Domain B - EU Gas and LNG Re-routing

Domain B — Results: EU Gas and LNG Re-routing

Computed 2026-08-04 against hypothesis.md (commit e4821a8, pre-registered before any value was fetched). Every number below is reproducible: python -m src.domains.domain_b_eu_gas.compute regenerates results.json from the staged fixtures in data/, which themselves record the exact SQL used to compute each period average server-side.

Headline finding, discovered during data fetch, not anticipated in the hypothesis: Eurostat suppresses bilateral partner attribution for gas trade at source, for multiple reporters, not just Germany. This turned out to matter more than either pre-registered hypothesis.

Anonymization findings (the load-bearing discovery of this domain)

• Germany discloses no partner-level gas import data at all. Every one of Germany's extra-EU pipeline-gas imports is filed under a single anonymized partner code, QZ — "Countries and territories not specified for commercial or military reasons." Its intra-EU pipeline-gas imports are filed under a parallel code, QY, with the same reasoning stated for intra-EU trade. No LNG import series exists for Germany at all (consistent with Germany having no LNG import terminal of its own until early 2023). Confirmed via a direct dimensions-table scan for every partner code attached to Germany's relevant series — not inferred from an empty query result.
• Poland's tracked partners are all declared zero. Russia, the US, Qatar, and Norway all show exactly €0 for Poland across nearly the entire window, despite Poland's publicly known long-term Qatari LNG contract and Norwegian pipeline gas via the Baltic Pipe (operational since October 2022). Poland has its own anonymized bucket, QU ("Countries and territories not specified"), which almost certainly holds the real volume.
• Belgium is the exception. Of the four reporters, only Belgium discloses real, partner-attributed values for every combination checked — which is the only reason this domain has any usable data at all.

This is a genuine limitation of Eurostat Comext for gas-origin analysis, applied at the source by Eurostat's own confidentiality rules — not a FactIQ ingestion gap.

H1 — Import composition shift away from Russia: criterion not evaluable

Pre-registered falsification: RU share falls for at least 3 of 4 reporters. Only 2 of 4 reporters produced any computable share at all (Germany: impossible, zero partner data; Poland: all tracked partners are zero in both periods, share is 0/0 and undefined). The 3-of-4 criterion cannot be evaluated as designed — reported as a data-availability finding, not forced toward a verdict either way.

Of the 2 reporters that could be tested, the results disagree:

• Netherlands: RU share fell sharply — 38.1% (pre) → 10.2% (post). Directionally consistent with H1's expectation. Caveat: based on very thin data (mostly single-month pre-period observations per partner — see caveats).
• Belgium: RU share rose — 10.9% (pre) → 13.8% (post). The opposite of H1's expectation, on well-populated data (30-44 observations per series). This is plausible and real, not an anomaly: Russian LNG (unlike pipeline gas) was not sanctioned after 2022, and Zeebrugge — Belgium's LNG terminal — is a known transshipment point for Russian Yamal LNG. Belgium's RU LNG import value average roughly quadrupled post-2022 (€35.3M → €133.0M/month).

H1 is not falsified, but it is not confirmed either — it is unresolvable with this data. Two reporters, one confirming, one contradicting, is not a basis for either verdict.

H2, leg 1 — Re-routing signature (trade value rising post-2022)

• Belgium → Germany pipeline gas: confirmed, strongly. Average monthly value rose from €82.2M (pre, n=31) to €701.2M (post, n=43) — an 8.5x increase. Well-populated on both sides of the window; this is the cleanest single result in this domain.
• Netherlands → Germany pipeline gas: cannot be tested. The full series (2002–2026) has only 28 total data points, almost entirely clustered in 2002–2003 and January–April 2026. Zero pre-period observations exist for the 2019–2022 window. Reported as data-insufficient, not computed, not estimated.

H2, leg 2 — Mirror-statistic reconciliation: the pre-registered pairs are impossible; a substitute pair fails to reconcile

The pre-registered check (NL/BE export to Germany vs. Germany's import from NL/BE) cannot be computed at all. Germany's import side is 100% anonymized under QY (see above) — there is no DE-side series attributed to NL or BE to compare against. This was the domain's defining method per the original spec, and it is blocked at the data source for every Germany-anchored pair.

Substitute, run on the best available real pair instead: Belgium's exports to the Netherlands vs. the Netherlands' imports from Belgium (same CN8 code, same flow, both sides genuinely partner-attributed — the only such pair found). Result: the two sides do not reconcile.

• Average relative difference across 16 overlapping months (12 in 2022, 4 in 2026): 69.7% — far beyond the pre-registered 20% threshold that was set to allow for the ordinary Eurostat CIF/FOB valuation gap.
• 15 of 16 months exceed the 20% threshold; the one exception (2022-10, 19.3%) is the closest the two sides ever come to agreeing.
• The direction of the mismatch isn't even stable: in every 2022 month, Belgium's reported export value is 2-5x larger than the Netherlands' reported import value for the same nominal flow. In January 2026, this reverses — the Netherlands' import figure is larger than Belgium's export figure. A systematic valuation convention (like CIF/FOB) produces a stable, one-directional gap; this does not look like that.

This is the domain's second major finding: naive bilateral mirror statistics for piped natural gas do not reconcile, even approximately, for at least this pair. The CIF/FOB explanation pre-registered in the antithesis section does not fit a 70% average gap with directional reversals — something else is going on (see Antithesis).

Antithesis checks performed

• CIF/FOB valuation convention — pre-registered as the expected source of some mirror gap. Ruled out as sufficient explanation for the BE/NL result: CIF/FOB gaps are conventionally a few percent, not 70%, and CIF/FOB alone cannot explain the direction of the gap flipping between 2022 and 2026. Something else — differing customs procedures, hub-based trading/swap arrangements not captured as simple bilateral flows, or transit/re-export accounting — is more likely, but this analysis cannot distinguish which. Flagged as open, not resolved.
• Mirror-statistic check: this domain's defining method. Run exactly as designed on the only pair where it could be run at all. The result — failure to reconcile — is reported as a finding in its own right, not discarded for being inconvenient.
• Re-export vs. genuine consumption — flagged in advance, unresolved: this data cannot distinguish Germany consuming re-routed gas from Germany passing it onward again. Still true, and now compounded by the fact that Germany's import side can't be observed directly at all.
• Price vs. volume — all figures are trade value (EUR), not physical quantity. The 8.5x BE→DE increase and the RU-share results both sit inside the same 2021-2023 window as Europe's gas price spike; some portion of every value-based shift reported here is price, not volume or routing. Not decomposed in this pass.
• Base-year changes, Chinese New Year timing: confirmed not applicable, as pre-registered.
• FactIQ vintage caveat. Every Comext value used here is FactIQ's current, revised print (Phase 1 finding, reconfirmed by FactIQ directly), not what Eurostat originally published for that month.

Caveats (restated from hypothesis.md, plus what was learned)

• Only 5 extra-EU partners were tracked; this turned out not to be the binding constraint — partner-level suppression was, for 2 of 4 reporters, well beyond what the hypothesis anticipated.
• Poland's Baltic Pipe (Norway) leg cannot be mirror-checked (Norway is not a Comext reporter) — moot in practice, since Poland's Norway leg also returned zero under the tracked partner codes, consistent with the QU-anonymization finding.
• The Netherlands' data is thin across the board (mostly single-month pre-period observations for H1, essentially absent for H2 leg 1) — not the confidentiality-blackout Germany shows, but not usable as a full data source either. Worth a direct follow-up question to FactIQ or Eurostat about NL's actual reporting completeness for this CN8 code.
• Value (EUR), not physical volume — see Price vs. volume above.
• FactIQ's Comext data is not vintage-aware — see FactIQ vintage caveat above.

Bottom line

• The domain's most useful output is not either pre-registered hypothesis — it's the discovery that gas-partner attribution is suppressed at source for multiple EU reporters, and that even where both sides of a bilateral gas flow are visible, they don't reconcile. Anyone building further on Comext gas data needs to know this before trusting a bilateral partner figure at face value.
• H1 is unresolvable as designed — 2 of 4 reporters testable, and they disagree. Not a confirmation, not a clean falsification.
• H2 leg 1 (re-routing signature) is confirmed for Belgium (8.5x increase, well-populated data) and untestable for the Netherlands (no pre-period data).
• H2 leg 2 (mirror reconciliation) could not be run as designed (Germany's side is fully anonymized) and failed on the best available substitute pair — a genuine, reportable data-integrity finding, which is exactly what this method exists to catch.

Pre-registered hypothesis

Domain B — EU Gas and LNG Re-routing

Status: PRE-REGISTERED, 2026-08-04. No comparison values have been pulled at the time this file is committed — only schema structure and CN8 product-code identifiers (metadata, not results). Per project principle, this hypothesis will not be edited after results are computed; if it fails, the failure is recorded, not the hypothesis.

Data sources (exact identifiers confirmed present as of 2026-08-04)

FactIQ eu_comext_<iso2> schemas (Eurostat Comext, one physical schema per EU reporter, CN8 product detail, monthly, value in EUR). Series ID pattern, confirmed via search_series: eu_comext_{M|X}_{reporter}_{partner}_{te|ti}_p1_cn8_{product_code}_eur (M=import, X=export, te=extra-EU partner, ti=intra-EU partner).

Reporters (4, chosen in advance for a specific story, not scanned for best fit):

• de (Germany) — largest EU gas consumer; had no operating LNG import terminal until its first floating terminals came online in early 2023; historically dependent on Russian pipeline gas via Nord Stream.
• nl (Netherlands) — established LNG import hub (Gate terminal, Rotterdam); candidate re-export gateway to landlocked consumers.
• be (Belgium) — established LNG import hub (Zeebrugge); candidate re-export gateway alongside NL.
• pl (Poland) — earlier diversifier: Świnoujście LNG terminal (operating since 2016) plus the Baltic Pipe from Norway (operational October 2022); a within-window pivot away from Russian supply that predates most of the rest of the EU's adjustment.

Products (CN8, confirmed via eu_comext_lookup.product_codes):

• 27111100 — Natural gas, liquefied (LNG)
• 27112100 — Natural gas in gaseous state (pipeline gas)

Extra-EU supplier partners tracked for import composition (te): ru (Russia), us (United States), qa (Qatar), no (Norway), dz (Algeria) — the major external suppliers to Europe. Any import value from a partner outside this list is not captured; this is a declared, bounded set, not an exhaustive one.

Intra-EU mirror pairs tracked (ti), pipeline gas only:

• NL→DE: eu_comext_X_nl_de_ti_p1_cn8_27112100_eur (NL's reported export) vs. eu_comext_M_de_nl_ti_p1_cn8_27112100_eur (DE's reported import)
• BE→DE: eu_comext_X_be_de_ti_p1_cn8_27112100_eur (BE's reported export) vs. eu_comext_M_de_be_ti_p1_cn8_27112100_eur (DE's reported import)

Norway is not an EU member and has no eu_comext_no schema, so Poland's Baltic Pipe leg (import from Norway) cannot get a mirror check — it can only ever be observed one-sided, from the Polish import side. Declared here as a known, structural limitation, not an oversight to fix later.

Time windows (declared in advance)

• Pre-period: 2019-01 through 2022-01 (last full month before the 2022-02-24 invasion of Ukraine).
• Transition period, excluded from period-average comparisons: 2022-02 through 2022-09 (the acute shock window — sanctions escalation through the September 2022 Nord Stream pipeline sabotage). Included in any month-by-month chart, but not used in the pre/post averages below, to avoid a single volatile stretch dominating a period mean.
• Post-period: 2022-10 onward through the latest available month.

H1 — Import composition shift away from Russia

For each of the 4 reporters, Russia's share of tracked extra-EU gas import value (LNG + pipeline gas combined, RU value ÷ sum of RU+US+QA+NO+DZ value) should be lower in the post-period average than the pre-period average.

Falsification: if Russia's post-period share is not lower than its pre-period share for at least 3 of the 4 reporters, H1 is falsified.

H2 — Re-routing via NL/BE to Germany (mirror-statistic check)

If Dutch and Belgian LNG terminals became a re-export gateway for supplying Germany after Russian pipeline gas was cut, then:

1. NL→DE and BE→DE pipeline-gas trade value should be higher in the post-period average than the pre-period average (the re-routing signature itself).
2. The mirror pair should reconcile reasonably well: for each pair, in months where both sides report a value, the relative difference |export_value − import_value| / max(export_value, import_value) should mostly stay under 20% — a threshold chosen in advance to allow for the standard Eurostat CIF/FOB valuation gap (imports are valued CIF — cost+insurance+freight — exports FOB — free on board — so some gap is expected by convention, not evidence of error; see Antithesis).

Falsification: if NL→DE and BE→DE pipeline-gas trade does not rise from pre to post period, H2's re-routing claim is falsified. Separately and regardless of that outcome, if the mirror pairs' relative difference average is under 20%, the underlying data is corroborated as internally consistent; if it is not, that is reported as a data-quality finding in its own right, not folded into or used to explain away H2's result either way.

Antithesis — required counter-checks before any result is called a finding

• CIF/FOB valuation convention. Eurostat values all EU imports CIF and all exports FOB, by definition, for every product and every reporter — a systematic reason for import value to run a few percent above the mirroring export value even with perfect, error-free reporting on both sides. The 20% mirror-check threshold above is set well above the typical CIF/FOB gap so it isn't mistaken for a data-quality problem, but any observed gap must be read against this convention before being called a discrepancy.
• Mirror-statistic check: this domain's defining method, not a secondary check — required by the original spec for exactly this domain. Applied to the two available intra-EU pairs (NL→DE, BE→DE); not applicable to any extra-EU leg (Russia, US, Qatar, Norway, Algeria are not in the warehouse as reporters).
• Composition/aggregation effects. LNG and pipeline gas are tracked and reported separately throughout — never summed into one "gas" number before analysis — because the re-routing story specifically depends on the mix shifting from pipeline to LNG-then-repiped, not just on a total gas volume changing.
• Re-export vs. genuine consumption. A rise in NL/BE→DE pipeline gas trade value could reflect NL/BE using Germany as a further pass-through to other markets, not DE consumption — this data alone cannot distinguish "Germany consumes it" from "Germany re-routes it onward again." Flagged, not resolved.
• Price vs. volume. All series here are trade value (EUR), not physical quantity. European gas prices moved enormously across this window (the 2022 price spike, later normalization) independent of any volume/routing change. A value-based composition shift could partly or wholly reflect price effects rather than routing changes. This is a real limitation of using _eur series instead of _kg/_su — declared here in advance: value series are used because the mirror-statistic check in H2 specifically requires matching valuation methodology (CIF/FOB) between the two sides, which is a value-side concept; a quantity-based cross-check is flagged as a natural extension, not done here.
• Base-year changes: not applicable. Comext CN8 codes and EUR valuation do not rebase the way an index-linked series would.
• Chinese New Year timing: not applicable — no Asian Jan/Feb data in this domain.
• FactIQ vintage caveat. Comext data pulled through FactIQ carries no vintage guarantee (Phase 1 finding, reconfirmed by FactIQ directly). Every value used here is FactIQ's current, revised print for that month, not what Eurostat originally published.

Caveats (known before results; will be restated in the results writeup)

• Only 5 extra-EU supplier partners are tracked; any import value from a partner outside {RU, US, QA, NO, DZ} is invisible to H1's share calculation. If a reporter sources meaningfully from elsewhere (e.g. Nigeria, Trinidad), H1's denominator understates true total imports.
• Only 2 mirror pairs are checked (NL→DE, BE→DE) out of many possible intra-EU legs — chosen because they match the specific re-routing story in the original spec, not because they are the only legs that exist.
• Poland's Norway leg (Baltic Pipe) cannot be mirror-checked — Norway is not an EU Comext reporter.
• Value (EUR), not physical volume — see Price vs. volume above.
• Comext's own reporting lag (~2 months per Phase 0) means the most recent 1-2 months in any window may be thin or provisional.
• FactIQ's Comext data is not vintage-aware — see FactIQ vintage caveat above.