Is the ban leaking? Xinjiang cotton after UFLPA — the origin moved, the feedstock didn't

Trade-flow companion to the price wedge (R72). This case asks the standard question — is the control actually holding, or is restricted material re-entering the buyer through a laundered origin? — of the one control in the corpus that is not a mineral export ban: the US Uyghur Forced Labor Prevention Act (Pub. L. 117-78, rebuttable presumption in force 21 June 2022). DUAL-SCORE / alternative-track signal, never folded into any Tier-1 exposure score. Research, not investment advice; nothing below asserts that any named company shipped forced-labour cotton — the inference is structural (flows and ownership), not per-shipment.

Verdict

The last hop moved. The feedstock hop concentrated harder on China.

US apparel imports re-flagged away from China at scale after UFLPA took effect — China's share of US clothing imports halved, and Vietnam overtook it as the #1 supplier in 2025. But over the same window, China's exports of the inputs those hubs sew — cotton yarn/fabric and knitted fabric — grew faster to the three hubs than to the world, lifting the hubs' share of China's textile-input exports from 34.3% to 40.8%. This is the corpus's mode-C (transformation-washing) fingerprint in its purest agricultural form: cut-and-sew in Vietnam legally confers Vietnamese origin, but the cotton inside the fabric was grown where it always was. In 2024 Xinjiang was 92.2% of China's cotton crop — a record high, per China's own statistics bureau.

The volume-implausibility test inverts exactly as it did on Russian crude → Indian diesel: not "you cannot produce this garment" (Vietnam plainly can) but "you cannot grow this cotton" — Vietnam's domestic crop is under 1% of the cotton its mills consume.

Table 1 — the origin line moved

US apparel imports, HS chapters 61 + 62 (knitted + woven clothing), customs value:

Source: UN Comtrade, reporter USA, flow M, annual, HS ch. 61+62, pulled from the free public preview API (2025 annual is complete — December 2025 present).

Vietnam displaced China as the #1 declared origin of US clothing in 2025. On the stated-origin reading alone, a risk team would conclude US apparel had substantially de-risked from China.

Table 2 — the feedstock line did not

China's exports of textile intermediates — HS ch. 52 (cotton, incl. cotton yarn and woven cotton fabric) + HS ch. 60 (knitted/crocheted fabric) — to the same cut-and-sew hubs:

Source: UN Comtrade, reporter China, flow X, annual, HS ch. 52 + 60. China's 2025 annual is not yet in Comtrade, so this leg ends at 2024 — one year short of Table 1. For China→Viet Nam 2024, HS-4 detail confirms ch. 52 is ~98% yarn and woven fabric (5205–5207 $0.15bn + 5208–5212 $1.51bn of the $1.69bn total), not raw cotton lint.

The divergence is the signal. China's finished-garment line to the US fell 55% while its fabric line into the countries that now sew those garments grew 26% — more than four times its growth to the rest of the world.

The feedstock-implausibility leg

• Viet Nam grows essentially no cotton. USDA FAS (Cotton and Products Annual,

Hanoi, VM2025-0013) puts 2024/25 imports and consumption at a record 7.4m bales, with domestic cotton under 1% of total market demand — imports are "practically all cotton consumed."

• Crucially, Vietnam's raw cotton is not Chinese: its largest lint suppliers

are Australia, Brazil and the US (same source). So the raw-cotton channel is clean, and the exposure travels one step downstream — as yarn and fabric.

• China's cotton is Xinjiang's cotton. China's National Bureau of Statistics

(Bulletin on National Cotton Output, 27 Dec 2024) reports 2024 national output of 6.164 Mt, of which Xinjiang produced 5.686 Mt = 92.2%, a record share.

• China also imported 2.609 Mt of raw cotton in 2024 (UN Comtrade, HS 5201), so

the mill pool is not purely domestic. Even so, on in-year supply (production + imports, ignoring stock draw — an approximation), Xinjiang lint is ~65% of the cotton available to Chinese mills.

• Vietnam-reported customs data (UN Comtrade, reporter Viet Nam, HS ch. 52+60) puts

China at 46.3% (2019) and 48.1% (2021) of Vietnam's fabric-and-yarn imports. Roughly half of what Vietnam sews is Chinese cloth. (2023+ Vietnam-reported rows contain duplicate customs-procedure splits in the preview API and were discarded rather than double-counted — hence the older years here.)

Chain the three together: about half of Vietnam's cloth is Chinese, and about two-thirds of Chinese mill cotton is Xinjiang-grown. A garment that changed its flag from "China" to "Viet Nam" did not necessarily change the field its cotton came from.

The corporate pipe (the common-ownership tell)

The strongest single tell in this corpus is a controlled producer operating on both sides of the origin line under one parent. Cotton has a clean, publicly documented example.

Huafu Fashion Co., Ltd. (SZSE: 002042) was added to the UFLPA Entity List with 25 of its subsidiaries effective 15 January 2025 (90 FR 3899, FLETF notice 2025-00901) — the largest single tranche to date, taking the list to 144 PRC-based entities. The Federal Register describes the parent verbatim as:

> "a company based in Anhui Province, China that is vertically integrated from > cotton planting and processing to yarn spinning to textiles manufacturing… the > United States Government has reasonable cause to believe… that Huafu Fashion Co., > Ltd. sources material from the Xinjiang Uyghur Autonomous Region." > — 90 FR 3899, §2(d)(2)(B)(v) designation

The same notice lists the group's XUAR arms by name, among them Aksu Huafu Color Spinning Co., Ltd. (Aksu Prefecture), Awati Huafu Textile, Xinjiang Huafu Color Spinning Group, Xinjiang Huafu Cotton Industry Group, Xinjiang Huafu Hengfeng Cotton Industry, Xinjiang Huafu Hongfeng Agricultural Development, Xinjiang Huafu Textile, Shihezi Huafu Hongfeng / Hongsheng Cotton Industry and Huyanghe Huafu Hongsheng Cotton Industry — plus non-XUAR group nodes Ningbo Huafu Donghao Industrial (materials trading) and Ninghai Huafu Textile.

The offshore leg. In June 2020 — two years before the UFLPA presumption bit — Huafu announced a CNY 1.5bn / US$212m yarn plant in Long An province, Viet Nam, 500,000 spindles (300,000 in phase 1). Its stated rationale, as reported at the time, was that the plant "will also mean more of its products come from Vietnam, allowing it [to] dodge tariffs imposed on countries not involved in the CPTPP," and would let it "expand production while skirting Asian trade tariffs imposed on Chinese clothing" (Yicai Global, 8 June 2020).

That capacity is real and operating, not announced-and-shelved: responding on the Shenzhen exchange's investor-interaction platform, the company stated its current Vietnam capacity is 290,000 spindles (29万锭), with the Vietnamese subsidiary posting net profit of CNY 51.19m in 2022 and CNY 28.56m in H1-2023 (Jiemian, 14 Dec 2023). Against a group total of ~2.06m spindles, roughly one spindle in seven now turns outside China.

The perimeter gap. Every one of the 40 entities named in the January 2025 notice is located in the PRC — Urumqi, Aksu, Kashgar, Shihezi, Wujiaqu, Huyanghe, Altay, Kizilsu, Baotou, Ningbo, Jiangsu, Anhui. No third-country entity appears on the list, because the Entity List is by design a register of PRC-based companies. So Huafu's Vietnamese spinning arm sits inside the corporate group and outside the name-match perimeter. The rebuttable presumption still reaches goods produced "in whole or in part" with a listed entity's inputs regardless of shipping country — but reaching them requires tracing an input, not matching a name on a manifest.

What this tell is, and is not. This is a mode-B (capacity relocation) tell, not a proven mode-A relabel: because Vietnam's raw lint comes from Australia, Brazil and the US, Huafu's Vietnamese spindles may well run on non-Chinese cotton. What demonstrably relocated is the company and the flag, and — per its own 2020 framing — that was the point. The molecule-level exposure travels on the separate, much larger channel in Table 2: $14.1bn/yr of Chinese fabric into the three hubs.

The enforcer's own data points at the transit hub

CBP publishes a UFLPA Statistics Dashboard by fiscal quarter, industry and country of origin. For calendar 2023, in the apparel/footwear/textiles category, the value of shipments stopped under UFLPA whose declared origin was Viet Nam ($19.14m) exceeded China ($17.70m); 786 shipments in the category were reviewed and 430 (55%) denied (CBP dashboard figures as reported by just-style, 16 Jan 2024). Later reporting on the same dashboard puts Viet Nam at the highest value of any country subject to UFLPA review for Oct-2023→Aug-2025.

An enforcement regime aimed at Xinjiang finding more suspect value arriving from Viet Nam than from China is the enforcement-side echo of Table 1 and Table 2.

Honesty note: `cbp.gov` returns HTTP 403 to this host, so the dashboard was not read directly — these figures are second-hand from named trade press citing it, and are deliberately kept out of the tables above. Verifying them against the live dashboard is the first item on the GATE-1 list below.

Transmission chain

`` Xinjiang cotton fields (92.2% of China's 2024 crop, NBS) → XUAR ginning + spinning (e.g. Aksu Huafu Color Spinning — 90 FR 3899) → eastern-China / group mills, blended with imported lint (2.6 Mt imports 2024) → HS 52 + 60 fabric exports: $14.1bn/yr into VN + BD + KH (+26% since 2019, vs +6% to the world) ← mode C → cut-and-sew in Viet Nam / Bangladesh / Cambodia (substantial transformation → new country of origin) → US imports declared "Made in Viet Nam": $17.0bn in 2025, now #1 ⟂ parallel corporate leg: Huafu Vietnam, Long An, 290k operating spindles, same listed parent ← mode B ``

Why it matters for the buyer

1. Stated origin has stopped tracking exposure. A brand or fund reading "China is now only 14.6% of US apparel imports, down from 30.2%" would score the Xinjiang dependency as more than halved. The fabric data says a large part of what changed is the country where the last stitch happened. Any exposure model keyed on declared country-of-origin is compromised for a substantial-transformation good; the feedstock cross-check is the correction. 2. The regulatory perimeter is narrower than the risk. The Entity List names PRC companies. The exposure now sits in third-country subsidiaries of those same companies and in third-country mills buying their cloth — reachable in law, invisible to a name-match screen. This is the same structural gap the EU Forced Labour Regulation (2024/3015, application 2027) will inherit. 3. It generalises the mode-C rung. Russian crude → Indian diesel established that a legal transformation can wash economic origin. Xinjiang cotton → Vietnamese T-shirt is the same rung on an agricultural, labour-based control rather than an energy embargo — evidence the mode is about legal substantial-transformation rules, not about any one commodity.

Method & honesty rails

• Trade data: UN Comtrade free public preview API. Table 1 = reporter USA,

flow M, HS ch. 61+62. Table 2 = reporter China, flow X, HS ch. 52+60. Nominal USD, value terms only — chapter-level Chinese reporting returns no net weight, so none of this is deflated or volume-normalised. Some of the +26% in Table 2 is price, not tonnage.

• Confounders are large and are not decomposed. The 2021→2025 collapse in

China's US apparel share is not attributable to UFLPA alone. Section 301 tariffs, the 2025 reciprocal-tariff round, the repeal of the $800 de minimis exemption for China, and ordinary wage-driven relocation all push the same way. This case claims a direction and a divergence, not a causal share.

• An unexplained mirror gap, flagged not used. For 2024, China reports $34.6bn

of ch. 61+62 exports to the US; the US reports $18.4bn of imports from China — a ~$16bn discrepancy. De minimis e-commerce parcels (recorded as Chinese exports, historically outside formal US import statistics) and entrepôt/valuation differences are plausible explanations, but this is unresolved and appears in no table here. It is a second blind spot worth its own case.

• Inference, not accusation. No shipment, and no named company, is asserted to

have moved forced-labour cotton. The Huafu material is quoted from the Federal Register's own findings and the company's own public disclosures.

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

exposure score.

Gate: GATE 0 — what is missing

Every figure in both tables traces to a named free public dataset, and every ownership hop to the Federal Register or a company disclosure. The gate is held at 0 because the pipe is not closed:

1. No traced shipment. Nothing links Huafu Vietnam yarn to a specific US-bound garment, the way the antimony case's consignor/consignee chain does. The tell is structural, not per-shipment. 2. No lint provenance for the Vietnam spindles. Whether Huafu Vietnam spins Chinese-origin or Australian/Brazilian/US lint is unknown from public sources — and it matters, because it decides whether the Huafu leg is mode B alone or mode B + mode A. 3. CBP figures are second-hand (403 from this host); they need reading off the live dashboard. 4. Table 2 ends a year early (China's 2025 annual not yet in Comtrade), so the feedstock leg does not yet cover the year Vietnam overtook China. 5. No fabric-level Xinjiang share. "~65% of Chinese mill cotton is Xinjiang" is a supply-pool approximation, not a measured share of exported fabric.

Lift to GATE 1 by closing (1) or (2) with a registry/filing/customs trace, and by re-reading Table 2 once China's 2025 annual lands.