The Appliance Layer of Oil: Crack Spreads, Shadow Fleets, and the Aschenbrenner Trade

2026-05-19 · 21 min read · 5202 words

Field statement. When a utility-layer chokepoint is disrupted, the appliance layer captures the margin that would otherwise have accrued to the utility's owner. This is true of electricity (Westinghouse and Tesla absorbed the spread that Edison tried to keep). It is true of oil in April 2026 (Indian and Türkish refineries absorbed the EUR-billion-per-month margin that Western sanctions tried to deny Russian crude). And it is, the Aschenbrenner 13F argues at USD 13.7 bn AUM, true of AI in 2026–2030: the chip utility is being shorted, the power-and-deployment appliance is being bought.

The Mercantile Thesis essay published on this blog on 6 May 2026¹ argued that today's market overprices the utility layer of AI (foundation-model labs) and underprices the appliance layer (sovereign deployment, multi-agent orchestration, hardware-native runtime). That argument used the canonical 1880s electrical precedent: Edison's DC kit empire losing to Westinghouse and Tesla's AC substrate. The argument was structural and used 150-year-old data because the AI appliance layer is six months old.

This essay does the same audit against a system that has 150 years of its own data, a real-time natural experiment running, and a single sophisticated counterparty putting USD 13.7 bn on the trade in a publicly filed 13F. The system is oil. The natural experiment is the 2026 Iran war and the Hormuz blockade. The counterparty is Leopold Aschenbrenner's Situational Awareness LP.

The merchant question is the same one the Mercantile Thesis asks of AI: what is the flow, where is the bottleneck, who owns the bottleneck, who is taxing the spread, and what is the durable position? The oil answer, in April 2026, is sharp enough to read with the lights off. I'll work it through, and then return to AI with whatever the oil case forces us to update.

I. Refining is the canonical appliance layer

Crude oil is the utility input. It is fungible (a barrel of Urals and a barrel of Arab Light are physically interchangeable for most refining configurations), metered (loaded by the tonne at port terminals and tracked by tanker AIS pings), and transported as a bulk commodity (VLCCs at roughly 2 million-barrel capacity per hull). The utility-layer value of crude is the regional price differential (Brent vs WTI vs Dubai vs Urals), which is what commodity-trading houses spend their lives arbitraging.

Refined products (gasoline, diesel, jet, naphtha, petrochemical feedstock) are the appliance loads. They are load-shaped, in the literal sense: the refinery's distillation columns and crackers transform the crude utility into specific molecules customers buy. Each refined product carries its own price and its own end-use market. The customer of the diesel does not interact with the crude; the customer of the petrochemical feedstock does not interact with the diesel. The refinery is the appliance that reshapes the bulk utility into the specific loads.

The crack spread is the appliance-level value capture: the price differential between a barrel of crude (utility input) and the basket of refined products (appliance output) it yields. The 3-2-1 crack (three barrels of crude in, two barrels of gasoline plus one barrel of distillate, diesel and heating oil, out) is the standard benchmark. When the crack spread widens, refineries make money even if crude prices fall. When it narrows, refineries lose money even if crude prices rise. The crack spread is the structural margin of the appliance layer; the crude price is the variable cost of the utility input.

This is the cleanest pre-electrical instance of the topology that the Mercantile Thesis applies to AI. The merchant principle reads refining as a tollbooth-class position in the oil flow graph². A refinery sits between the upstream utility (crude flow from sovereign producers) and the downstream demand (specific load-shaped products to end consumers). Whoever owns the refinery captures the crack-spread margin without owning the underlying crude or the downstream market. Reliance's Jamnagar complex (~1.4 million bpd nameplate, the world's largest single-site refinery) and Nayara Energy's Vadinar refinery (~400 000 bpd) gate most of India's seaborne-crude refining capacity. Their throughput decisions move the regional crack spread. They are the appliance layer.

The historical precedent is Standard Oil. Rockefeller did not own Pennsylvania crude; he owned the refineries that turned crude into the standardized, brand-marked product the lighting market would pay for, plus the rail rebates and pipeline infrastructure that gated the flow into and out of those refineries³. The refining-as-bottleneck reading is not new. What is new is that the 2026 sanctions architecture has converted Indian and Türkish refineries into the same structural position Standard Oil occupied in the 1880s: the tollbooth at which the geopolitical origin of crude becomes negotiable. The refinery is now the origin-laundering appliance, and the crack spread now carries a sanctions-arbitrage premium on top of its traditional refining margin.

II. The April 2026 CREA data is the empirical signature

The Centre for Research on Energy and Clean Air publishes a monthly Russia Fossil Fuel Tracker that combines Kpler tanker-tracking data with ship-to-ship transfer records and refinery throughput estimates. The April 2026 report is the cleanest single-month primary-source documentation of the appliance-layer-of-oil reading.⁴ The headline figures:

• Shadow fleet share of Russian seaborne crude: 54% of Russian seaborne oil was transported by sanctioned "shadow" tankers in April 2026, up from 48% in March 2026, the highest measured share on record. A further 38% used G7+ tankers; the remainder used non-sanctioned shadow vessels.
• Russian fossil-fuel export revenue: EUR 734 million per day, the highest daily revenue rate in two and a half years, up 4% month-on-month despite sanctions and a price-cap regime nominally targeting the same flows.
• Refined-product re-export laundering loop: refineries using Russian crude in India, Türkiye, Brunei, and Georgia exported EUR 760 million of oil products to sanctioning countries in April 2026, of which approximately EUR 232 million was derived from Russian-origin crude. India-origin refined-product exports to the EU specifically totaled EUR 145 million for the month.
• The top three buyers (China, India, Türkiye) accounted for 93% of Russian crude exports. China took EUR 7.3 bn (41% of top-five importers), India EUR 5.0 bn (28%), Türkiye EUR 3.0 bn (17%).
• 47 shadow vessels operating under false flags at month-end. Five vessels delivered EUR 236 million of Russian hydrocarbons under false flags in April alone.

The structural reading is in the gap between two of those numbers. The G7 sanctions regime, the EU import ban on direct Russian oil, and the USD 60/barrel price cap collectively destroyed the utility-layer value of Russian crude for direct G7 sales. And yet Russian fossil-fuel export revenue is at a 2.5-year high. The crude reaches the end consumer; it just transits through Indian, Türkish, and South-East-Asian refineries first. The refinery is the appliance layer; the crack spread is its margin; the sanctions architecture is the friction that fattens the spread for the appliance owner.

The Iran-side mirror of the same loop runs through Chinese teapot refineries with crude relabeled as Malaysian / Indonesian / Omani via ship-to-ship transfers off Malaysia. The Columbia University Center on Global Energy Policy documents the structural parallel from 2025 data: China imported approximately 1.3 million barrels per day of "Malaysian" crude (more than twice Malaysia's own production at ~535 000 bpd in 2024) and Indonesian exports to China surged from under 3 000 bpd to 291 000 bpd despite Indonesia being a net oil importer.⁵ The Iranian and Venezuelan crude is rebranded at the ship-to-ship transfer point, then ingested by Chinese teapot refineries that are more risk-tolerant than state-owned enterprises and dependent on the sanctioned-crude discount.

Two laundering loops, mirror-symmetric. Russian crude via India to the EU; Iranian / Venezuelan crude via Chinese teapots. Both running at a measured intensity that puts the appliance layer of the global oil system in clearer focus than at any time since the Marc Rich era of the 1970s and 1980s⁶. The refinery is the legible boundary at which provenance becomes negotiable. The crack spread is the margin the appliance layer captures for performing the trick.

III. The 2026 Iran war is the natural experiment

On 28 February 2026 the United States and Israel launched an air war against Iran. The opening campaign included the targeted assassination of Iran's supreme leader. On 4 March 2026 Iran closed the Strait of Hormuz to commercial shipping. On 13 April 2026 the United States launched a counter-blockade of the strait. Brent crude surged past USD 120 per barrel. The International Energy Agency characterized the disruption as the largest supply disruption in the history of the global oil market.⁹

This is the natural experiment the merchant lens needed. The Hormuz chokepoint moves approximately 20% of global seaborne crude on a normal day. A disruption of that magnitude is the canonical utility-layer shock. The first-order prediction from any non-merchant frame is that downstream refined-product prices spike, the end consumer pays the disruption tax, and the system rebalances slowly and painfully over months.

What actually happened is the appliance-layer prediction. Brent spiked, yes, but Indian, Türkish, Chinese, and South-East-Asian refineries kept running. Crack spreads widened materially for refiners on the receiving end. Aggregate refined-product flow to the end consumer was substantively preserved. The refined-product re-export laundering loops kept operating; some of them (the Indian-to-EU loop in particular) ran harder during the disruption because the price-arbitrage opportunity widened.

Iran's own export-revenue capacity was materially degraded for a window of weeks (China-Iran total trade fell by reported figures in the high-double-digit-percent range January–March 2026; the oil-only figure is harder to pin from public reporting because shadow-fleet flows are deliberately illegible). The Iranian utility layer (the crude flow at the chokepoint) was disrupted at the regulatory level. The appliance layer (shadow-fleet logistics, third-country refineries, teapot ingestion in China, refined-product re-export) captured the margin that would otherwise have accrued to Iran's national oil company. China continued to import Iranian crude at approximately 1.4 million barrels per day on shadow tankers through a direct diplomatic carve-out, but the margin on those flows now sits with the refiners and the trading houses, not with the producer.

This is the structural pattern. Utility-layer disruption produces appliance-layer capture. The merchant principle has been right about this since the 1973 OPEC embargo (Marc Rich made his career on it⁷) and the 1980–88 Iran-Iraq tanker war (Vitol, Trafigura, and Glencore's institutional ancestors made theirs in the same way⁸). The 2026 Iran war is the same pattern at modern scale, with cleaner primary data than any prior disruption because of Kpler tanker-tracking, CREA monthly reporting, and the SEC 13F filing infrastructure that lets us watch a sophisticated counterparty position around the same thesis in real time.

Which brings us to the counterparty.

IV. The Aschenbrenner trade is the AI version

Leopold Aschenbrenner published Situational Awareness: The Decade Ahead in June 2024 as a 165-page self-published essay arguing that trillion-dollar compute clusters, AGI by 2027, and the national-security implications of frontier AI are the dominant geopolitical and economic forces of the next decade.¹⁰ The third chapter of that essay is the energy chapter, and it argues that power is the binding constraint: chip production can scale faster than electricity generation, and the next-leg AI bottleneck migrates from "we don't have enough chips" (the bottleneck of 2023–2025) to "we don't have enough power" (the bottleneck of 2026–2030).

In late 2024 Aschenbrenner founded Situational Awareness LP as the operational expression of the essay's thesis. The fund is co-managed with Carl Shulman, seed-backed by Patrick and John Collison, Daniel Gross, and Nat Friedman. The investment strategy is the picks-and-shovels trade of artificial intelligence, the physical things AI compute cannot run without.

The fund's Q1 2026 13F-HR filing (period 31 March 2026, filed 18 May 2026, SEC CIK 0002045724) is the canonical primary-source artifact of the bottleneck-shift trade. The headline figures, in USD millions of market value at the close of the quarter:¹¹

| Category | Position | Size (USD mn) | Direction | |-—|-—|-—|-—| | Chip utility: SHORT | SMH puts | 2,040 | put | | Chip utility: SHORT | NVDA puts | 1,570 | put | | Chip utility: SHORT | ORCL puts | 1,070 | put | | Chip utility: SHORT | AVGO puts | 1,010 | put | | Chip utility: SHORT | MU puts | 422 | put | | Chip utility: SHORT | INTC puts | 159 | put | | Chip utility: aggregate notional short | | ~6,271 | | | Power appliance: LONG | BE (Bloom Energy) | 879 | long equity | | Power appliance: LONG | IREN Limited | 401 | long equity | | Power appliance: LONG | CLSK (CleanSpark) | 104 | long equity | | Power appliance: aggregate long | | ~1,384 | | | Foundry-survivor: LONG | TSM calls | 355 | call |

The portfolio is USD 13.68 bn total across 42 positions. It approximately doubled in one quarter; the prior 13F covering 31 December 2025 reported USD 5.52 bn AUM across 29 positions. The pivot toward concentrated chip-sector puts is the new positioning, not a continuation.

The trade is single-thesis. It is not diversified, not opportunistic, not multi-strat. It is one structural claim (the binding constraint on AI scaling moves from chips to power between 2026 and 2030) expressed at USD 13.7 bn AUM through six put positions on the major US-listed semiconductor names and three long positions on distributed-power and miner-pivoting-to-AI-hosting names. Bloom Energy's USD 879 million long is the trade in clearest form: Bloom sells distributed solid-oxide fuel cells, the canonical behind-the-meter power solution for data centers that cannot wait for grid interconnection. If the bottleneck is power, Bloom is the appliance vendor.

The merchant-principle reading: Aschenbrenner is not taking a bottleneck position himself. He is taking a positional bet on which layer of the stack the bottleneck migrates to. The chip-sector puts express the view that the chip utility commoditizes. The Bloom / IREN / CLSK longs express the view that the power-and-data-center appliance captures the next-leg margin. The trade is the AI version of the oil-refining trade.

That is what makes the 13F load-bearing. It is the cleanest 2026 public-record market-priced version of the bottleneck-shift reading the Mercantile Thesis develops. Where the Mercantile Thesis argues the utility-vs-appliance asymmetry from structural first principles, the 13F is the empirical evidence that a sophisticated counterparty with USD 13.7 bn of AUM has reached the same conclusion and is willing to express it at scale.

The signal here is the structure of the trade more than its directional content. A long-Bloom or short-NVDA position by itself is a directional call any number of investors are taking on either side. The structure (chip-utility short paired against power-appliance long, sized at multi-billion notional, in a fund whose published thesis is the energy-bottleneck argument) is the merchant version. Position size, position concentration, and the explicit thesis-paper that pre-committed the reasoning. The trade is legible in a way most macro positioning is not.

V. What the oil case forces us to update on AI

The Mercantile Thesis treated the AI utility-and-appliance reading as a structural prediction with weak short-horizon empirical support: six months of open-weight model releases versus 150 years of electrical-distribution history. The oil case adds an empirical bridge. The same topology produces the same signature in a system with 150 years of its own data. That is not proof that AI will rhyme. It is evidence that the topology is general.

Four updates the oil case forces on the AI reading:

Update 1: The appliance layer captures during utility disruption, not despite it. The Mercantile Thesis was largely written before the April 2026 CREA data and the Iran-war disruption produced their clear signal. The signal is that disruption to the utility layer fattens the appliance layer's margin. Applied to AI: a chip-supply disruption (export controls, TSMC capacity constraint, a Taiwan crisis) should increase the margin available to the appliance layer (vertically integrated inference stacks, deployment infrastructure, behind-the-meter compute), not decrease it. The Aschenbrenner trade is positioned for exactly this: the chip puts pay off in a chip-supply shock; the power-infrastructure longs benefit from the bottleneck migration that shock would accelerate.

Update 2: Origin-laundering is a real appliance-layer business model. I had not previously written about this. The Indian / Türkish refining loop is a USD-billion-per-month business of converting politically inconvenient utility input (Russian crude) into politically legible appliance output (Indian-origin diesel). The AI analogue is the inference-laundering business that emerges when foundation-model weights become geopolitically segregated: PRC-origin weights running through US-jurisdiction inference infrastructure with no model-card disclosure to the end consumer; or US-origin weights running through EU-jurisdiction inference for AI Act compliance. The structural opportunity for an origin-laundering inference appliance is the analogue of the Indian-refinery business. I have not seen this written about, and the Mercantile Thesis canon should add it as a vector.

Update 3: The natural experiment matters more than the structural argument. I spent the Mercantile Thesis arguing the appliance-layer claim from structural first principles. The April 2026 CREA data and the Iran-war disruption did more for the argument than another 3 000 words of structural reasoning could have. The lesson is that the canon should be primed to absorb natural experiments when they occur. If TSMC capacity becomes constrained, if a Taiwan crisis disrupts chip flow, if the EU AI Act produces a forced segregation event, the canon should be ready to read those events through the appliance-layer frame in real time and update.

Update 4: A counterparty taking the trade at scale is itself evidence. Aschenbrenner is not a price-taker. At USD 13.7 bn AUM expressed as a single-thesis trade, his position itself moves the price of the names involved, and his public 13F is read by the market as a signal. The merchant-principle reading does not give weight to crowd consensus; it gives weight to high-information counterparties expressing high-conviction positions through capital allocation. The Aschenbrenner 13F qualifies. The Mercantile Thesis canon should add 13F-tracking of high-information AI-energy-positioned funds as a standing reference signal.

VI. The Cynic's Audit

"You're reading too much into a single 13F. Aschenbrenner could be wrong."

Type I (overclaim) concession. Yes, he could. The argument of this essay is not that the chip-sector reprices on Aschenbrenner's timeline; the argument is that the structural reading is shared by at least one high-information counterparty with USD 13.7 bn of capital expressing it at scale. That is a different claim from "the trade works." The Mercantile-Thesis argument is structural over a ten-to-fifteen-year horizon. The Aschenbrenner trade is positioned for that horizon, but options have expiry dates and the put positions could be repriced or expire worthless before the structural thesis plays out. If the chip-sector puts roll off in 2027 at a loss because the bottleneck migration is slower than expected, that does not refute the structural reading; it only refutes the timing of the trade. The Mercantile Thesis explicitly states that the merchant lens is structural-not-tactical and gives no guidance on which month the repricing happens.

"The Indian refinery laundering loop is just sanctions arbitrage. It collapses the moment sanctions are lifted. You're calling a temporary arbitrage a structural appliance position."

Type II (missed-risk) concession. The objection is partly correct. The sanctions-arbitrage premium on the Indian refining loop is conjunctural and will compress as soon as the EU and G7 either lift sanctions or close the refinery-laundering loophole. Either is possible. The appliance-layer argument does not depend on the arbitrage premium persisting; it depends on the refinery itself being the appliance layer of the oil flow graph. That is true with or without the sanctions premium. The historical record (Standard Oil 1880s, Marc Rich 1970s, BlasFarchy-documented trading houses 2000s) confirms the refining-as-bottleneck position is structural across the sanctions cycle. The 2026 sanctions premium is a bonus on the structural position, not the structural position itself. If the EU closes the loophole tomorrow, the Indian refining business is still the Indian refining business; the crack spread is still the appliance margin.

"The Iran-China-US-shale leverage narrative in the popular discourse is not the appliance-layer story you're telling. You're cherry-picking the data."

Both Type I and Type II. This deserves a direct answer. The popular discourse (including the Grok-thread voice that originally motivated this essay) has framed the 2026 ME tensions as US shale leverage to force China onto US-origin oil and capture the crack spread on Texas crude. The Columbia CGEP 2025 data does not support that framing. US crude imports to China fell 76% in 2025. China is decoupling from US oil, not being routed onto it. The US-shale-leverage reading is a popular-discourse overlay that the primary data does not confirm. The appliance-layer reading I have run here is what the data actually shows: China is buying Iranian and Venezuelan crude via teapot refineries; India is buying Russian crude via Reliance and Nayara; the appliance layer (refining, shadow fleet, ship-to-ship transfer) is capturing the margin; the utility layer (sovereign producer revenue) is degraded. I have flagged this as a Type-1 catch on the popular discourse in the experiment register; readers should check the CGEP report directly and not take my characterization on trust.¹²

VII. Honest limitations

Five limitations the essay does not pretend to have resolved:

1. The CREA April 2026 figures are the most recent monthly snapshot available at writing time. The shadow-fleet share, the refinery-laundering volumes, and the Russian export-revenue figures may revise materially in subsequent months as Kpler updates and CREA re-runs its methodology. The structural reading should hold across reasonable revisions; the specific numbers may not.
2. The Aschenbrenner trade may unwind before it is judged. Q1 2026 13F filings show the position at quarter-end. The Q2 2026 13F (covering 30 June 2026, filed ≈ 14 August 2026) may show the chip-sector puts already reduced or rolled. A short-cycle reversal would not refute the structural reading but would weaken the the-trade-is-on-the-tape element of this essay.
3. The Iran war is ongoing at writing time. The resolution path (negotiated settlement, escalation, regime change, prolonged blockade) materially changes the natural-experiment reading. The current essay reads the disruption through April 2026 data; the May–December 2026 window may produce data that revises the reading.
4. The popular-discourse Type-1 catch on US-shale-leverage is my reading of the primary data; it is contestable. The CGEP report is one primary source; the China Select Committee "Crude Intentions" report (House of Representatives, 2026) reads the same data with a different policy frame. A reader committed to the US-leverage reading can argue that the 76% decline in US-to-China crude is itself the leverage being exercised. I find this less convincing than the appliance-layer-capture reading, but it is not refuted by the data I have surfaced.
5. The AI-side analogue (inference-laundering as appliance business model) is speculative. I have flagged it as Update 2 above, but I do not have an example to point to. The structural prediction is that if PRC-origin and US-origin AI weights become geopolitically segregated under export-control regimes, then an inference-laundering appliance business emerges in third-country jurisdictions. The prediction is falsifiable in 2027–2028 if and only if the segregation event occurs.

VIII. Falsifiable bets

Three predictions with timestamps and falsification criteria, matching the format of the Mercantile Thesis falsifiable-bets section.

Bet 1 (Q4 2026, high confidence). Russian shadow-fleet share of seaborne crude exports remains above 50% in every monthly CREA tracker through December 2026, AND aggregate India-origin refined-product exports to the EU exceed EUR 100 million in every monthly tracker through the same window. Falsification: if any single monthly CREA report between May and December 2026 shows shadow-fleet share below 50% OR India-to-EU refined-product exports below EUR 100 million, the structural-persistence reading of the appliance-layer-capture pattern is wrong on its stated timeline.

Bet 2 (Q4 2027, medium confidence). Situational Awareness LP's chip-sector aggregate put exposure (measured as the sum of put-position market values on NVDA / SMH / AVGO / ORCL / MU / INTC across the four 2027 quarterly 13F filings) remains net negative; i.e., the fund continues to hold a meaningful net-short position on the chip utility layer through 2027. Falsification: if the position is unwound to neutral or net long by any 2027 13F filing, the thesis-persistence reading of the Aschenbrenner trade is weaker than this essay claims. Aschenbrenner himself rotating out of the trade is the strongest possible signal that the structural reading was wrong.

Bet 3 (Q4 2028, lower confidence). At least one named inference-laundering business (a third-country inference provider that hosts geopolitically-segregated weights for end users in jurisdictions where direct use of those weights would be regulatorily inconvenient) operates at meaningful scale (USD 50 million annualized run-rate revenue or higher) by 31 December 2028. Falsification: if no such business exists or is identifiable in public reporting by that date, the AI-side analogue of the refining-laundering loop has not materialized on the stated timeline. This bet is the lower-confidence speculative one because the precondition (export-control-driven weight segregation) is itself uncertain.

IX. What this is and what it is not

This essay is the Mercantile Thesis canon's first natural-experiment confirmation case. The Mercantile Thesis argued the utility-and-appliance frame from structural first principles using the 1880s electrical precedent. This essay tests the same frame against the 2026 oil system and finds the signal stronger than the Mercantile Thesis argued. The Aschenbrenner 13F is the bridge: a USD 13.7 bn capital expression of the same reading, applied to the AI stack the Mercantile Thesis was originally about.

It is not a trade recommendation. The merchant lens predicts structural positions, not entry points; the 13F is evidence the structural reading is being underwritten at scale, not evidence the trade is timing-right today.

It is not a refutation of the geopolitical-discourse framing. The popular-discourse reading of the 2026 Iran war (US-shale-leverage on China via crack spreads on Texas crude) is not confirmed by the primary data I have surfaced; the data confirms a different reading (appliance-layer-capture by Indian / Chinese / Türkish refiners and the trading houses behind them). Both readings can be partially correct; the appliance-layer reading is the one with the stronger primary-source case.

It is the canon's first essay that treats a single 13F filing as load-bearing primary-source evidence. The discipline that requires is 13F-tracking as a standing reference signal for high-information AI-energy-positioned funds. The Codex now carries the Aschenbrenner profile and the Situational Awareness LP filing record; future Mercantile Thesis canon updates should cite the Q2 2026 and subsequent filings as they appear.

Receipts at the time of writing

The load-bearing claims of this essay and the primary sources backing them:

• CREA Russia Fossil Fuel Tracker, April 2026 monthly report: shadow-fleet 54%, refined-product re-export EUR 760 mn, India-to-EU EUR 145 mn, Russian revenue EUR 734 mn/day, top-three buyer concentration 93%. Published May 2026 at energyandcleanair.org.
• SEC EDGAR, Situational Awareness LP, 13F-HR Q1 2026 (period 31 March 2026, filed 18 May 2026, CIK 0002045724): USD 13.68 bn AUM, 42 positions, USD ~6.27 bn aggregate chip-sector put notional, USD ~1.38 bn power-appliance long. Available via SEC EDGAR; parsed at 13f.info and last10k.com.
• Columbia University Center on Global Energy Policy, "Where China Gets Its Oil" (2026): China imported 1.38 mn bpd Iranian crude in 2025, "Malaysian" crude rebranding loop (1.3 mn bpd vs 535 000 bpd Malaysian production), Indonesian-export-surge anomaly.
• 2026 Strait of Hormuz crisis primary sources: Wikipedia compilation with citations to US Congressional Research Service R45281, UK House of Commons Library CBP-10636, Al Jazeera 24 April 2026 coverage, American Action Forum analysis.
• Aschenbrenner, *Situational Awareness: The Decade Ahead* (June 2024): situational-awareness.ai, Chapter III (power as binding constraint) is the thesis-paper underwriting the trade.
• Pre-registered experiment registers (stax-experiment lane qm-oil-appliance): four registers pre-registered, three verdicts logged: Aschenbrenner 13F confirmed, shadow-fleet/India-laundering confirmed, Iran-China-crack-spread inconclusive with explicit Type-1 catch on the popular-discourse US-shale-leverage framing.

The receipts are the audit trail. If the essay's reading is wrong, the path to refuting it runs through these specific primary sources, not through opinion.