Did the Kachin rare-earth cut leak? Myanmar heavy-REE feedstock after the Oct-2024 KIO takeover

Trade-flow companion to the price wedge (R72), producer-side variant — and the *upstream twin* of the rare-earth-magnet case. The magnet case ([remagnet](2025-rare-earth-magnets-licensing-circumvention-transshipment.md)) sits at the finished-good end of the heavy-rare-earth (HREE) chain, where China's Apr-2025 export licence bites. This case sits at the feedstock end, where the raw ion-adsorption-clay (IAC) concentrate that makes China's ~90–95% terbium/dysprosium/holmium dominance possible is actually dug — two townships in Myanmar's Kachin State. This is a DUAL-SCORE / alternative-track signal, never folded into any Tier-1 exposure score. The divergence between "the origin that stopped mining" and "who suddenly started shipping to the buyer" IS the signal. Research, not investment advice; origin-relabelling is INFERRED from statistical implausibility plus public records, never asserted as smuggling on any single shipment.

Verdict — GATE 0 (producer-side; detector-negative on cross-border laundering — land-border clay cannot re-flag — with a NEW twist: the real laundering channel here sits below the resolution of country-of-origin trade data)

In October 2024 the Kachin Independence Army (KIA) seized Chipwi and Pangwa townships — the world's most consequential HREE mining district outside China — and suspended all mining and export. The Kachin Independence Organisation (KIO) reactivated licensed exports on 27 March 2025 and formalised a permit- and-tax regime (~35,000 CNY/t levy plus a two-in-ten in-kind cut) in its October 2025 Rare Earth Mining Management Regulation (anchor action). Those Kachin sites supply an estimated 60–70% of China's heavy-rare-earth-oxide imports — the proximate basis for China's ~95% world share in terbium, dysprosium and holmium.

The buyer here is China (the sole at-scale HREE separator), and China does publish granular import-by-origin. Its imports of rare-earth compounds (HS 2846) from Myanmar show the producer-side cut landing hard:

Source: UN Comtrade, reporter China (156), flow imports (M), annual, HS 2846 (compounds of rare-earth metals, yttrium, scandium), pulled via the free public preview API; partner Myanmar = M49 104. `netWgt` is gross compound weight (oxide/carbonate, not contained REO), so contained-metal moves are approximate; direction and magnitude are unambiguous.

The 2025 partial year confirms the decline continued past the March reopening. Chinese customs (GACC), via ISP-Myanmar's conflict-economy tracker and Reuters reporting: Jan–Sep 2025 ≈ 28,000 t / $624 M (53% of China's total RE imports of >52,000 t), versus Jan–Sep 2024 > 38,000 t / $724 M — a ~26% volume and ~$100 M value drop; Reuters put the first five months of 2025 down "over a third" year-on-year. (A widely-cited heavy-REO-specific series — 19,500 t (2021) → 41,700 t / ~$1.4 bn (2023) — tracks the same shape on the narrower heavy-only subset; see the basis note below.)

> Updated 2026-08-29. Findings 1–3 below stand as written. Public reporting > published since has established that the intra-origin blind spot is not one > boundary but at least four armed-group jurisdictions, which changes how the > headline decline should be read — see the section "Update 2026-08-29 — the > blind spot did not stay one blind spot" below (findings 4–6, incl. mode J — > intra-flag relocation).

The circumvention read — three findings, none of which is a phantom flag

1. Mode-A (trans-shipment relabelling) is structurally impossible to test — and there is no surger. Myanmar HREE is ion-adsorption clay hauled across a land border (Kachin → Yunnan smelters). Like Wa-State tin concentrate ([tin](2025-tin-wa-state-ban-circumvention-transshipment.md)), land-border concentrate cannot re-label through a third country — there is no maritime consignment to re-consign, no ~0%-capacity transit country that can suddenly "export" IAC clay it has no deposits for. The antimony volume-implausibility test ("can the transit country even make this?") has no purchase here. No phantom origin appears in China's 2846 partner data.

2. The one real substitution is GENUINE mode-B, not laundering — Laos. The material story is that Chinese operators relocated IAC mining into Laos, which has genuine ion-adsorption-clay deposits of its own. China's HS 2846 imports from Laos: 2023 = 11,308 t / $104 M → 2024 = 10,748 t / $260 M (flat volume, value up on price). Laos is not a re-flagged Myanmar molecule — it is real Lao clay dug under Chinese operation. This is the same mode-B relocation shape seen in nickel/bauxite (Chinese-owned capacity rebuilt under a new-but-genuine flag), not the antimony trans-shipment shape.

3. The genuinely NEW lesson — the real laundering channel sits *below* country-of-origin resolution. The anchor action itself asks the operative question: > "How does the KIO levy interact with Chinese customs declarations — are > exports recorded as originating from KIO-controlled territory or laundered > through SAC channels?"

Both a KIO-permitted export and an export papered under the rival Myanmar SAC (junta) Notification 93/2024 licence cross the border as "Myanmar (104)" in Chinese customs. The circumvention that matters here — jurisdictional rent-capture relabelling between an armed non-state actor and the junta it displaced — happens entirely within a single country-of-origin code. Country-level trade data has no resolution below the flag, so the detector is structurally blind to this intra-origin laundering. Every prior case in this corpus relabels across a border (antimony CN→TH, solar CN→SE-Asia); this is the first where the relabelling is sub-national and therefore invisible to the flow leg by construction.

Common-ownership tell: N. There is no namesake shell (no Youngsun→Unipet pipe). The "pipe" is a state/militia-commercial arrangement — hundreds of Chinese-operated extraction sites paying the KIO levy and hauling to Yunnan smelters that Chinese firms also run. Vertical integration across a land border, not a re-export front company.

Why it matters for the buyer

1. This is the upstream anchor of the magnet control. A risk team reading "China controls ~90% of heavy-REE separation" is really exposed to two Kachin townships governed by a non-state armed group. The Tier-1 exposure score keys on China-origin for the separated oxide and finished magnet — but the single point of failure is upstream, foreign, and in an active conflict zone. The Oct-2024 pause spiked Tb/Dy spot prices; a KIO permit revocation or ceasefire breakdown is a live single-source disruption behind an apparently "Chinese" supply chain. 2. Official data is right about the flag and wrong about the origin. The separated oxide and the magnet ship as Chinese; the molecules originate in Kachin. Any model keyed on the stated origin of the finished good misreads where the chokepoint actually is. 3. Not all circumvention crosses a border. When rent-capture relabelling is intra-origin (militia levy vs junta licence), the honest signal is the direct-line collapse + the price channel (Tb/Dy spikes on each Kachin disruption), not a phantom-flag surge — because the phantom flag cannot exist. The detector must know its own blind spot.

Update 2026-08-29 — the blind spot did not stay one blind spot: four armed-group jurisdictions, one customs code

When this case was written (Jul-2026) the intra-origin laundering channel had two sides to it: a KIO permit or an SAC (junta) licence, both crossing as "Myanmar (104)". Public reporting published since has established that Myanmar HREE extraction is no longer a Kachin story at all. At least four distinct non-state armed groups now host ion-adsorption-clay mining, in four different districts, each running its own taxation regime — and every one of them exports under the same single country-of-origin code.

The TNLA's entry has a stated commercial motive on the public record: the group ceded ruby-rich Mogok back to the Myanmar military under its October 2025 bilateral ceasefire, losing that revenue and the taxed trade routes with it — and the rare-earth infrastructure appears in its territory at about that moment (Stimson, 21 Jul 2026).

Finding 4 — the flag-level decline is an upper bound on the disruption, not a measure of it

The trade spine above reads the −38% volume drop (2023→2024) and the −26% Jan–Sep-2025 fall as the Kachin cut landing hard. That reading now needs a correction, and it cuts against the alarming interpretation rather than for it. Three of the four districts came online inside the same window the Kachin line was falling — NDAA sites scaling through May-2025, UWSA pools through Feb-2025, TNLA pools from Dec-2025. Under one customs code a district-level shutdown and a district-level start-up are arithmetically indistinguishable: the flag nets them against each other before the number is published. So the observed decline is the net of a Kachin loss and an at-least-three-district gain, which makes it a floor on the gross Kachin disruption and a ceiling on the net supply loss. A risk model that read "Myanmar HREE −38%, therefore the chokepoint tightened by 38%" has the sign right and the magnitude unknowable from this data.

Finding 5 — mode J: intra-flag relocation

This is a mode the corpus has not previously named, and it is the sub-national analogue of mode B (capacity relocation). In mode B the controlled producer rebuilds real capacity in a different country and ships under a new-but-genuine flag (BTR at Morowali; Chinese IAC operators in Laos, in this very case). In mode J the extraction restarts in a different sub-national jurisdiction under a different governing authority — but inside the same country-of-origin code, so no new flag ever appears and the relocation is invisible in the flow data by construction.

The three preconditions that make mode J available here are general, and they are what a detector should screen for elsewhere:

1. Low capex, no fixed plant. In-situ ammonium-sulphate leaching of ion-adsorption clay needs pools and hoses, not a refinery. A site can be stood up in months — which is exactly what the Mantong imagery shows between Nov-2025 and the Jul-2026 report. 2. A land border, so no maritime consignment exists to re-consign. As established above, this is why mode A cannot fire — and it is the same fact that makes mode J attractive instead. 3. Fragmented territorial control, so several authorities can each host and tax the same activity. The levy payer changes; the flag does not.

Where all three hold, disruption relocates rather than stops, and country-level trade data will systematically understate supply resilience.

Finding 6 — the number of sub-national failure points went from 2 to 4 in ~13 months

The original case warned that "China controls ~90% of heavy-REE separation" really resolves to two townships governed by a non-state armed group. The correct count today is at least four armed-group jurisdictions across two states — each with its own levy, its own ceasefire exposure, its own Chinese operating counterparties. That is worse for single-point-of-failure analysis in one respect (four active conflict-zone jurisdictions to track, none of which files production statistics) and better in another (no single district's closure now severs the feedstock the way Chipwi/Pangwa did in Oct-2024). Both directions are real; neither is visible in the flag.

Cross-check: the GACC-derived reporting and Comtrade agree

An independent validation of the trade spine. Shan Herald, citing ISP-Myanmar, reports Myanmar exported ">$4 billion worth of rare earth minerals to China between 2017 and 2024", of which 84% ($3.6 bn) after the February 2021 coup, with "in 2023 alone, export values peaked at a record $1.4 billion". The Comtrade pull in the table above — a different dataset, reporter China, HS 2846 — puts 2023 at $1,441.9 M. Two independently-sourced series landing on the same peak year at the same magnitude is a meaningful check on the spine.

What is NOT established by this update — labelled gaps

• Site counts are not tonnage. "12 sites", "19 sites", "20 leaching pools" are

satellite- and video-derived counts of infrastructure, not production. No per-district output figure is published by anyone, and none is inferred here. The share of the Myanmar line attributable to Shan State remains unquantified in free data — absent, not zero, and explicitly not scored as small.

• The substitution is timing-consistent, not causally established. That three

districts scaled while Kachin fell is a coincidence of window. No public source traces a specific volume from Chipwi to Mantong or Mong Yawng, and this case does not assert one.

• No common-ownership shell in the new districts either. Reuters places "a

Chinese company using Chinese-speaking managers" at a UWSA site, but names no entity, so no registry hop can be run. Layer 2 stays N — for want of a named counterparty, not for want of looking.

• The 2025 annual trade line is now closed (re-checked 23 Sep 2026 —

Comtrade now carries China's 2025 annual HS 2846; see the 2026-09-23 update below). China still files no monthly series to Comtrade (Sep–Dec 2025 monthly queries return zero rows), so intra-year timing remains GACC-only.

Update 2026-09-23 — full-year 2025 lands; the Kachin gate reopening cannot be isolated

Trigger: circumvention-leads — China's Yunnan side reportedly reopened up to eight KIA-controlled Kachin border gates on 26 Oct 2025 after a year-long closure (Mizzima 30-Oct-2025; Irrawaddy, "China reopens four border gates in Myanmar's Kachin State controlled by KIA"; Myanmar Now). No Chinese government notice or gazette confirms it — Myanmar-side/NGO reporting only, and the count differs by outlet (four vs eight gates) — so it is recorded here as reported context, not filed as a policy action.

UN Comtrade now returns reporter China's 2025 annual HS 2846 imports:

Source: UN Comtrade subscription API, reporter China (156), flow M, annual 2025, HS 2846, partners 104 / 418 / 0 (World); all rows `partner2Code=0`, `motCode=0`, `customsCode=C00` (totals, no dimension replication). Pulled 2026-09-23.

What it says:

1. The 2025 Myanmar decline was shallow (−9.5% wt), not the −26% the Jan–Sep GACC/ISP figure implied. Arithmetic on the two sources: ~40.1 kt full-year minus ~28 kt Jan–Sep leaves ~12 kt for Oct–Dec 2025 (~4.0 kt/month) vs ~3.1 kt/month for Jan–Sep — a ~30% higher Q4 run-rate. That fits in time with a late-2025 easing (the reported 26-Oct gate reopening, the KIO's Oct-2025 permit regime, and TNLA pools coming online from Dec-2025), but it cannot be attributed to any one of them: (a) the two figures are different sources and possibly different baskets (see method note — the ISP figure is GACC-derived "rare-earth imports", the Comtrade line is HS 2846), so the Q4 residual is indicative only; (b) the reopening reportedly covered consumer goods and construction materials, not ore; (c) all Myanmar flows share one origin code (finding 3). Detector read on the lead: unresolvable in free data — not negative, not positive. 2. Mode-B (Laos) is now the growth leg: +55% by weight, +64% by value, to 21.7% of China's HS 2846 imports by weight (16,629 / 76,619). Laos is still read as genuine relocated IAC capacity, not a re-flag — there is no public evidence of Myanmar clay crossing into Laos, and Laos has its own deposits. The flat world total with Myanmar down and Laos up is the substitution signature finding 2 predicted. 3. World value +12.8% on flat volume — unit value rose from ~$18.0/kg to ~$20.4/kg. Mix and price are not separable in HS 2846; recorded, not interpreted.

Common-ownership tell: still N. No named entity sits on the reopened-gate corridor in any source read.

Gate stays GATE 0. Every figure above traces to a named public source, but the two conditions that blocked GATE 1 in Jul-2026 both still hold and a third has joined them: (a) no common-ownership shell exists to trace, (b) the operative relabelling is sub-national and below customs resolution, and now (c) the new-district evidence is site-count imagery, not trade or production data, so the relocation is documented but not measured.

Method & honesty rails

• Trade data: UN Comtrade, reporter China (156), imports, annual, HS 2846

(rare-earth compounds) for the 2021–2024 spine; GACC-sourced 2025 partial-year figures via ISP-Myanmar and Reuters/CNBC reporting. HS 2846 pools all rare-earth compounds (light + heavy, oxide + carbonate); it is broader than the heavy-REO-only series some analysts cite, which is why the Comtrade annual levels (34.7–72.1 kt) sit above the partial-year heavy-only figures (28–38 kt). The two are different baskets on the same underlying flow; the post-takeover decline is consistent across both. Contained-REO tonnage is not isolable in free data.

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

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

• Inference, not accusation: the intra-Myanmar (KIO/SAC) laundering channel is

flagged as unquantifiable in free data, not asserted. No individual shipment is called illegal. Laos is presented as genuine relocated capacity, not a re-flag.

• New-district evidence (2026-08-29 update): Stimson Center (21 Jul 2026),

Shan Human Rights Foundation (25 Aug 2025), Reuters (12 Jun 2025) and ISP-Myanmar research via Shan Herald. All are satellite-imagery, video or field-reporting derived site counts — infrastructure, not output. No district-level tonnage is published and none is inferred; the Shan-State share of the Myanmar line is labelled unquantified, never treated as small.

• GATE 0 by construction: every table figure is traced (Comtrade + GACC/ISP +

the named reports above), but (a) the operative laundering channel is sub-national and below customs resolution, so no laundered fraction is quantifiable; (b) there is no common-ownership shell to trace — the coupling is state/militia-commercial, and the one Chinese operator placed at a UWSA site by Reuters is unnamed, so no registry hop can be run; (c) the mode-J relocation across four armed-group jurisdictions is documented but not measured. The honest signal is the direct-line collapse read as a net of loss and relocation, the mode-B Laos leg, the mode-J intra-flag leg, and the price channel — not a phantom-flag surge.