Domain E - EU Waste-Oil / Biodiesel Feedstock Trade Transparency Monitor

Domain E — Results: EU Waste-Oil / Biodiesel Feedstock Trade Transparency Monitor

Computed 2026-08-05 against hypothesis.md (commits cd26894, 6ace438, pre-registered before any monthly value was pulled). Every number below is reproducible: python -m src.domains.domain_e_eu_waste_oil_transparency.compute regenerates results.json from the staged fixtures in data/.

Read this before the numbers below: this is a monitor, not a verdict. Nothing here proves mislabeling, fraud, or circumvention. It flags patterns that match the shape of what real investigative and regulatory scrutiny of this sector has looked for. A flag is a prompt to look closer, never a conclusion — see Antithesis for the alternative explanations that must be ruled out before reading any of this as more than that.

H1 — Price-convergence signal: flagged for Spain, Belgium, and the Netherlands

The pre-registered threshold was a waste_price / virgin_price ratio above 0.70, sustained for 3+ consecutive available months.

Spain is the standout. Its "waste" import price from China ran at parity with, and often above, virgin technical-grade palm oil from Indonesia for essentially the entire observed window (mean ratio 1.01 — the waste category was, on average, not discounted at all). The lowest reading in the whole series (0.59, March 2025) is the only material dip; every other month sits at or above the flagged threshold. Belgium and the Netherlands show the same pattern on thinner data. Italy's two series never had overlapping months in this window, so H1 could not be evaluated for it at all — reported as a gap, not a zero.

This is the single most notable result in this domain: genuine waste/residue material is not supposed to price like virgin oil. That it does, consistently, across three independent reporters, is exactly the shape of signal this monitor was built to surface.

H2 — Volume-plausibility signal: consistent across all four reporters

"Waste" (CN8 15180095) import value grew faster than the virgin-oil benchmark in every one of the 4 core reporters, 2023 → 2025:

As declared in advance, this has no fixed pass/fail threshold — it's a comparison, not a test. But the direction is unanimous: in every reporter checked, the "waste" category is growing faster than, or moving opposite to, the legitimate virgin-oil trade it's nominally a byproduct of.

H3 — Mirror-statistic reconciliation: fails everywhere, but differently than gas did

Using quantity (kg), not value, as the reconciliation metric — this avoids currency mismatches, since China reports in USD and the EU reporters in EUR:

Same qualitative finding as Domain B's gas check: China's declared export quantity and each EU reporter's declared import quantity for the same nominal flow do not reconcile. But the pattern is different — gas showed a consistently large gap every month; this shows high variance, swinging between near-perfect matches (Spain: 1.5% in March 2026, 4.1% in May 2025) and near-total mismatches (Spain: 89% in April 2026) in adjacent or nearby months. That variance pattern is more consistent with shipment-timing misalignment (a single large cargo recorded by the exporter in one month and the importer in the next) than with a stable measurement-convention offset — a real, different-shaped finding from Domain B's, not the same one repeated.

H4 — Disclosure map: no anonymization found for this product, but one real absence

Unlike Domain B's gas finding, none of the reporters checked anonymize this specific trade line. Every reporter's series for CN8 1518 00 95 carries real, named partner attribution — including Germany, which suppressed gas partner data completely but discloses this product normally (just at negligible volume, €21,875/month average).

• Core (real, material): Spain, Belgium, Netherlands, Italy.
• Real but smaller-scale, not analyzed further this pass: Portugal (36 obs, €557K/month average), Sweden (8 obs, €3.83M/month average — worth a closer look in a future pass).
• Real but too thin to trend: Germany, France, Ireland, Austria, Lithuania, Denmark.
• Real but anomalous: Romania — only 2 observations, but both very large (€6.9M average) — reads like isolated bulk shipments, not a trend. Flagged, not investigated further.
• No series found at all: Poland — 24 real partner series exist for this exact product (Indonesia, India, Malaysia, Thailand, Vietnam, Russia, the US, and others, plus an anonymized "not specified" bucket), but none for China specifically. This reads as Poland genuinely never recording direct China-origin trade in this code, not suppression — worth noting since it means Poland can't be added to the core set even though it clearly participates in this broader trade.

Ireland (checked at direct request, 2026-08-05)

Confirmed immaterial. 5 observations from China (avg €15,034/month), 1 from India (€50 total). China Customs has no dedicated export series to Ireland at all for this HS code. Ireland's real, indexed trade in biodiesel-adjacent products is intra-EU petroleum blending stock from Spain/Netherlands/France (a different CN8 family entirely), not this waste-oil line. Included in the disclosure map, not the core analysis.

Antithesis — read before treating H1/H2 as more than a flag

• Composition effects, unresolved. CN8 1518 00 95 comingles genuine waste cooking oil, palm fatty acid distillate, and other inedible fat/oil blends of varying legitimate quality and price. A price at parity with virgin oil could mean a higher-value sub-blend within this code, not necessarily virgin oil relabeled as waste. This analysis cannot distinguish those cases — it can only say the average declared price for the whole code doesn't look like cheap waste.
• Benchmark choice matters. Technical-grade crude palm oil (1511 10 10) was chosen as the closer comparator to non-food-grade waste material, but it is still a specific, single benchmark. A different reference (rapeseed oil, tallow, a blended feedstock index) could shift the ratio materially. This should be treated as a first pass, not a settled benchmark choice.
• Real regulatory/market events can produce a similar shape of data. The EU's anti-dumping scrutiny of Chinese biodiesel is real and active over this exact window; front-loading ahead of duties, or genuine price convergence from other causes, is not ruled out by this analysis. No specific EU decision date was sourced with a citation in this pass (attempted via search_news, not found) — none is used here as a reference marker.
• Mirror non-reconciliation is not itself evidence of anything — see H3 above; the variance pattern found here is more consistent with timing misalignment than a Domain-B-style stable measurement gap, and neither indicates wrongdoing on its own.
• CIF/FOB valuation convention contributes some expected gap to both the H1 price comparison and the H3 quantity check's value-adjacent interpretation — though H3 specifically used quantity to avoid this, H1's price ratio is value-based and remains subject to it.
• FactIQ vintage caveat. Every value used here is FactIQ's current, revised print (Phase 1 finding), not what Eurostat/China Customs originally published for that month.

Caveats

• Analysis window is 2023-01 onward, chosen for manageable scope and policy relevance — not the full available history (Spain's series goes back to 2011). A longer-window pass is a natural extension, not done here.
• Indonesia and Malaysia remain invisible as EU-side reporters — this entire analysis is bounded to what China (and, thinly, India) declare.
• H2 has no independently-calibrated numeric threshold, by design — it's a directional comparison, not a pass/fail test.
• Portugal and Sweden showed real, non-trivial data not incorporated into the core H1–H3 analysis this pass — a defensible next expansion.
• This monitor is not yet wired into the daily harvester as of this commit — see the project README for the plan to add it.

Bottom line

• The most notable, best-evidenced finding: "waste" oil from China is priced at or above virgin palm oil in Spain, and similarly in Belgium and the Netherlands — not the discount genuine waste material should carry. This is consistent across independent reporters and a real number of months, not a single outlier.
• Volume in this category is growing faster than the legitimate benchmark trade everywhere checked.
• Declared quantities between China and each EU reporter don't reconcile — differently-shaped from Domain B's gas finding, more consistent with timing noise than a stable bias, but still a real, reportable data-integrity gap.
• No anonymization found for this specific product — a genuine contrast with Domain B's gas finding, and worth stating plainly since it means this monitor's numbers, unlike much of the gas story, are not blocked by suppressed data.
• None of this proves anything about any specific shipment or company. It says the aggregate trade pattern in this sector, over this window, looks like the pattern real scrutiny of this sector has flagged before — which is exactly what a transparency monitor is for.

Pre-registered hypothesis

Domain E — EU Waste-Oil / Biodiesel Feedstock Trade Transparency Monitor

Status: PRE-REGISTERED, 2026-08-05. No monthly price/volume panel has been pulled at the time this file is committed — only series existence, reporter coverage, and magnitude checks (metadata, not results). Per project principle, this hypothesis will not be edited after results are computed; if a check finds nothing notable, that is recorded as the result, not adjusted.

Purpose and framing — read this before the method

This is explicitly not a trading domain, unlike A–D. No output of this domain will ever claim to prove fraud, mislabeling, or circumvention. The purpose is narrower and different: build a reproducible, ongoing monitor of publicly-derived trade patterns in a sector (EU imports of "waste"-classified fats/oils and biodiesel from Asia) that has already drawn real regulatory and investigative scrutiny, using the same quantitative techniques that scrutiny has used — price-benchmark comparison, volume-plausibility comparison, mirror-statistic reconciliation, and disclosure-transparency mapping — and to say plainly, every time, what the data does and does not show. A flagged pattern is a prompt for a human to look further, never a conclusion.

Because this is a monitor, not a one-off backtest, it will be wired into the existing daily harvester (src/harvest/) after this initial pass, so the same pre-declared method re-applies to new data as it arrives — future monthly updates are not new hypotheses, they are the same method run again.

Data sources (confirmed present as of 2026-08-05)

"Waste"/residue proxy — FactIQ eu_comext_<iso2> schemas: CN8 1518 00 95 — "Inedible mixtures or preparations of animal fats and oils and their fractions, whether or not containing vegetable or microbial fats or oils." This is the specific line real-world reporting on suspected mislabeled UCO/palm-residue imports has used. Value (_eur) and quantity (_kg) both available.

Biodiesel — same schemas: CN8 3826 00 10 (pure FAME, ≥96.5% ester content) and 3826 00 90 (biodiesel mixtures, <70% petroleum content). Value and quantity both available.

Virgin-oil benchmark — same schemas, from Indonesia: CN8 1511 10 10 (crude palm oil, technical/industrial use — the closer comparator, since "waste" material is itself non-food-grade) and 1511 10 90 (crude palm oil, standard/food-adjacent grade — secondary comparator). Confirmed present for the Netherlands, 2002–2025/2026, value and quantity both available.

Reporters — confirmed via direct coverage check (COUNT/AVG/date range on each reporter's own data_points), not assumed:

Core reporter set for H1–H3: ES, BE, NL, IT. DE, FR, and IE are carried into H4 (the disclosure map) only, with their thinness stated as the finding, not hidden.

Partners: China (cn) primary; India (in) secondary, checked but not yet quantified. Indonesia and Malaysia have no FactIQ trade schema of their own (Phase 0 finding, still true) and cannot be observed directly as EU-side partners — but China's own customs data discloses partner country by name for this exact HS code, including Indonesia, on both its import and export sides, which is the closest available window onto that leg of the chain.

Mirror-statistic data — FactIQ china_customs schema: China's HS8 15180000 export series are partner-disaggregated. Confirmed exact series IDs exist for exports to Spain, Netherlands, Belgium, Italy, Germany, France, Poland, and Lithuania (value in USD, plus a _qty companion in kg), e.g. china_customs_X_8d_15180000_312 = China's exports to Spain.

H1 — Price-convergence signal

Method: for each core reporter, compute implied unit price (value ÷ quantity, €/kg) monthly for (a) CN8 1518 00 95 from China and (b) CN8 1511 10 10 from Indonesia (the technical-grade virgin benchmark, same reporter). Compute the ratio waste_price / virgin_price each month.

Pre-declared monitoring threshold (not a classical falsification criterion, since this is a monitor, not a backtest — but declared now, in writing, before any monthly value is pulled, for the same no-p-hacking reason): genuine waste/residue material is expected to trade at a real discount to virgin oil. A ratio above 0.70, sustained for 3 or more consecutive months, is flagged as notable in this monitor — worth a human's attention, not proof of anything. This threshold is informed by publicly documented historical UCO-to-virgin-oil discount ranges (typically 20–40% below virgin vegetable oil); it is a round, conservative number chosen in advance, not fitted to this data.

H2 — Volume-plausibility signal

Method: for each core reporter, compute the year-over-year growth rate of CN8 1518 00 95 import value from China, and separately the year-over-year growth rate of CN8 1511 10 10/90 (virgin palm oil) and CN8 3826 00 10/90 (biodiesel) import value from the same or comparable partners, over the same window. Flag: any period where the "waste" category's growth rate exceeds every one of these comparators by a wide margin — no fixed numeric threshold is pre-declared here, since there is no independently verified "plausible waste generation" series in FactIQ to calibrate one against; this leg is reported as a relative comparison chart, not a pass/fail test, and that limitation is stated here in advance rather than papered over with a fabricated threshold.

H3 — Mirror-statistic reconciliation

Method: identical to Domain B's approach. For each core reporter, compare China's declared export value/quantity (china_customs_X_8d_15180000_<partner>) against that reporter's declared import value/quantity (eu_comext_M_<reporter>_cn_te_p1_cn8_15180095_<eur|kg>) for the same nominal flow, month by month where both sides report. Relative difference = |export − import| / max(export, import). Domain B found this method fails to reconcile for gas (69.7% average gap) — this domain re-runs the same test on a structurally different commodity (bulk liquid moved by tanker, not a piped grid commodity), with no assumption about whether it will reconcile better or worse.

H4 — Disclosure/anonymization map

Method: repeat Domain B's anonymization-discovery method (scan the dimensions table for every partner code attached to the relevant import series) across a broader reporter set: the 4 core reporters plus DE, FR (already found thin, checked for anonymization status regardless), and PL, LT, PT, RO, DK, SE, AT — for both CN8 1518 00 95 and 3826 00 10/90. Output is a standing table: which reporters disclose real partner data for this trade, which anonymize, refreshed on every harvest. A reporter that changes status between harvests is itself the notable event.

Antithesis — required counter-checks before any pattern is called notable

• HS 1518 is not UCO/POME-exclusive. It comingles genuine waste cooking oil, palm fatty acid distillate, other inedible fat/oil blends, and legitimate technical-use products. A volume or price signal in this code is a prompt to look closer, not identification of any specific product.
• Real regulatory and market events can produce the same shape of data as the pattern this monitor watches for. The EU has active anti-dumping scrutiny of Chinese biodiesel; front-loading ahead of duties, or genuine trade diversion after them, can look similar to a mislabeling signal in aggregate trade data. Any exact EU decision date needs sourcing with a citation before being used as a reference marker on a chart — not asserted from memory. (Not sourced as of this commit; to be added only if found with a citable source during the fetch phase.)
• A real, sourced, forward-looking confound exists independent of any fraud story: Indonesia launched a 50% domestic palm-biodiesel blend mandate ("B50") on 2026-07-09 (Bloomberg, via search_news), which will divert palm oil toward domestic use and reduce export availability going forward. A future decline in Indonesia-linked feedstock volume, or a future price rise in genuine palm oil, should be read against this policy first, not assumed to indicate anything about the waste-oil question.
• Mirror-statistic non-reconciliation, if found, is not itself proof of fraud — Domain B already showed a legitimate commodity (gas) can fail to reconcile for structural/measurement reasons unrelated to any wrongdoing. The same caution applies here by default.
• CIF/FOB valuation convention applies here exactly as in Domain B — a stable, expected, non-suspicious source of some mirror-check gap.
• Composition/grade effects. CN8 1511 10 10 (technical-grade) and 1511 10 90 (food-grade) crude palm oil carry different legitimate price levels; the benchmark choice must be stated on every chart, and both should be shown if they diverge meaningfully.
• FactIQ vintage caveat. Comext and China Customs data pulled through FactIQ carry no vintage guarantee (Phase 1 finding). Every value used here is FactIQ's current, revised print, not what was originally published for that month.

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

• Indonesia and Malaysia — the two largest actual source countries for genuine palm residue — have no FactIQ trade schema. Only China's and India's declared trade is directly visible; Indonesia is visible only as a named partner inside China's own customs data, one step removed.
• Germany and France were checked and found unusable for this specific trade line (immaterial volume / too sparse) — the core analysis rests on 4 reporters (ES, BE, NL, IT), not all of the EU.
• H2 has no independently-calibrated numeric threshold — it is a comparison chart, not a pass/fail test, by design and stated here in advance.
• No FactIQ dataset can verify actual UCO/waste-oil production capacity in China or India — any "implausible volume" read is relative to other trade categories, never an absolute, independently-verified ceiling.
• This monitor sits inside a live, evolving policy environment (EU anti-dumping scrutiny, Indonesia's new B50 mandate) — every future reading should be interpreted against contemporaneous policy context, not treated as a clean, isolated signal.
• FactIQ's trade data is not vintage-aware — see FactIQ vintage caveat above.