

exposure — it sits beside them, like the China–West price wedge.

• Inference, not accusation: no shipment is asserted to be relabelled; the

Zambian-surge fingerprint is pre-computed as what we would look for, and explicitly marked unobservable in this window. The anticipated-smuggling note is attributed to named public reporting, not asserted as fact.

• Distinct shape (R72): completes the corpus's producer-side flank. Antimony

= leak; bismuth = ally-substitution; tellurium/molybdenum = detector-negative by prior resilience; gallium/indium = HS-basket wall. Cobalt = reporter wall + inverted (producer-side) geometry — the case where the control is provably binding yet the origin channel is dark by construction.