Is the Russian-LNG ban leaking? The blank-origin case

Trade-flow companion to the price wedge (R72), buyer-side. This is a DUAL-SCORE / alternative-track signal, never folded into any Tier-1 exposure score. Research, not investment advice; nothing here is asserted as smuggling or as a breach by any named party.

Every other case in this register starts by asking which implausible origin surged. This one could not get that far. Before the detector can test an origin's production capacity, the buyer's statistics have to record an origin — and for roughly 30% of the EU's LNG imports they do not. That is the finding. It is not a phantom flag. It is no flag, in the EU's own books, concentrated in three member states, and present in both of the EU's official datasets by two different mechanisms.

Verdict — GATE 1, three results, one of them a clean positive

Layer 2 (common-ownership tell): N — not traced. See Layer 2 below for what was and was not established, and why its absence is not informative here.

Layer 0 — the origin field is blank, and it is blank in the same three places twice

The EU publishes two official, free, monthly series that should both answer "where did this LNG come from". They disagree, and where they disagree they are blind in the same three member states.

Dataset 1 — Eurostat `nrg_ti_gasm` (imports of natural gas by partner country, monthly), siec=G3200 (liquefied natural gas), unit=MIO_M3. Dataset last updated 2026-08-25; series runs to 2026-07.

Reproduce: https://ec.europa.eu/eurostat/api/dissemination/statistics/1.0/data/nrg_ti_gasm?format=JSON&lang=en&unit=MIO_M3&siec=G3200&geo=<GEO>&partner=<PARTNER>&sinceTimePeriod=2021-01

The "Not specified" bucket went 477 → 16,263 Mio m³, a 34× rise in four years, and is now a larger line than Nigeria or Qatar. It is not spread across the bloc:

2024: 33,394 / 117,590 = 28.4%. H1-2026: 30.7%.

France's case is the odd one, because France does report a partner split — it just is not an origin split:

Two things are wrong on the face of it. France's Algerian line is pinned to three significant figures for three consecutive years (18,182 / 18,845 / 18,885) — a measurement does not do that; a contract allocation does. And France reports zero from the United States in 2025, the year the US supplied 55,373 Mio m³ (37.8%) of all EU LNG, into a country whose Dunkerque, Montoir and Fos terminals are among the largest regasification capacity in the bloc. France's declared mix sums exactly to its total (18,885 + 8,308 = 27,193), so this is not suppression of small partners — it is the whole book.

Dataset 2 — Eurostat Comext `DS-045409`, the EU's customs trade database, CN 2711 11 00 (natural gas, liquefied), imports, quantity. Dataset last updated 2026-08-14.

Reproduce: https://ec.europa.eu/eurostat/api/comext/dissemination/statistics/1.0/data/DS-045409?format=JSON&freq=A&reporter=<ISO>&partner=<ISO|WORLD>&product=27111100&flow=1&indicators=QUANTITY_IN_100KG&time=<year>

"No rows" is Comext's distinct state from a reported zero: nothing is returned at all. So in the EU's customs database, France declares zero LNG imports of any origin in a year it landed ~27 bcm, and Germany and Poland do not appear on the line at all.

The two datasets are blind in exactly the same three member states, by two unrelated mechanisms. That correlation is the load-bearing fact. A single dataset with a gap is a data-quality note; the same three importers missing from both the energy return and the customs return means the EU has no series in which those cargoes' origin can be tested.

What follows arithmetically

The EU27 "imports of Russian LNG" line — the number quoted whenever the bloc's remaining Russian gas exposure is discussed — is, for 2025, exactly and only Belgium + Spain + Netherlands + Portugal + Italy:

> 5,936 + 3,792 + 2,503 + 294 + 207 = 12,731 Mio m³ = the published EU27 total, > to the last unit.

France, Germany and Poland contribute nothing to it — not a small number, structurally nothing, because none of the three reports an origin. The published Russian share is therefore a lower bound over a partial denominator, not a measurement of the bloc. How much of a lower bound is not knowable from EU sources; what is knowable is that the question cannot be closed with them.

The mirror, flagged as an estimate. UN Comtrade carries a France series for HS 271111 built from partners' export declarations, not from French returns — every row is flagged isReported: false, isNetWgtEstimated: true, legacyEstimationFlag: 6. It reads, for 2025: United States 10.099 Mt, Russian Federation 5.427 Mt, Algeria 2.626 Mt, Nigeria 0.810 Mt, world 20.581 Mt — close to the inverse of France's own return. Reproduce: https://comtradeapi.un.org/public/v1/preview/C/A/HS?reporterCode=251&period=2025&cmdCode=271111&flowCode=M (the preview returns rows split by customsCode/motCode/partner2Code; take the widest slice per partner). This is a UNSD estimate and is recorded as one. It is quoted here only to establish that the gap is large, not to size it: a number no member state reported cannot be added to a number they did.

Layer 1 proper — where a control was testable, it worked

The 14th package's trans-shipment ban is the one leg of this file where the detector has a complete series and a clean answer. Eurostat nrg_te_gasm (exports of natural gas by partner), siec=G3200, EU27, Mio m³:

These are reported zeros, not absent data — the China, Singapore, Taiwan and Korea series each return a value in all 25 months from 2024-06 to 2026-06, and the last non-zero month for the whole Asian reload route is February 2025, one month before the ban's late-March effective date. The residual EU LNG export book is now almost entirely intra-EU (Belgium 725, Germany 244, Italy 218, Finland 193).

The register's own action file for the 14th package predicted this leg ("removes the EU port arbitrage that has propped up Yamal LNG's winter delivery schedule to Asia. Watch Asian-spot LNG basis spreads from Q2 2025"). The flow data grades it correct, and the wind-down ran ahead of the deadline — consistent with terminal operators dropping reload service outright rather than verifying cargo origin per-cargo, which is a compliance overshoot: all EU-origin reloads to Asia stopped, not only Russian ones.

The cliff that wasn't — read the carve-out, not the headline

The 19th package (anchor action) bans imports of Russian-origin LNG, with short-term contracts prohibited from 25 April 2026 and long-term contracts (>1 year, executed before 17 June 2025) running to 1 January 2027. The headline date has passed. The flow did not move:

Russian LNG's share of EU imports in the two months after the ban's first deadline is the highest since 2023. Nothing about that is unlawful: the surviving volume is long-term-contract volume, which the instrument expressly permits until 1 January 2027. This is the nitrogen-fertiliser rule reproducing out of sample — check the escalator against the line it will actually hit. There, an 8× duty ramp was aimed at a heading the lowest rung had already emptied. Here, a headline "full import ban" took effect against the smaller of the two contract books, and the register's own action file recorded the carve-out correctly at filing time. A risk desk that read "EU bans Russian LNG, effective 25 April 2026" and marked the exposure closed is wrong by at least eight months — 1 January 2027 is the date that matters, and the observable test of whether that cliff binds will run into a statistical system that, on today's evidence, cannot see a third of the cargo.

Layer 2 — the common-ownership tell: N, not traced

No corporate pipe was established for this case. What the public record supports is only that the volume is contractual rather than opportunistic (long-term offtake agreements predating 17 June 2025 are the instrument's own stated category), which is why the flow survives the first cliff. No entity chain, registry hop, or beneficial-ownership link is asserted here.

Two rails on that N. First, unlike the Russian-seafood row — where the absence of an ownership tell was informative, because a lawful channel needs no shell — the N here is uninformative, and for the same reason the rest of this file exists: layer 2 was never reached, because layer 1 could not be completed on 30% of the flow. An untested hypothesis is not a negative result. Second, this file deliberately does not convert the juxtaposition "France reports no LNG origin" + "French companies hold equity in Russian LNG projects" into an implication. Both may be true and unrelated; the reporting gap covers all origins including the United States, which is exactly what a mechanical reporting practice looks like and exactly what a targeted concealment would not.

Open, and the highest-value next step for this row: obtain a published methodological explanation. Eurostat's ESMS metadata for nrg_ti_gasm, and Regulation (EC) No 1099/2008 on energy statistics, should define what the partner field means for natural gas and whether member-state derogations or confidentiality flags apply. Until that is read, the cause of the blindness is unknown and is recorded as unknown — this file establishes only that the blindness exists, is large, is growing, and is correlated across two datasets.

Mode — (I) reporter-blind origin (new; a statistical mode, not a behavioural one)

Modes A–H all describe something an exporter or importer does: relabel (A), relocate (B), transform (C/C⁻/C⁰), reroute (D/D⁰), falsify an end-user (E), shift scope (F), drop to an upstream heading (G), license a passthrough (H). Every one of them assumes the buyer's statistics record the field the detector reads.

Mode I is the failure of that assumption. Nobody has to lie. The importing member state simply does not report the origin — or reports something that is not an origin — and the flow becomes untestable at the point of entry. It belongs in the taxonomy because it is prior to every other mode: A cannot be detected, C cannot be detected, and neither can their absence, in a jurisdiction operating under mode I. It is the register's first entry where the blind spot is on the buyer's own statistical system rather than in the flow.

Three transferable rules:

1. Test the denominator before testing the origin. Every share this register quotes has an implicit denominator of reported trade. Before concluding "origin X supplies n% of imports", pull the unattributed and non-reporting share. On EU LNG that is ~30% — large enough to invert conclusions, and growing every year since 2022 (0.6% → 11.1% on the "Not specified" line alone). 2. Cross the energy return against the customs return. They are compiled by different national bodies under different legal instruments and fail independently. Where they are blind in the same member states, as here, the gap is structural rather than a vintage artefact — and no third EU source closes it. 3. A flat series is a contract, not a measurement. France's Algerian line at 18,182 / 18,845 / 18,885 across three years is the tell that the field carries a commercial counterparty rather than a loading port. Screen partner series for implausible year-on-year stability before trusting them; physical trade is noisy and a three-year plateau in a spot-exposed commodity is a red flag on the statistic, not on the trade.

Why it matters for the buyer

1. The number everyone quotes has a hole in it. "Russian LNG is 8.7% of EU imports" (2025) is arithmetically the Belgian, Spanish, Dutch, Portuguese and Italian book and nothing else. A corporate or financial exposure model keyed to the EU27 Russian share is keyed to a partial census. State it as a lower bound or do not state it. 2. The exposure did not close in April 2026. It is contracted to close on 1 January 2027, and the flow through the first cliff went up. Any position, covenant or disclosure that treated 25 April 2026 as the end of EU Russian-LNG exposure is mispriced. 3. One control in this file worked, and it is the cheap kind. The trans-shipment ban severed a service (port reload) rather than a molecule, and went to zero a month early with no phantom flag anywhere. Controls attached to a physical service performed inside the enforcing jurisdiction are the ones this register keeps finding to be binding; controls attached to a contract are the ones with the carve-outs.

Method & honesty rails

• Data: Eurostat nrg_ti_gasm and nrg_te_gasm (updated 2026-08-25, series

to 2026-07), siec=G3200, unit=MIO_M3; Eurostat Comext DS-045409 (updated 2026-08-14), CN 2711 11 00; UN Comtrade free public preview API, HS 271111. Reproduce URLs are printed above each table. All three are free and key-less. Intra-EU partners are 0.3–0.5% of the EU27 LNG import total, so the aggregate is effectively extra-EU.

• Absent ≠ zero. Comext "no rows" (Germany, Poland) is reported distinctly

from France's declared 0.000, and the export-side zeros in the Layer-1 table were checked to be reported zeros with values present in all 25 months. Nothing missing is scored as good news.

• Estimates are labelled. The UN Comtrade France series carries

isReported: false and is described as a UNSD mirror estimate everywhere it appears. It is not added to any EU total.

• Register-state, not world-state. This file establishes that the EU's

published series cannot resolve the origin of ~30% of its LNG. It does not establish that those cargoes are Russian, that any party mis-declared anything, or that any sanction was breached.

• Inference, not accusation. No entity is named as a circumvention vehicle.

Layer 2 returns N because it was not traced — not because it was tested and found clean.

• Alternative-track only: never touches buyerRelativeScore or any Tier-1

exposure score.

• Not yet in a committed fetcher. Every figure in this case was pulled ad hoc

against the URLs above; there is no data/intelligence/lng-circumvention.json artifact yet.