Is the wall leaking? Silicon metal and the origin ladder

Trade-flow companion to the price wedge (R72). DUAL-SCORE / alternative-track — never folded into any Tier-1 exposure score. The wedge measures scarcity; this desk measures the other half: is the control actually holding, or is the restricted trade re-entering the buyer under a new flag? Research, not investment advice. Origin-relabelling is INFERRED from statistical implausibility plus public corporate records, never asserted as evasion of any particular consignment.

Verdict

Silicon metal is the corpus's longest-running closed door: the US antidumping order on silicon metal from China (A-570-806) has been in force since 10 June 1991 at a 139.49% cash-deposit rate, continued through a fifth sunset review in 2023. The direct line is genuinely shut — China supplied 46 t of the ~101,000 t the US imported in 2023, and even the 2025 rebound (877 t) is under 1% of the market.

What the 35-year record shows is not leakage through one door but an origin ladder: each time a duty wave names a cohort of non-China origins, that cohort's line goes to zero within months, and a new cohort — including origins that appear nowhere in the USGS silicon production table — appears from ~zero to take its place. This case traces three waves, and finds that the volume-implausibility detector splits the newest cohort cleanly along the line the regulator later drew:

Petition-cohort totals: Ferroglobe USA / Mississippi Silicon petition, 24 Apr 2025, as summarised by [International Trade Insights](https://www.internationaltradeinsights.com/2025/04/petition-summary-silicon-metal-from-the-republic-of-angola-australia-the-lao-peoples-democratic-republic-norway-and-the-kingdom-of-thailand/). Our independent UN Comtrade pull reproduces the 2023 figure exactly (17,277 t) and the 2024 figure to within 0.9% (39,123 t vs 38,768 t) — see Method.

Layer 1 — volume implausibility, and the split it predicts

US imports of silicon metal (HS 280469, net weight, tonnes), by origin, UN Comtrade:

Trade data: UN Comtrade free public preview API, reporter USA (842), flow imports (M), HS 280469 ("silicon containing < 99.99% by weight of silicon" — the metallurgical-grade line; Commerce's scope is US statistical subheadings 2804.69.1000 and 2804.69.5000, explicitly excluding semiconductor-grade 2804.61.0000). World totals for 2020–22 and 2025 were not returned by the preview tier and are shown as n/a rather than estimated. Production shares: [USGS Mineral Commodity Summaries, February 2026, silicon](https://pubs.usgs.gov/periodicals/mcs2026/mcs2026-silicon.pdf) — world silicon-metal production 4,600 kt in 2025, of which China 4,000 kt (~87%).

The 2025 petition named five origins. The USGS table splits them in half: Australia and Norway are listed producers; Angola, Laos and Thailand appear nowhere in it, and sit inside a residual "Other countries" bucket of just 42–46 kt for the entire rest of the world. Angola is the sharpest instance — zero tonnes to the US in every year through 2023, then 1,675 t in 2024 and 2,833 t in 2025, from a country with no silicon smelter in the USGS record and whose two publicly-announced silicon projects (a Sonangol / Qinghai Lihao polysilicon MOU, and the MN Kitota / ST New Materials ferro-silicon plant in Kwanza Norte) were still pre-production announcements over that window.

Commerce and the ITC then drew the same line. In the final determinations (23 Feb 2026) neither Angolan nor the Lao respondent filed a questionnaire response, so all rates are adverse facts available: Laos 94.44% AD + 69.10% CVD; Angola 68.45% AD. Commerce additionally classified both Laos and Angola as non-market economies for future proceedings. On 6 April 2026 the ITC found Laos materially injuring, Angola threatening material injury, and terminated the Thailand CVD case as negligible. The Australia and Norway legs — the two USGS-listed producers — were still in a supplemental final phase in July 2026. The origins that could not plausibly have made the material did not defend themselves; the ones that could, litigated.

This is the second time in the corpus that the detector has been externally graded (after hardwood plywood), and the first time it has been graded on a split cohort rather than a single origin.

Layer 2 — the five-month Hong Kong burst

The monthly series is where the petition-perimeter rule from the plywood case shows up again. Hong Kong has no silicon smelting capacity of any kind and had never appeared as an origin of US silicon-metal imports:

Tonnes, HS 280469, UN Comtrade monthly. The petition was filed 24 April 2025; Commerce initiated 21 May 2025 and issued preliminary determinations from 30 Sep 2025.

Hong Kong's line switches on in the month the petition was filed, runs for five months totalling 2,812 t, and switches off again as preliminary duties land and the named cohort goes dark. That shape is consistent with material moving through an un-named entrepôt during the window between a petition and the first cash-deposit requirement — the classic gap that the perimeter rule predicts. It is not consistent with Hong Kong having become a producer.

The origin still rising through 2026 is Viet Nam (323 t in 2024 → 1,738 t in 2025, with 925 t in the first five months of 2026) — also absent from the USGS table, also un-named by any petition. On the plywood corollary, that is the cheapest remaining door.

The corporate pipe (the ownership tell — partially traced)

Commerce named the "Angolan" producers/exporters in the final AD determination:

> PC Silicon Co. Limited and Wanhongda International Limited > — FR 2026-03477, 23 Feb 2026

Neither carries an Angolan corporate form (Lda. / S.A.). Both carry Hong Kong private-company forms, and Wanhongda International Limited is indexed in public mirrors of the Hong Kong Companies Registry as a HK private company limited by shares, CR no. 2304691, incorporated 6 November 2015, registered in Wan Chai. PC Silicon Co., Limited is likewise indexed in Hong Kong company directories. We were unable to render either record directly from the Companies Registry's ICRIS Cyber Search Centre from this host — see the gate note below.

The Lao leg routes to the same jurisdiction. The sole Lao respondent, Lao Silicon Co., Ltd. — described in trade press reproducing the company's own introduction as "founded in 2006 and invested by A&C (Far East) Industrial Ltd Hong Kong", producing east of Vientiane since 2007 and the only silicon metal manufacturer in Laos (Ferro-Alloys).

So all three implausible or newly-appearing origins in this cohort — Angola's two named exporters, Laos's sole producer, and the 2025 entrepôt itself — point at Hong Kong. That is a domicile pattern, and it is presented as one: we have not traced any of these entities to a specific Chinese parent, smelter, or shipment, and no public document we found asserts one.

Which mode is this?

Two, adjudicated apart inside one petition:

• Mode B (capacity relocation) — Laos. Real furnaces, real output, foreign

(Hong Kong) investment, new flag. Commerce's 69.10% CVD finding against Lao government subsidy programmes and its NME classification of Laos both presuppose actual Lao production. Nothing here is phantom; the economic origin has simply moved and the legal origin followed it.

• Mode A (relabelling), inferred — Angola and Hong Kong. Angola: no smelter in the

USGS record, zero → 4,508 t across two years, exporters with Hong Kong corporate forms that declined to answer Commerce. Hong Kong: zero capacity, a five-month burst bounded precisely by the petition and the preliminary duties. Both are inference from implausibility, not from any traced consignment.

The new structural finding this case adds to the corpus is the origin ladder: where the plywood case showed one hop (China → Viet Nam) graded by one regulator, silicon metal shows the hop repeating on a cycle. Each petition costs the domestic industry roughly five years and produces a cohort that goes to zero; the ladder then adds a rung. The 2021 wave took out real producers (Iceland, Kazakhstan) alongside a closed plant (Bosnia); the 2026 wave took out a relocated producer (Laos) and a paper origin (Angola) while the genuine producers (Australia, Norway) were still litigating. The remedy perimeter moves one cohort behind the flow, every time.

Why it matters for the buyer

1. Silicon is now a US critical mineral. It was added to the Final 2025 List of Critical Minerals published 7 November 2025 (USGS MCS 2026). A buyer's silicon exposure model keyed on stated country-of-origin has been wrong in a different direction every five years for three decades. 2. "Not from China" is doing no work here. The direct China line has been closed since 1991 and is under 1% of US imports — yet China makes ~87% of the world's silicon metal. Any diversification claim resting on the China line being small is reading the one number the control was designed to move. 3. The genuine base is narrow and concentrated. Strip out every origin that failed the plausibility test and US supply is Brazil, Canada and Norway — about 88% of 2024 imports from three countries. That is the number a risk team should be carrying, not the 15-origin import table. 4. A newly-charged origin is the safest origin (plywood corollary, holding here): Laos and Angola now carry 68–94% duties and NME status. The un-named, un-charged, USGS-absent origins — Viet Nam today — are where the next cohort forms.

Method & honesty rails

• Trade data: UN Comtrade free public preview API

(comtradeapi.un.org/public/v1/preview), reporter USA (842), flow M, HS 280469, annual 2019–2025 and monthly 2024-01 → 2026-05. netWgt is net weight in kg, converted to tonnes. Months and world totals the preview tier did not return are shown as blank/n/a and never interpolated. Late months lag ~2–3 months.

• Independent cross-check: our 2023 petition-cohort total (Angola + Australia + Laos

+ Norway + Thailand = 17,277 t) matches the petition's own stated 17,277 t exactly, and 2024 matches to within 0.9%. The dataset is therefore validated against a figure computed by a third party from US official statistics.

• Regulatory record: Federal Register texts retrieved from govinfo.gov

(Angola AD final, Laos AD final + NME, Laos CVD final, ITC finals / orders, 9 Apr 2026).

• Production shares: USGS Mineral Commodity Summaries, February 2026, silicon.
• Alternative-track only: this never touches buyerRelativeScore or any Tier-1

exposure score. It sits beside them, like the China–West price wedge.

• Inference, not accusation: no individual shipment is asserted to be illegal or

mislabelled. Where a regulator has made a finding (Laos, Angola), that finding is quoted as the regulator's, not ours. Hong Kong and Viet Nam carry no finding at all and are presented purely as an unexplained shape in the data.

Why this is GATE 0, and what would close it

Every figure in every table above traces to a named free public dataset. The ownership layer does not fully close, so the case is filed GATE 0:

1. PC Silicon Co., Limited — indexed in Hong Kong company directories, but no CR number, incorporation date or registered address was retrieved. Unverified. 2. Wanhongda International Limited (CR 2304691) — read from third-party mirrors of the Hong Kong Companies Registry, not rendered directly from ICRIS (the mirrors are behind a bot wall from this host). Needs a direct ICRIS confirmation. 3. No link is traced from any of these entities to a Chinese parent or smelter. The Hong Kong domicile pattern is real; the hop behind it is an open question, not a finding. 4. A&C (Far East) Industrial Ltd (Lao Silicon's stated investor) is sourced to trade press reproducing company self-description, not to a registry filing. 5. Lao Silicon's nameplate capacity was not found, so we cannot test whether its 7,758 t of 2024 US shipments exceed plausible single-plant output — i.e. whether the Lao leg is purely mode B or has a mode-A component on top. 6. Hong Kong's 2,812 t has no consignor names, and HK's own re-export statistics (Census & Statistics Department) were not cross-pulled against the US import line.

Closing (1), (2) and (6) would move this to GATE 1 on the current claims. Closing (3) would be a materially stronger case than any currently in the corpus.

Register note

The controlling actions for this case — the US AD/CVD orders on silicon metal from Angola and Laos (Commerce finals 23 Feb 2026; ITC finals 6 Apr 2026; orders published 9 Apr 2026) — are not yet filed in the IPTM register. This case is anchored to the nearest filed action, Commerce's parallel preliminary AD determinations on crystalline silicon PV cells from India, Indonesia and Laos (2026-04-28), which sits one step downstream on the same silicon chain and names the same successor origin. The silicon metal orders should be filed and this case re-anchored.