Oil & the Two Wars

Where it stands — 14 readings against the two cases

So what — which case is materializing — each tile is one reading against author-set thresholds: red sits in the risk case, green in the recovery case, amber between. The round-two readings come first — the gas winter, the tanker war's tempo, the food bill, the political clock — and the round-one instruments the fork was defined on follow.

Thresholds are author-set and stated on every tile as base … · risk …; each tile links its source and, where the reading lives on another tab, that panel. Where a series is already on this page the tile draws its recent path through the zones (Brent daily; SPR and Cushing weekly; European gas and the food index monthly; storage and the attack tempo from their keyed rows); otherwise a level line marks the latest print. The round-two thresholds: European gas risk above $18/MMBtu and base below $12; storage risk below 70% full and base above 80%, against the 1 November norm; attack tempo risk at five or more ships fired on, seized or sunk in fourteen days and base at none; food risk above 130 on FAO's index and base below 125; the deadline tile turns red inside a week.

$71 → $138 → $96: Brent & WTI through the war (USD/bbl)

So what — two legs, and no round trip — the first spike broke on the expectation of reopening rather than on barrels arriving; the second began when that expectation died. Brent closed below its pre-war $71.32 on six sessions (Jun 26 – Jul 6, low $68.53 on Jul 2) and has not been below it since.

FRED daily spot (DCOILBRENTEU / DCOILWTICO), refreshed 2026-09-04; the series lags a few business days (last print Sep 1, $96.02). Shaded bands = war phases; numbered markers = the events tabulated on the Prices & futures tab, with the second leg's arithmetic.

The next thirty days, with this site's number on each

Ledger — expectations frozen when set — each carries the read this site already makes elsewhere — the gate nowcast, the SPR paces, the arrival projections, the strip's implied odds — stated once and graded here when the print arrives.

How to readOpen rows show the expectation and its band; landed rows show the print beside it with a grade. Bands are each model's own backtested or historical width, not chosen. Schedule rows carry dates only. A green chip means the print landed inside the band; amber, outside but on the side called; red, outside; grey, no expectation or the source moved its baseline. Hover a row for the expectation in full, how the number was made, and its source.

Six editions of the official supply forecast against what settled — and this site's read of August (mb/d)

Nothing here forecasts the strait. September's supply row is two branches for exactly that reason, and the market row is the strip's opinion, not this site's.

H2 2026–2027: the market vs the forecasters (Brent, USD/bbl)

So what — the market disagrees — the official forecast is conditioned on the war ending; the strip is not, and the gap between them is the price of that disagreement. The base case (EIA, de-escalation-conditioned) glides back toward $70 and the deferred glut once the risk premium unwinds; the risk case (scenario A) sends winter product markets into a depleted buffer — ~13 months of endurance now vs ~19.5 in February, so the same shock produces a bigger price response. All four futures are quantified on the Scenarios tab, including a 24-month chapter-by-chapter timeline.

How to readSolid is observed daily Brent; every dashed path is a forecast — the EIA line plots quarterly midpoints, the strip one session's closing settlements.

The ≈$15 strip-vs-EIA gap is the market's persistence pricing; bounds are analyst quotes (Goldman severe $115; Fink $40 / $150+). The gray dashed vintages — $58 (Feb) and $96 (Apr, the day of the peak) — are regime-chasing, not conservative.

Methodology & caveats (paths, vintages)

Actuals: FRED daily. EIA path: the latest STEO vintage's quarterly midpoints (currently the Aug 11 edition: $85.26 Q3 / $78.00 Q4 / $74.00 Q1-27 / $65.00 Q4-27 — completed Aug 6, so it embeds July's re-escalation but pre-dates the Aug 7–18 rally). Strip: ICE settlements of the Sep 1 session. Each STEO vintage extrapolated the regime it was published in — too low before the war, too high at the peak, and, if the strip's premium is right, still too low now.

Every price the strait sets, indexed to its own February — ranked by how far it went

So what — crude is not where the shock landed hardest — one axis, one base: each row is that price divided by its own pre-war February level, so a barrel, a tonne of urea and a gallon at an American pump are finally comparable. The report's own subject does not top its own board.

Methodology & caveats (why these are comparable, and where they are not)

Every row is a monthly average indexed to its own February 2026 value. Monthly averages are what make the rows comparable — a weekly peak read against a monthly mean would manufacture divergence — so the pump row is EIA weekly retail averaged to months rather than its $4.50 weekly peak (11 May), and Brent is the daily spot the price chart above draws, averaged to months, which is why Brent peaks at 165 (April 2026) here — a month's mean — and at $138.21, a single close, there. Rows are ranked at the last month every series prints; a row that has printed since ends at that newer reading — drawn hollow while the month is in progress, since it is a mean to date rather than a complete month — outside the ranking. Fuels are IMF Primary Commodity Prices via FRED (Asian LNG and European gas are the free proxies for JKM and TTF, not the assessed markers); fertiliser and the food index are World Bank Pink Sheet and FAO. Where the World Bank Pink Sheet has printed a month the IMF series has not yet reached and the two agree within 3% over the past year — European gas — that month is the Pink Sheet's print, named as such in the hovercard and replaced by the IMF figure when it arrives; coal and Henry Hub, whose Pink Sheet counterparts do not agree that closely, wait for the IMF print; Brent needs no lead, since its own daily series runs ahead of both. Each row's own basis is in its hovercard, and the monthly paths behind these endpoints are on the fuel-divergence chart and the urea arc. This is a ranking of price: it says nothing about how much of each is bought, which is the next panel's question.

Who absorbed it — the same shortage, four lived experiences

So what — the market cleared on poverty, not on price — only 25–30% of global gasoline and gasoil demand sees full market pass-through (IEA). The other two-thirds never got a price signal: it adjusted through subsidy budgets, FX crises, rationing decrees and queues.

Red = rationed by decree or absence, amber = industry shut on margin, purple = the state stopped absorbing it — subsidy or stabilisation buffer cut, capped or lifted, so the price arrived late and all at once, green = merely trimmed by price. Each tile states its own basis and as-of date and links its sources.

Five gates, one globe

So what — every barrel clears at least one of these — the five chokepoints that shape the war's maritime geography, from the strait itself to the farthest relay.

Basemap tiles: OpenStreetMap contributors. Gate readings are the latest 7-day tanker deadweight as a share of each gate's own 2H-2025 baseline, from IMF PortWatch — the same series the line chart below draws. Route line widths are proportional to current flow volume (mb/d): the thickest line is Russia→Asia at ~3.58 mb/d (four-week average to Aug 16, Bloomberg tanker tracking); the thinnest is the Hormuz corridor, still near-shut. Hover any route for its volume and transit time; zoom in for transit-time labels on each leg (zoom 3+) and pipeline/canal detail with flow vs capacity (zoom 4+) — Petroline, ADCOP, SUMED and the Suez Canal with their current throughput and rated capacity. Only the five documented gates are numbered; waypoints are approximate sea lanes, not AIS tracks. One Russia→Asia flow is off this frame entirely: ESPO crude loads at Kozmino and reaches China on a short Pacific hop, touching none of these gates — which is part of why Russia's export line is the tab's most resilient.

The routes, spelled out: five gates, one gradient

So what — a barrel must clear every gate on its route — and AIS comes back on somewhere along the voyage, so the farther a gate sits from the war, the more honest its count.

How to readEvery line is a gate's 7-day tanker deadweight as a share of its own 2H-2025 baseline — the departures floor's arithmetic applied at each gate; 100 = normal. The mb/d view scales each share by that gate's own flow anchor (EIA chokepoint estimates, 1H2025 total petroleum — crude plus products, so Malacca reads high; Hormuz stays on the ~15 exports-basis anchor so its line is exactly the departures floor). A derived floor on the same caveats: dark transits invisible, deadweight counts both directions. Counts and dwt are both shown on the board's gate chips because their disagreement is signal.

Chip colorsEach chip's left edge scores that gate against its own normal — red below 25%, amber 25–89%, green 90% and up.

Reading the board. Each row is a route, its gates in voyage order; a chip is that gate's latest reading against its own normal — deadweight share of the 2H-2025 baseline at sea gates (ship-count share in parentheses), mb/d crude-equivalent against pre-war at origins and arrivals. The relay is the method: a lifting that leaves the Gulf dark re-lights somewhere along the voyage — open-source AIS watchers now log Gulf liftings staying dark until south of India — so Hormuz is the AIS floor's weakest point and Malacca, where everything has re-lit, its strongest. Two count-vs-dwt gaps carry the story: Malacca keeps ~85% of its ships but only ~65% of its deadweight (the missing VLCCs are the missing Gulf flow), and the Cape runs more hulls than baseline at less deadweight (the vessel-class step-down: Suezmax parcels where VLCCs went). Gate series are IMF PortWatch, direction- and cargo-agnostic — shape signals, not flow measurements; the directional read lives in the arrival panels below.

One route, one clock: each corridor's gates aligned to its departure date

So what — the relay, aligned

How to readEvery series is % of its own normal (gates: dwt vs 2H-2025; arrivals/exports: 7-day mean vs pre-war). The route's clock gate is drawn bold and unshifted; each downstream series is shifted back by a pinned voyage lag with ±3–4 days of blur (voyages are distributions, not constants). The r beside each lag is computed on every refresh (Jun 7 – Aug 10 window) — and the fit quality is the route's finding, not a defect: strong on the Gulf→East Asia relay, noise where a clock doesn't feed the arrival line (Suez→Europe, Cape→US), and negative on Gulf→India, whose arrivals are Russian-fed.

The armada, landing: daily tanker arrivals — five demand centers, with the coming week

So what — arrivals audit departures — a departure can go dark; a discharge cannot. Whatever leaves the Gulf must berth somewhere lit within a voyage time.

Departures — the strait itself: AIS-implied flow floor, with every claim at its date (mb/d)

How to readThe flow line is derived — pre-war ~15 mb/d scaled by the seven-day share of baseline tanker deadweight transiting; deadweight counts both directions. The dark-corrected line adds the barrels that left dark and re-lit at Malacca ~13 days downstream (Malacca's dwt-implied flow, minus its ~12.3 mb/d non-Gulf baseline, minus the lit cohort) — an estimate, not a floor, drawn by loading date, which is why it ends ~2 weeks before the lit line: a dark barrel is only visible once it re-lights. The wedge between the lines is never zero, and should not be: Iran's sanctioned exports (~2.2 mb/d pre-crisis) move dark by trade design even in peacetime — which is also why the lit floor read only ~12.7 against the ~15 pre-war anchor. But the wedge is not all covert flow, and its composition flips across the window: during the healthy memorandum weeks Kpler put the truly-dark share of Gulf liftings at just ~5% (the June–July wedge is mostly the armada exit double-counted at both gates, plus non-Gulf barrels crossing Malacca), rising to ~66% by the expiry week — when the wedge turns mostly real. Read it as barrels the strait's lit count misses, of mixed provenance; its memorandum-window average lands within ~1 mb/d of Kpler's print, while its August window has slipped to ~2 mb/d, below the trackers' last public ~4 — either the outflow is fading past their mid-August estimates, or dark barrels now stay dark beyond Malacca. Malacca is the dark trade's re-light gate because its demand side is Asian (the Shandong discharge points) — so the wedge is a floor on dark eastbound flow: dark discharge short of Malacca (India, floating storage) is missed, while dark flow west is structurally near-zero (Atlantic ports cannot take dark cargo — insurance, port-state control, sanctions enforcement). The dashed extension is a model fill, not data: lit is observed through the data edge and only the not-yet-re-lit dark residual is held at its last week's level; the shaded band is the same rule replayed at every wall-clock of the war window and scored against what actually re-lit (p10–p90 of realized errors, widening with horizon). Each weekly print converts a slice of the dashed line into the solid one — the model grades itself in public. The two official claims are graded on this chart, over the 7-day windows they described, against the generous dark-corrected basis: the claimed weeks measured 2.0× and 4.9× below the claims. The third claim is a single day — “more than 17 million barrels” by ship on Monday Aug 31, said on Sep 1–2 — drawn as a red point at its date: it sits above the ~15 pre-war anchor and above the strait's best lit day of the whole war (Jul 6, memorandum open, ~15 mb/d-e), and it is graded when the solid line reaches Aug 31 (~Sep 15) and by the lit day's own print (~Sep 8). The two graded claims missed by 2–5×; applied to 17, that record implies ~5.5 Mb for the day (range ~3.6–8.5), and TankerTrackers' own same-day count (9.1, with a 7-day average of 8.3) sits at the top of that range. One such day adds ~2.4 mb/d to a 7-day mean — inside the band's upper half. The modal outcome, on the Aug 21 precedent, is a convoy-night pulse the week does not keep: the day partly real, the week near the trackers' figure, and the statistic quoted next time different again. The later June rise to 7+ does not vindicate the Jun-12 claim — it began with the Jun-18 memorandum, five days after the claim, and its tallest reach is the armada double-count.

What the gates imply for world supply. On the dark-corrected outflow — … mb/d so far in August against ~14.4 pre-war — the pass-through calibrated on the March–June collapse (a Gulf barrel that fails to sail costs ~one barrel of world supply) puts August world supply near … mb/d (range …; EIA STEO total-liquids basis), the low side of EIA's own … August forecast. September is a strait scenario, not a forecast: flows holding at August's level read …; memorandum-window flows resuming (~… mb/d) read … — EIA's September … sits inside that reopening case, and its path back to ~107 by early 2027 assumes the reopening compounds from October.

Russia — seaborne exports: the other departure side (mb/d crude-equivalent)

All five demand centers, summed — with the world's two sides (mb/d crude-equivalent)

How to readEvery arrivals panel is IMF PortWatch tonnage, all tanker classes at ~7.33 bbl/t — a shape proxy, not a crude level; trailing ~4 days are preliminary. The world pair (teal departures, amber arrivals — real dates, no cohort shift, since they mix every route) shares one denominator and may be compared to each other: the gap between them tracks on-water build and draw, and departures count every loading on a barrel's journey — hub re-exports at Singapore, Fujairah or Rotterdam, coastal legs — not just the first. The blue five-center sum is a ~60% coverage subset: compare it to its own pre-war line only; its level gap to the world lines is coverage, not deficit.

China — every port, summed (mb/d crude-equivalent; all tanker classes)

United States (mb/d crude-equivalent)

India — the fastest referee: 5–7 days from the Gulf (mb/d crude-equivalent)

East Asia — Japan, Korea and Taiwan, summed (mb/d crude-equivalent)

Europe — 10 largest seaborne importers, summed (mb/d crude-equivalent)

The audit. On AIS-derived port arrivals, China lands … mb/d of crude-equivalent (last final week) against a pre-war ~12.4 — …% below in that week, with the third, reroute wave landing; its war-era baseline runs ~40% below pre-war, the same cut its customs tonnage shows. The US Energy Secretary's Aug 11 claim — ~9 mb/d out through Hormuz plus 5–7 via bypasses, ≈15 mb/d total, with trackers' lower counts (~4 mb/d Kpler/Windward; ~7 peak, Commodity Context) dismissed as covert vessels — implies a China arrival level near … mb/d-e. The gap is ≈… mb/d of crude-equivalent that would have to be berthing unseen — about twice any plausible dark-discharge undercount (~1 mb/d of Iranian barrels into Shandong), and narrowing only because the rerouted barrels are landing. Timing matters both ways: today's arrivals reflect late-July departures, when the trackers said ~4 — and they match the trackers. Each claim is drawn on the chart as a red segment in its own arrival-test window (claim date + voyage), graded once the window passes. The June 12 claim has already been graded: its test window (Jul 2–9) implied ~1.35 and measured 1.01 — the war baseline — while the strait itself moved 1, 0, 0, 2, 3, 0 tankers per day between the claim and the memorandum; flows only rose after the Jun 17 memorandum, and the arrival wave came one voyage after that. The Secretary's Aug 4–11 average gets its test in the window Aug 24 – Sep 5: if ~9 mb/d really left dark, China's line must climb past the reroute wave's crest and hold near 12 by then. The newest claim — “more than 17 million barrels” through the strait by ship on Monday Aug 31, said on Sep 1–2 — is a single day rather than an average, the third statistic in three claims, and is drawn on the departures chart at its date. Its arrival footprint is small (India's share ~1.8 Mb over Sep 5–8, about one weekly sigma, drawn as a weak test), so the departures chart's dark-corrected line, reaching Aug 31 around Sep 15, is its referee. And the claims have a track record — the ledger below puts each quantified claim beside the lit strait traffic of its own week. The June pattern is the tell: when flows genuinely recovered under the MoU, AIS saw the climb in real time (2% → 38% of baseline dwt in four weeks), so "trackers miss the covert recovery" is contradicted by the one period when recovery actually happened. Kpler's post-mortem of the sixty-day memorandum window (19 Aug) is the tracker-side bookend: ~374 Mb of crude cleared the Gulf across the window — ~6.1 mb/d, about 40% of the ~15 mb/d Hormuz averaged in 2025 — with more than half of it moving in the first three weeks; by the expiry week 66% of a shrinking total had no confirmed loading port, against ~28% before the memorandum. That last figure is the honest size of the "covert vessels" argument: real and growing — but attached to a total far below the claimed level, not hiding on top of it.

The claims ledger (who said what, and what the lit fleet was doing that week)

Date	Claimant	Claim	Lit strait that week¹	What the data showed	Source

¹ 7-day mean of AIS-visible tanker transits (count and deadweight) through Hormuz vs the 2H-2025 baseline — IMF PortWatch. AIS is a floor; the June recovery shows the floor rises when flows do.

Methodology & caveats (proxy basis, AIS floor, the forecast model)

What the series is: IMF PortWatch's daily port-level import_tanker tonnage summed across every port in the country — ALL tanker classes (crude, products, chemicals, LPG), so a shape proxy rather than a crude level; tons convert at ~7.33 bbl/t where crude-equivalents are quoted. AIS floor: dark discharges are undercounted, which is why the audit carries an explicit ~1 mb/d allowance rather than pretending the floor is a ceiling. The trailing ~4 days are preliminary — measured against archived prints they are noisy but not biased, so the model projections use them while the quoted last-week figures still exclude them. The forecast is scenario modeling, not a promise: voyage arithmetic on cargoes already loaded (strait-direct arrivals ended ~Aug 12, last loadings Jul 18 + ~25 days; what can land next week is short-haul Russian, mid-July Suez/Cape Saudi, and the non-Mideast barrels China ramped in July; the US pulse maps to late-June loadings and looks mostly landed — ~Aug 27 is the outer bound, and the EIA weekly import series is the cross-check). Bands, not points.

The routes west: Cape of Good Hope and Suez, daily tanker deadweight

So what — more hulls, not more tonnage — every Gulf barrel bound for the Atlantic sails one of these two corridors, so what they carry now is what the Atlantic lands 30–45 days later.

Reading the corridors. By deadweight, Cape of Good Hope tanker traffic runs …% … its 2H-2025 baseline (… vs … million dwt a day) and Suez …% … — yet by hull count both run above baseline, the Cape …% (…/day vs …) and Suez …%. The gap is vessel class: the westbound reroute is sailing in Suezmax and Aframax parcels where the strait sent VLCCs — the Suez draft caps the westbound parcel size, and the Sumed shuttle rides the canal as lightened transits — so the deadweight the Gulf used to send west has been re-hulled, not replaced. The Cape's own baseline is already diversion-elevated by the Red Sea years, which makes its shortfall a demanding comparison. Deadweight is the default view because a single day's count can spike when a convoy of small tankers is released together, while the tonnage barely moves. The mb/d view is the same derived floor the departures chart and the gates chart use — each corridor's flow anchor (EIA 1H2025 total petroleum: Cape …, Suez … mb/d, the latter including the Sumed leg the canal count cannot see) scaled by its deadweight share — so it reads against those two, not against the arrivals panels, which convert observed cargo tonnes; on it the Cape's floor is … mb/d. Both series are direction- and cargo-agnostic — all tanker classes, both ways; the public layer carries no east/west split, and the arrival panels above are the directional instrument instead (westbound Cape traffic must surface in the EU/US panels, eastbound in China/India/East Asia, one voyage later) — so they are a shape signal, not a flow measurement: what they lead is Atlantic arrivals ~30–45 days out. The gate model behind the European and US forecast panels reads this deadweight, not the hull count — each corridor's dwt share of baseline times its flow anchor, the Cape 22 days ahead of US arrivals and 14 ahead of Europe, Suez 12 — so the Cape's tonnage shortfall is already in those projections: both lines are drawn flat near their war baselines, not rising into a second, reroute-fed pulse. What would change that is a deadweight recovery on the Cape; a hull rise on its own does not, and a Cape still running smaller ships into October means the reroute west is a parcel-size story, not a volume one.

The attack log: every ship fired on, seized or sunk

So what — the ordnance under every premium above — hull by hull: who fired, where, and what route the ship was on when the war found it.

How to readFour bars per month, one per severity class, worst on the left — hull lost, seized, damaged or disabled, near-miss — each at its own true height, never summed, so a class can be followed across the war: the red bar is the story's floor (never above six), the amber bar is April's seizure campaign, and the pink bar's July tower is mostly faded — the only other visual coding — meaning it rests on a single party's claim or was never confirmed (July's faded pink is the Azov block, where Ukraine's drone force claimed ~35 hulls in 96 hours and only the two lead vessels are corroborated beyond its own videos; unconfirmed claims of a strike sit in damaged-or-disabled, faded, as what they claim to be). The hover does the rest: the red bar names its lost hulls — sunk or written off — and the space between bars gives the month's totals, theater split and claim count. August ends at the 27th.

Is the count real? Counting a near-miss beside a sinking is deliberate — the freight and insurance panels above reprice on the shot, not the hit — but the thin red base of each bar is where to look when the number feels too big.

The full log — … vessels, sortable (date, vessel, route, area, outcome, fleet, confidence)

Date	Theater	Vessel	Type · flag	Fleet	Attacker · method	Area	Route	Cargo	Outcome	Conf.

Methodology & caveats (what counts as an attack; attribution; the route column)

$71 → $138 → $96: Brent & WTI through the war (USD/bbl)

So what — two legs, and no round trip — the first spike broke on the expectation of reopening rather than on barrels arriving; the second began when that expectation died. Brent closed below its pre-war $71.32 on six sessions (Jun 26 – Jul 6, low $68.53 on Jul 2) and has not been below it since.

Table view (monthly averages)

Month	Brent avg	Brent range	WTI avg

FRED daily spot (DCOILBRENTEU / DCOILWTICO), refreshed 2026-09-04; the series lags a few business days (last print Sep 1, $96.02). The second leg peaked at $105.32 on Jul 23 before giving back $9.30. Shaded bands = war phases; numbered markers are keyed in the event table below.

Benchmarks inverted: differentials to Brent, monthly ($/bbl)

So what — quality broke first — the barrels were still being produced; what failed was which grade could reach which refinery, and both benchmarks priced that before any volume was lost.

How to readEach line is that benchmark minus Brent — above the zero line the grade trades over Brent; below it, at a discount.

Table view (monthly differentials)

Month	Dubai − Brent	WTI − Brent

Both inversions in one picture — quality and geography broke before quantity did. Dubai's March +$23.6 came with the benchmark itself breaking: Upper Zakum, Al Shaheen and Fateh dropped out as deliverable grades and price discovery migrated to Murban futures. WTI–Brent hit a daily extreme of −$25.94 on Apr 8 as Atlantic light sweet was pulled East. Both are FRED monthly series. Urals is not drawn — no continuous public series exists, only sparse third-party assessments (the last is Reuters, 9 June) — but its story is told where it is sourced: the April–May premium over Brent that overran the EU's $44.10 cap, and the grade-mix discussion.

Market-implied probability the disruption persists

Scenario modeling: two-state risk-neutral mixture, p = (futures − EIA base) / (severe − EIA base). Assumes exactly two outcomes, risk-neutral pricing, and no variance premium — a calibration device, not a forecast. Sensitivity across severe anchors $105 (Goldman-adjacent) to $150 (Fink bound) shown as columns. Anchors: EIA STEO Aug 11 base ($78.00 Q4); futures are the Sep 1 ICE settlements. The strip settles daily while FRED's spot series lags a few business days (last print Sep 1, $96.02), so the two sides of this comparison can be days apart. Against that curve, spot peaked at $105.32 (Jul 23), $10 above the front contract, and last closed at $88.24, $7 below it.

The implied price/supply curve: what each lost mb/d cost — 2026 vs history’s supply shocks

So what — the price of a barrel not there — and the curve is steeper going up than coming down: the same shortfall costs more while it is worsening than while it is healing.

What it actually cost at the pump ($/gallon, US retail gasoline & diesel)

So what — crude stopped explaining it — the pump does not trail crude: week to week the two move together, and they troughed in the same week. What they did instead is come apart. About half of a crude move reaches the pump inside the week, and since late April the pump has held a widening margin above anything crude alone accounts for.

Table view (weekly pump price, the crude-implied line, and the gap)

Week	Gasoline ($/gal)	Diesel ($/gal)	Implied by crude ($/gal)	Gap (¢)	Brent ($/bbl)

Methodology & caveats (pass-through, and what the wedge is not)

The pump does not lag crude; it diverged from it. The two series troughed in the same week, and the cross-correlation of their weekly changes is largest at lag zero (r = 0.74), collapsing by one week (r = 0.39) — the peaks land five weeks apart not because one trails the other but because the two paths diverged from late April.

Pass-through is the slope of weekly pump changes on contemporaneous weekly Brent changes: 1.32¢/gal per $1/bbl (95% CI 0.93 to 1.71, r² = 0.55, n = 38) against the 2.38¢ of the 42-gallon arithmetic. That is a short-run figure — roughly half a crude move lands inside the week; cumulative pass-through over months can be larger and is not estimated here.

What the wedge is not. This panel cannot separate refining margin from tax, retail margin or grade mix, so the 69¢ is an observation and not an attribution. It is consistent with products running tighter than crude because refining was hit on both fronts (the refining ratchet), but it does not establish it. GASREGW is a national average across grades and formulations; state and grade spreads are far wider than the moves drawn here. Brent is averaged to the same Monday weeks GASREGW is surveyed on — that alignment is what makes “same week” meaningful, and it makes this crude line smoother than the daily spot elsewhere on this tab.

What refiners were paid: product crack spreads over Brent ($/bbl, weekly)

So what — the fuel, not the barrel, carried the premium — a crack is the gap between a fuel's spot price and the crude it is made from. When crude is short, the crack narrows; when refining is short, it widens. Which one moved says where the shock landed.

How to readProduct spot × 42 gallons − Brent spot, averaged by week. It is a margin proxy on public markers (Atlantic-basin products against a European crude), not any refiner's realised margin: no freight, operating cost or crude slate. Heating oil stands in for diesel and sits a few cents below ULSD, so the distillate crack shown is a floor.

Table view (weekly crack spreads and Brent, $/bbl)

Week of	Gasoline	Distillate	Jet	3-2-1	Brent

Methodology & caveats (which markers, why heating oil, what a crack is not)

Series. FRED daily spot: DGASNYH (conventional gasoline, New York Harbor), DHOILNYH (No. 2 heating oil, New York Harbor), DJFUELUSGULF (kerosene-type jet, US Gulf Coast), DCOILBRENTEU (Brent). Each product is converted at 42 gal/bbl and Brent subtracted on the same trading day; weeks are Monday-labelled means of the trading days available. The 3-2-1 composite is (2 × gasoline + 1 × distillate) ÷ 3, the textbook proxy for a simple refinery's margin. What it is not. A crack against Brent for US products carries the Brent–WTI spread inside it, so its level is not a US refiner's margin; the changes are what this card reads. Heating oil is used because FRED carries no free ULSD series; the two track closely and ULSD trades a few cents above. Singapore and European cracks, where the Gulf shortfall landed first, are not free daily series and are quoted on the Supply tab from press reports, dated.

The same barrel in the buyer's currency: Brent indexed to February, in dollars and six importer currencies

So what — the exchange rate is part of the bill — a currency that weakened against the dollar during the war made the barrel dearer than the Brent chart shows; one that strengthened made it cheaper. The gap between each local line and the dollar line is that effect alone.

How to read100 = each series' own February 2026 average. Above the dollar line, the currency weakened and the barrel cost more in local terms than the dollar move alone; below it, the currency strengthened and absorbed part of the shock. Rates are Fed H.10 noon buying rates; the real is in the table but not drawn.

Table view (rate change since February, and where each local Brent index stands)

Currency	Feb 2026 rate (per $)	Latest rate	Change vs $	Brent index, $	Brent index, local	FX wedge (pts)	Local peak

Methodology & caveats (what the index isolates, and what it leaves out)

Construction. Brent spot (FRED DCOILBRENTEU) × the local units per dollar on the same day (Fed H.10 via FRED: DEXINUS, DEXJPUS, DEXKOUS, DEXCHUS, DEXBZUS; DEXUSEU inverted for the euro), each divided by its own February 2026 average. Days missing either series are dropped before Monday-labelled weekly means are taken, so the wedge is never a timing artefact. Leaves out. Taxes, subsidies, term-contract pricing and the Dubai basis Asian buyers actually pay — the Overview's import-bill panel handles volumes and Dubai; this card isolates one thing, the currency.

What the SPR did to pump prices — and what refilling it will do (¢/gallon, US retail gasoline)

So what — asymmetric by design — a release is fast and lands on a high price; a refill is slow and lands on a low one, so putting the barrels back moves the pump far less than taking them out did. That asymmetry is the case for holding a reserve — and the reason the bill for using one is deferred rather than avoided.

How to readBars are model ranges, not data — each spans the de-escalation and escalation fits, dot at the central fit; the Treasury row is an outside benchmark.

Table view (rates, price bases, and effects)

Scenario	Rate (mb/d)	Δ Brent, mid ($/bbl)	Pump effect (¢/gal, lo–hi)	Price basis

Who actually gives up a lost barrel — every elasticity from the war months, market vs household

So what — the pump balances almost nothing — the price system cleared the shortfall, but almost none of the clearing happened at the pump. The gap between the market's response and the household's is inventories, reserves and rerouting doing the absorbing.

How to readEach dot or bar is an elasticity — % of quantity given up per 1% of price, magnitudes on one axis; rows come from different fits at different lags and never sum.

Table view (each link of the chain: estimate, lag, fit)

Link	Elasticity	Lag (wks)	Fit	Method

The demand loop: pump price vs gasoline demand — and the six-week re-pairing that closes it

So what — the loop is the lag — paired week-for-week the path orbits the line, because households answer a price about six weeks after it is set; hand each month's driving the price from six weeks earlier and the orbit collapses onto the elasticity the ladder above publishes.

How to readDots are months joined in date order, both axes % deviations from the 2021–25 same-week norm; the dashed line is the fitted elasticity, identical in both views — only the price pairing changes.

Table view (monthly deviations, both pairings)

Month (2026)	Gasoline demand (% dev)	Pump price, same week (% dev)	Pump price, 6 wks earlier (% dev)	Weekly prints

World total liquids supply: 2026 by month, and scenario paths into 2027 (mb/d)

So what — the rebound was permission, not repair 106.9 → 94.5 → 101.5 mb/d: the largest drop ever recorded, and the rebound rode the reopening

Table view (observed months + scenario paths)

Month (2026)	World supply (mb/d)	m/m	Status
Jan	106.5	−0.4	historical
Feb	106.9	+0.4	historical
Mar	97.0	−9.9 (largest ever)	estimate
Apr	95.1	−1.9	estimate
May	94.5	−0.6 (trough)	estimate
Jun	98.8	+4.3	preliminary
Jul	101.5	+2.7	preliminary (OMR 12 Aug)

Scenario path (author, from Jul 101.5)	Aug 2026	Dec 2026	Jun 2027	Anchor

IEA total liquids, monthly; scenario paths (author scenario model — toggles above) are author scenarios branching from the last published month, colored by the scenario ladder. Price consequences of each branch are quantified on the Scenarios tab.

Methodology & caveats (baselines, path construction)

IEA total liquids = crude + condensate + NGLs + processing gains + biofuels. m/m deltas computed from the level column; IEA's own quoted deltas (−10.1 Mar, +4.1 Jun) come from different vintages. Press "9.4 mb/d below pre-war" for June uses a different baseline — vs Feb 106.9 the gap is 8.1. Aug-OMR forecast: 2026 average 102 mb/d (−4.3 y/y, cut a further 0.6 on 12 Aug, with 3Q26 alone cut 1.7 vs the July report); 2027 rebound to 110.3 contingent on de-escalation. EIA's quarterly equivalents (Aug STEO): Q1 104.02 → Q2 96.14 → Q3f 99.67 → Q4f 103.66. Paths branch from the observed July print of 101.5 (IEA OMR, 12 Aug). July looks high because production is not exports, and July's monthly average is mostly pre-closure days — PortWatch has Hormuz tanker capacity at 37% of the 2H25 baseline in the week of Jul 5 and ~1% only from the week of Jul 19, while Bloomberg's 4 Aug survey already had OPEC crude +1.16 m/m. The re-escalation hit therefore lands in August (~97.3, giving back ~60% of the 7.0 mb/d recovered since the May trough), on round one's demonstrated lag: the 28 Feb closure took the full quarter to convert into shut-ins (Mar 97.0 → Apr 95.1 → May 94.5). The two war-expands paths take their levels straight from the closure-scenario supply panel: Feb baseline minus the headline shut-in (A: 106.9 − 9.5 = 97.4; B: 106.9 − 15.5 = 91.4, below the May trough), using the round-1 calibration that shut-in maps ~1:1 to measured world shortfall (June: 8.3 shut in, 8.1 measured). The recovery path climbs inside the demonstrated June ramp (+3.5 m/m against +4.3 observed) but stays ~1 mb/d under EIA's Q4 base until war damage repairs; frozen conflict is May-style escorted transit. Scenario A undershoots its own 97.4 level in August because the bypasses take a month to ramp, as they did in round one.

Where the 12.4 mb/d went — and what came back, by producer (mb/d)

So what — the offset was a seventh Five Gulf producers lost >1 mb/d each; record light-sweet output offset barely a seventh of it

…and who stopped receiving them: February → April, by buyer (mb/d)

The same missing barrels as the bars above, counted on the buyer's side — never add the two charts. April is the only month all four buyers are measured, so it is the only single-month decomposition that sums; China's own trough came later and deeper (June; its first month back up only arrived with the July customs print). Four buyers only: Europe took ~4% of pre-war Hormuz crude exports but no citable monthly volume exists, so this understates the demand-side cut rather than measures all of it. Severity as a share of each buyer's own inflow — where Japan, not China, leads — is the dumbbell below; how China bridged the gap is the China-buffer panel.

…and how much of it buyers have gotten back: April → each buyer's latest print (mb/d)

Read the composition, not just the sum. Japan, Korea and India together clawed back ≈2.4 mb/d of their April cuts, and China — which slid another 2.3 to its own June trough after April — finally turned in July: the customs print rebounded +1.3 m/m to 8.45 (still −24% y/y, and still ≈1.0 below its April level). Net across the four buyers: ≈+1.4, with the laggard now moving. Each bar ends at that buyer's latest print, so the months differ by row (hover for each basis) — this chart deliberately does not sum to a single-month total the way the April one above does.

What was hit, how big, and for how long — capacity offline, a waterfall on the calendar

So what — nothing has net healed Nothing has net healed: 8.3 mb/d is still offline at the era's end, within one healed pipeline spike of the 9.3 peak — every export terminal that was hit came back, not one Russian refinery did

Every outage window is two steps: up when the capacity goes off, down (hollow) when it comes back, so the track is mb/d affected and still offline at that instant. Solid = physical damage; hatched = undamaged but blocked or precautionarily shut; lighter = partial, and each partial's figure declares in its hovercard whether it is sourced or an author estimate. Read the two panels against each other: round one is spikes that retrace and ends at less than half its peak; the war panel ends within one healed port spike of its peak. That is the repair clock — every Russian export terminal struck since February is back, and not one Russian refinery is. The eras are toggled separately because their scales differ — round one never exceeds 1.9 — so the war panel carries round one's peak as a dashed line for comparison. Hormuz is excluded from the total and rides its own strip — at ~11 mb/d it is 6× the largest damaged asset and would flatten every step below it.

Methodology & caveats (units, partials, omissions)

Gas converted to ~mboe/d at 5.8 mcf/boe so the steps add. Partial windows carry an explicit per-window offline figure rather than a blanket multiplier, because "partial" means different things across the table: for Mina al-Ahmadi and Kstovo the capacity quoted is already the damaged units, while for Ryazan (Aug-2025) and CPC it is the whole asset. Where a source quantifies the partial it is used (CPC "−30–40%" → 0.53; Ryazan "~half of capacity" → 0.17); where none does, the convention is half the asset — the same author estimate the ratchet note applies to Volgograd. Every figure and its basis is in the hovercard. Windows dated "approx." are dated from qualitative reporting ("weeks", "days"). Single-unit, weeks-scale Gulf hits (SAMREF, Riyadh, Ju'aymah, Manifa/Khurais, UAE/Oman/Iraq) and diffusely-dated Russian units (Perm, Novokuibyshevsk, Syzran, Yaroslavl) are omitted, so the track is a floor — see the master table below (each highlighted asset links to its primary source). Per-asset durations are in that table and in each step's hovercard.

Refining offline worldwide: two wars, two signatures (mb/d of capacity)

So what — the bigger theater is the Gulf 6.5 mb/d of refining capacity is off across the two wars, and 3.5 of it is in the Gulf against 3.0 in Russia — the larger theater is the one the ratchet below does not cover, and the two have opposite shapes

How to readHeight is nameplate capacity offline, not lost runs — a dark refinery subtracts throughput at roughly its utilisation (world runs, a different basis, bottomed near 77 mb/d in April).

The two shapes are the finding. Russia never recovers — every step is still down, because Ukraine re-strikes faster than repairs land, which is what the ratchet below decomposes. The Gulf does: Ras Tanura came back in 24 days and Bazan Haifa in two weeks, so its line falls twice. Jazan (Jul 25) is expected back ~Aug 30 on a consultancy estimate Aramco has never confirmed — IIR pushed it back from ~Aug 15 after the 9 Aug re-strike, and no restart had been reported by Aug 18. Russia's line sits above the ratchet's ~2.6 because it also carries Ukraine's Kremenchuk, off since 2022.

By country (capacity offline at the latest window, mb/d)

Country	mb/d offline	Theater

Iran's own 1.8 mb/d system is the single largest entry and has been down since 28 February — larger than any national total on the Russian side. Windows and per-asset sources are in the inventory below.

The Russian refining ratchet — a waterfall on the calendar (named sustained outages, mb/d)

So what — nine of the ten fell since May 5 Ten named refineries, ~2.6 mb/d, still down — nine of them knocked out in the fourteen weeks since May 5

How to readEach bar is one refinery at its strike date, height = capacity, stacking cumulatively; the dashed lines are three different estimator classes, not a range.

Only the named, still-ongoing outages (re-verified 20 Aug 2026) — a floor, not the tracker total. The steps accelerate visibly: one outage added Nov 2025–Apr 2026, then nine in the fourteen weeks since May 5 — the re-hit-during-repairs cycle described below. The August additions differ in kind: Orsk's governor puts repairs at up to six months because the damaged units need imported parts sanctions block, and Volgograd arrived here by exhaustion — a year of partial halts ended in a full stop after the Jul 31 strike.

Methodology & caveats (exclusions, reference lines)

Single-unit damage at Perm, Novokuibyshevsk, Syzran, Yaroslavl and others isn't cleanly dateable and is excluded, as is TANECO Nizhnekamsk — struck Aug 10 and again Aug 19, the war's deepest strikes and its deadliest (13 killed), but with no confirmed damage to processing units. Volgograd's 2025–26 recurrent partials were excluded on the same grounds; its full stop after the Jul 31 strike is dateable and joins the waterfall at 0.30. The named floor (~2.6) now exceeds the ~2.14 mb/d tracker figure, which predates the August wave — the Kyiv General-Staff claim (~2.7) is the only reference the floor has not overtaken. Reference lines convert the three estimator classes onto the same ~6.4 mb/d design base implied by the tracker figure itself; they are different estimators, not bounds of one range. Each bar's hovercard names its source; asset links are in the master table above.

Russia's oil exports: tonnes shipped against dollars earned (CREA estimate, 7-day means)

So what — revenue moved with the price, not the volume — the sanctions question is whether Russia's barrels keep flowing while the money they earn falls. Putting the two on one chart is the only way to answer it.

How to readBoth lines are CREA's Russia Fossil Tracker for the oil group (crude and products, seaborne and pipeline), on CREA's "actual price" scenario — a third-party estimate, read for shape. Tonnes convert at 7.3 bbl/t, so the volume line is crude-equivalent and the implied $/bbl in the table is a group average.

Table view (monthly: revenue, volume, implied price, discount to Brent)

Month	Revenue ($ bn)	Volume (mb/d, crude-eq.)	Implied $/bbl	Brent avg ($/bbl)	Discount to Brent

Methodology & caveats (what CREA counts, and the 7.3 bbl/t proxy)

Source. CREA's counter API, daily by commodity group. CREA tracks vessels and pipelines for tonnage and applies its own price estimates by grade and destination; the "actual price" scenario is their central case. Proxy. The oil group mixes crude and products, which carry different barrels per tonne; 7.3 is a crude-equivalent convention, so the implied realised price is a group average and its discount to Brent is indicative, not the Urals discount (which the Supply tab quotes from press, dated). Why it is here. The Overview's Russia figure is tanker loadings alone; this card is the revenue side of the same flow, and the pairing is what the price cap is meant to act on.

Biggest movers: February → trough → latest, by producer (mb/d)

So what — depth is not the same as exposure — Saudi Arabia lost the most barrels, but Kuwait lost the largest share of itself and has neither a pipeline nor a second coast to lose it through. The two rows that go the other way, the United States and a deliberately flat China, are the scale the collapses should be read against.

○ Feb (pre-war) · ● trough · ● latest where recovery is visible. Bases are mixed by necessity (monthly crude surveys where they exist, EIA quarterly averages or liquids where not — hover a row); Russia is excluded here — its drone-strike decline rides in the waterfall's residual bar above, and its story is rerouting, not Hormuz. China is included precisely because it does not move: a flat row at ~5.5 mb/d is the control case, and the only large producer that had one.

Table view (monthly detail + notes)

Producer	Jan	Feb	Mar	Q2 avg / latest	Note
Saudi Arabia	10.30	10.40	7.25	6.57 May → 7.12 Jun → 7.40 Jul	May lowest since 1990; heavier offshore grades shut; Petroline bypass carries light grades only (Jul: Bloomberg Aug 4 survey)
Iraq	4.34	4.57	1.57	1.46 (Q2) → 2.30 Jul (survey)	>90% of exports via Hormuz; no bypass; most exposed producer — and oil is 88% of government revenue
Kuwait	2.60	2.54	1.19	0.66 (Q2) → 1.57 Jul (survey)	Exports literally zero in March; deepest relative cut of any producer — and zero bypass of any kind, no pipeline and no second coast
UAE	3.61	3.64	2.37	3.08 liquids (Q2)	Murban kept flowing via ADCOP to Fujairah; offshore shut. Quit OPEC effective May 1
Iran	3.45	3.69	3.63	2.33 May; 2.85 Q2 avg (EIA)	Kept exporting through "its own" closure until the US blockade (Apr 13–May 29) cut exports to ~65 kb/d; June exports rebounded to ~1.76 mb/d (UANI tanker tracking) as the blockade lifted
Qatar (liquids)	~1.9	—	shut-ins	0.35 (Q2)	Ras Laffan strikes; condensate/NGL loss ~1.1 mb/d
Russia	9.26	8.67	8.96	8.94 (Q2, EIA)	Output below its start-of-year level and falling again into June ("sixth straight month" of decline — Moscow Times, 11 Jun, a different measure/vintage than this row); ~690 kb/d under quota; drones force crude out as refining dies
United States (crude)	13.57 (Q1)		13.88 (Q2, record)		Record quarter; liquids 24.31 mb/d
Brazil / Guyana / Kazakhstan	records: Brazil 5.32, Guyana 0.91 (Q2 liquids); Kazakhstan 1.88 (Mar crude)				The light-sweet offset, pulled East (+3.5 mb/d Atlantic-to-Asia)

OPEC+ crude collapsed 43.4 (Feb) → 35.2 (Mar) mb/d. Quota hikes continued as "paper formality" (+206/+188 kb/d tranches; the Sep tranche approved Aug 2 completed the 1.65 mb/d rollback). EIA/OPEC series break at May 2026 when the UAE (~12% of OPEC output) exited.

Who stopped buying: February → trough → latest, as a share of each buyer's own pre-war inflow

So what — severity is not size Japan lost 70% of its crude inflow, the deepest cut of the four and its lowest level since the series began in 1979; China's larger 4.89 mb/d cut was 41% — and is the only one with nothing back — together the four cut 6.34 mb/d in April alone, the demand side of the same shock the producer bars above measure on the supply side.

○ pre-war — 100% for every row, so the red segment's length is the depth of the cut · ● trough · ● latest. Right-hand figures give the cut and how much of it has come back; volumes in mb/d are one hover away and in the table below. Scaled this way because size is not severity: in mb/d China owns the longest bar and the second-shallowest cut, which ranks the four by how big a buyer each is rather than by how hard it was hit. The sub-labels give the reason: Hormuz exposure orders the depth of the cut for three of the four rows — Korea is the exception, 70% exposed and only 35% cut, with no sourced explanation on record for why it outperformed its exposure. Read it against the producer chart above: those bars are barrels that stopped being pumped, these are barrels that stopped being bought. Every dot carries its own basis — customs tonnage for China, ministry kilolitres for Japan, a secured-cargo tally for Korea, a press mb/d print for India — and Hover a row for each dot's basis. Europe is absent, not zero: only ~10% of its seaborne crude came from the Middle East Gulf and it took ~4% of pre-war Hormuz crude exports, but no citable monthly European import volume exists.

Table view (per-dot basis, April-aligned total, why the troughs differ)

Importer	Pre-war	Trough	Latest	Cut at trough	% lost	% of cut back	Cut in April	via Hormuz	Basis: pre-war / trough / latest

Why there are two totals. The troughs do not share a month: Japan, Korea and India bottom in April at the blockade's peak, but China bottoms in June — its January–February stockpiling, then its own run cuts, pushed the low later. Summing four different worst months gives 8.59 mb/d, which was never true of any single month. April is the one month all four are measured in, and the April-aligned cut is 6.34 mb/d — that is the quotable figure.

China's cut is over half the total and is the one with somewhere to go: its inflow loss was absorbed by stopped stockpile builds, a product-export halt, run cuts and a genuine stock draw — decomposed in the China-buffer panel. India's line is the counter-example: it ends the period roughly where it began, having replaced Gulf barrels with a record ~2.6–2.7 mb/d from Russia.

Asia's four big buyers: fewer barrels, bigger bill (April 2026)

So what — the shortage was not a discount — the crude that did arrive cost so much more that landing a quarter fewer barrels still cost more money. This is the strait converted into cash: observed volumes on one side, the observed Dubai price on the other.

Table view (volumes, prices and the daily bill, per buyer)

Buyer	Pre-war mb/d	April mb/d	Barrels	Pre-war $mn/day	April $mn/day	Bill

Methodology & caveats (why Dubai, why April, and what the baseline assumes)

Priced on Dubai, not Brent. These are Asian sour barrels; they price off Dubai and the Gulf official selling prices set against it. The two benchmarks came apart in exactly this window — Dubai ran $23.60 above Brent in March and $12.00 below it in April (the differentials), so pricing April on Brent would overstate the bill by more than a tenth. The series is FRED POILDUBUSDM, the monthly average behind that same spread. April, because it is the only month all four buyers are measured in — the same volumes the inflow panel draws. The one counterfactual is the pre-war leg: each buyer's own pre-war volume valued at the February Dubai average — the same barrels at the price before the war. The April leg is observed on both sides. Volume bases differ by buyer (China customs tonnage, Japan a ministry return, Korea a 2025 average, India a press print) and each row's hovercard names its own. This is crude only — it excludes the LNG, LPG and products these same buyers import through the same strait, so it is a floor on what the strait cost them.

The wrong barrels came back: heavy crude out, light crude in (mb/d, Feb → May)

So what — quantity was replaced, quality was not ~12 mb/d of sour and condensate was locked in; ~3 mb/d of mostly light-sweet came back — the mismatch broke the benchmarks

Barrel counts regrouped from the producer waterfall above (same data, bucketed by grade class; the bypass pipelines carry only lighter grades, so what stayed locked in was the sour end). The largely-sour SPR barrels are one reason the coordinated release punched above its weight for refiners. Russia's medium-sour decline (waterfall residual) is excluded — those barrels rerouted, they weren't locked in. Consequences: Dubai benchmark "effectively broken" (basket cut to Murban+Oman); medium sour at times traded above light sweet; Urals discount $12.6 → $2–3 (WCS, by contrast, was assessed ~$14.40 under WTI for August barrels as of Jul 8); Venezuela's heavy sour got a policy boost (OFAC licenses, ~1.0–1.1 mb/d). Products worse than crude: Gulf product exports (3.3 mb/d) nearly all halted; >4 mb/d of Mideast refining shut; Singapore distillates hit $290/bbl; global runs bottomed ~77 mb/d in April. And products have a different bypass from crude: Saudi Arabia loads refined product at Jazan, Yasref and Rabigh on the Red Sea — though Jazan (400 kb/d) was struck on 24–25 July and shut, restart estimated ~15 August (IIR), and YASREF's and Rabigh's post-embargo status is unpublished — but Ruwais (837 kb/d), Al-Zour (615 kb/d), SATORP, Ras Tanura, Bandar Abbas and Abadan are all inside the strait — the product bypass belongs to one country and reroutes no Gulf-loaded barrel (the ledger).

Displacement & demand: oil avoided vs added back, 2022–2030 (mb/d) — the wars barely moved it, and 2026 runs backwards

So what — the transition is not a shock absorber — EVs and LNG substitution grind at a pace set by fleet turnover, which no price spike accelerates within a year. In the one year the market needed displacement most, it went the other way: war-priced gas pushed users toward oil.

How to readBars above zero are oil demand avoided, below zero demand added back; the diamond is each year's net, and faded years are author scenario.

Durable displacement is overwhelmingly structural (EVs above all; the wars' attributable share is a few hundred kb/d) and grinds upward regardless — but 2026 is the reversal year: war-priced gas is adding oil demand back via gas-to-oil switching and the un-done Gulf oil-burn program, knocking net displacement from ~2.4 to ~2.05 mb/d — the first year-on-year decline in the series.

Methodology & caveats (anchors, sources, the IEA–OPEC dispute)

Gas-to-oil switching adds back a few hundred kb/d (author estimate, keyed −0.2; the oft-quoted "up to +1 mb/d" is Energy Intelligence's September 2022 framework ceiling, not a 2026 observation) after the Hormuz closure knocked out ~20% of global LNG supply (TTF +32%, JKM +45% y/y); Gulf oil-burn reversal +0.3–0.5 mb/d (Rystad). The cheap-LNG-displaces-oil thesis is deferred to 2027–28 (Ras Laffan out 3–5 years; IEA: 140 bcm of 2026–30 LNG supply lost) and assumes the strait reliably reopens — faded bars are author scenario anchored to published endpoints. Related dispute: IEA vs OPEC disagree by a record 2.44 mb/d on the 2026 demand level (103.5 vs 105.94 mb/d; growth −1.0 vs +0.8, a ~1.8 mb/d growth-basis gap) — structural-vs-cyclical readings of the same weakness. Hover any bar for its status and source.

China's buffer: the burn that never happened vs the draw trackers saw (cumulative Mb since Mar 1)

So what — the reserve was never opened — the counterfactual line is what China would have had to burn to keep consuming at pre-war rates. It didn't. The reserve is still there, which is why China's buffer is a live variable in every scenario ahead rather than a spent one.

How to readBelow zero means stocks still building, above zero net drawing — the sign convention both cumulative lines share.

Table view (customs imports, shortfalls, and reserve-scale estimates)

Month (2026)	Customs imports (Mt)	mb/d	Shortfall vs Jan–Feb baseline	Basis

Reserve scale — third-party estimates (unpublished by China; each row its own estimator and basis, never blend)	Mb	Cover at observed ~1.4 mb/d war draw

Month	Absorption channel	mb/d	Basis

How the import shortfall is absorbed, channel by channel — reserves were the last resort

So what — rationing, not reserves — the two channels that cost nothing politically came first; the government reserve was never touched. Every barrel absorbed this way is one China does not have to buy back, and one the market does not get handed back when the war ends.

How to readA waterfall in mb/d: each channel steps the month's shortfall down until it reaches zero — the rows sum by construction. In July the storage bar points up: tanks were refilling, so rationing had to absorb more than the shortfall alone.

Inside the rationing bar: consumption readings vs the residual they must explain (mb/d)

So what — the grid cannot help, and the readings don't reach — the rest of the residual is petrochemical-feedstock runs idled at the teapots plus whatever product-stock draw is quietly bridging consumption, which no satellite can verify: product tanks have no floating roofs.

How to readEach bar is one estimator's reading of the demand-side squeeze, converted at a keyed EIA/IEA level — different periods and bases, so rows bound the residual and never sum to it. The dashed rules are the supply-side residuals the readings must explain.

Table view (each reading, its estimator and level basis)

Reading	mb/d	Basis

The tanks themselves: built cheap, flipped in May, drawn through the war

So what — the buffer worked exactly as designed, in public — crude stored at 2025's $60–75 was spent at war prices, not from government caverns but from the commercial and refinery tanks it was put in, every step visible to the satellites that track floating-roof lids.

The level: satellite-anchored onshore crude (Mb)

How to readThe line is the cumulative-change splice re-anchored to Vortexa's observed May 5 record; the May 25 print is reconstructed, not keyed, and lands exactly — the built-in consistency check. Axis cropped (1,090–1,290 Mb); the two dots are the only observed levels, everything else is derived.

The flow: monthly net change (mb/d)

How to readAbove zero = drawing stocks down, below zero = building — the same sign convention as the cumulative chart above. Bars = satellite splice (bases flagged per bar); ◆ diamonds = the implied balance from official flows, an independent cross-check. July's bar carries the implied figure — the satellite print isn't out yet.

Table view (levels, monthly net change, bases)

Date (2026)	Onshore level (Mb)	Basis

Month (2026)	Net change (mb/d, draw-positive)	Basis

What was never coming back: China's pre-war 2026 demand outlook, decomposed

So what — the rationing landed on fuels already in structural decline — none of that is war. EVs, LNG trucks, and the construction slide were shrinking China's fuel demand before the first strike, which is why June's rationing was cheap to bear — and it lowers, each year, the level a peace would hand back.

How to readA waterfall in kb/d of year-on-year change, on the pre-war outlook basis (Kpler, Jan 2026) — deliberately separate from the June shock ledger above, which is wartime arithmetic on a February baseline.

The grid fought this war without oil: July generation by source (% y/y)

So what — the war's only grid casualty was gas, and coal still fell — China generates essentially no electricity from oil, so the strait could not reach the grid. With solar, hydro and nuclear growing 7–10% into flat demand, even coal burn fell; the lost LNG was absorbed without a single oil-fired megawatt. Electrification reaches oil through vehicles, not power stations — that chart is above.

How to readBars are July 2026 vs July 2025 generation, CREA prints of official data; the dashed rule is total generation (+0.5%) — sources growing faster than it are gaining share.

US SPR: 2026 drawdown in detail, and the refill paths into 2027 (million barrels)

So what — the trough is fixed, the pace is a price call All three paths share the ~243 Mb trough and differ only in pace — which leg plays out is effectively a price call

How to readWeekly EIA actuals in Mb; everything right of the latest print is a path — a DOE claim, an analyst view, an author scenario — not data.

Table view (2026 actuals + scenarios; full 2022–26 history below)

Date	Actual (Mb)	DOE claim (program pace)	Analyst (announced schedule)	Extended-release (author)
2026-02-27 (pre-war peak)	415.4	—	—	—
2026-05-01	392.7	—	—	—
2026-08-28 (latest)	286.6	286.6	286.6	286.6
2026-10-02 (trough at program pace, 1.26 mb/d)	—	~243	—	—
2026-09-25 (trough on announced schedule, 1.54 mb/d)	—	refilling	~243	~243 (draws continue)
2026-12-31	—	~305	~243	~194
2027-02-26	—	—	—	~177 trough
2027-06-30	—	~405	~286	~177
2027-12-31	—	~443	~329	~177

Week ending	SPR (Mb)	Context
2022-01-07	593.4	Pre-Ukraine-war baseline
2022-04-01	564.6	180 Mb emergency sale announced Mar 31 (sold ~$95/bbl)
2022-12-30	372.4	~221 Mb drawn in 2022
2023-07-07	346.8	Post-Ukraine low (flat bottom Jul 7–28)
2024-12-27	393.6	Refill: 59 Mb bought <$76/bbl + 140 Mb mandated sales canceled
2025-06-13	402.3	12-day Israel–Iran war: no release (Hormuz never closed)
2026-02-27	415.4	Pre-war peak; 2026 war begins Feb 28
2026-05-01	392.7	172 Mb release (US share of IEA 400 Mb) delivering
2026-06-05	349.2
2026-07-10	316.5	−98.9 Mb from pre-war peak (holiday-dip week)
2026-07-17	311.4	−104.0 Mb from pre-war peak
2026-08-28	286.6	Lowest since November 1982; −128.8 Mb from pre-war peak

All three toggleable paths share the committed-release trough (~243 Mb = 286.6 − 43.2) and differ only in pace and whether the war ends.

Methodology & caveats (path construction, DOE-claim reconciliation)

Every pace shown is one DOE has actually run. When the committed leg completes is a pace call, with three answers: the remaining 43.2 Mb runs dry on September 25 only at the announced-schedule 1.54 mb/d — a pace now above even the observed program plateau of 1.26 mb/d (the May 15–Jun 19 average, used by the DOE-claim path), which lands October 2 — and as late as Dec 3 if the latest weekly pace of 0.45 mb/d holds. The DOE claim path takes DOE's "~200 Mb within the next year" of war end at face value — the volume is reachable only if the programme fully subscribes: the 133 Mb awarded so far returns ≈170 Mb with its premium, the full 172 Mb would return ≈219 Mb, and DOE's 200 sits between them (see the release chart above); the analyst path (Sparta/OilX) starts returns in 2027 and spreads the awarded book's ~170 Mb over two years — pre-war levels only by 2028. Both require the war to end and companies to have crude to return. The extended-release path is an author scenario for the July 8 re-escalation persisting: releases don't stop when the committed 172 Mb completes — halting the draw before the November 2026 midterms is assumed politically untenable — so draws continue at ~0.6 mb/d (scenario pace, just under the recent post-plateau average) through the elections, taper, and bottom at ~177 Mb in late February 2027, brushing the top of the 100–150 Mb analyst-estimated unusable band: the SPR effectively runs out of usable barrels. No returns while the war continues. Authorized capacity 713.5 Mb. Historical context (2022–26 drawdown and refill) in the table view above.

Weekly draw rate (mb/d) — how fast the barrels actually left

How to readEach bar is the level chart's weekly change ÷ 7 — a derivative, not a survey; a DOE-led week shows as a labelled provisional bar. The diamond line is the 4-week trailing mean — the shortest window this page quotes a trend from, because the weekly series is too noisy to carry one.

Table view (weekly draw rate since the drawdown began)

Week ending	SPR (Mb)	Draw rate (mb/d)	4-wk pace (mb/d)	Change (Mb over week)

The draw plateaued at ~1.26 mb/d through June 19, faded through July to 0.41, and briefly re-accelerated to 0.87 in early August. The week ending August 28 printed 3.1 Mb, 0.45 mb/d (286.6 Mb — lowest since November 1982), below the 0.73 mb/d the announced schedule implies, the 2 most recent weeks in a row. The four-week pace is 0.65 mb/d, still 49% off its 1.27 May peak — still well off the plateau. At the latest ~0.45 mb/d the remaining 43.2 Mb completes around Dec 3.

Methodology & caveats (basis, pace readings)

First derivative of the level chart (weekly change in EIA WCSSTUS1, ÷7 to mb/d; when DOE's early Monday print leads, it appears as a labelled provisional bar until EIA's Wednesday release). The post-plateau weeks ran 0.79, 0.89, 0.43, 0.72, 0.54, 0.41, 0.87, 0.76 and 0.45 — a series too noisy week-to-week to carry a trend claim, which is why the four-week pace is the shortest window quoted anywhere on this page. Four candidate causes, and the weekly series is net, so it cannot separate them: a deliberate stretch-out ahead of the statutory floor and the midterms; withdrawal-rate decay with fill and cavern condition; netting against early exchange returns (contractually post-war, and hostilities resumed 8 Jul); or an ordinary end-of-programme taper, which the timing argues against — this is day 142 of a "~120-day" programme with 36% undelivered. What would separate them: DOE's monthly SPR inventory and movement reports, or any DOE/GAO statement on achievable rate at sub-300 Mb fill. Weekly SPR movements reflect cavern delivery scheduling as much as policy intent (corroboration: factually.co, accessed 2026-07-21).

The 400 Mb collective release: who committed what — and the repayment arithmetic (Mb)

So what — the biggest release, and the repayment is still unsold The largest collective action ever ran 63% delivered by 12 June, and the loan book DOE says returns “~200 Mb within a year” is only 133 Mb awarded — ≈170 Mb with its premium — the release side is a measured fact and the return side is a contract that has to be signed before it can be broken; the two bars are drawn on one scale so the gap between them is the thing you see.

How to readThe vertical ticks are reference marks, not segments: delivered-so-far on the commitment bar, DOE's 200 Mb return claim on the loan-book bar.

Bottom bar: DOE's "~200 Mb returns within a year" is not bad arithmetic — at the achieved 26% premium the full 172 Mb would return ~219 Mb. It is an unsold loan book: only 133 Mb has been awarded, across tenders that cleared 52–87%, and the contracts return through September 2028 rather than within a year.

How the gap was bridged: production, stock draws, and oil-on-water, Feb–Jun 2026

So what — demand did most of the work Stocks and cargoes at sea covered at most 4.6 mb/d in any month; the shaded gap demand destruction absorbed runs 3.7–7.3 mb/d — the buffers bought time, but the market was balanced mostly by people buying less, which is a form of adjustment nobody chose and nobody can bank for next time.

How to readBars are month-on-month changes (left axis); lines are levels (right axis), all mb/d — hatched bars are inferred, the gray bar an unpublished split.

Table view

Month (2026)	Production shipped/stored	+ Onshore reserve draws	+ In transit (oil-on-water)	Net stock draw	= Delivered
Feb	106.9	~0	~0	~0	—
Mar	97.0	+0.4	+3.8 (draining to ports)	+4.2	101.2
Apr	95.1	+4.2 (inferred)	−1.8 (inferred)	+2.5	97.6
May	94.5	split not published; not inferable (no water figure)		+4.6	99.1
Jun	98.8	+3.2	−3.9 (armada reloading)	−0.7	98.1
Jul	101.5	+0.2	+2.0 (armada delivering)	+2.2	103.7

March's cushion was mostly pre-war cargoes draining into ports — tankers, not tanks. The shaded distance between delivered supply and the pre-war demand path is what demand destruction absorbed: 3.7–7.3 mb/d a month, more than stocks and water combined.

Methodology & caveats (inferred splits, vintages)

The blue bar is the month-on-month change in IEA world supply, net of shut-ins — the −9.9 in March is the largest monthly drop ever recorded; green is what onshore reserve draws added; orange is the oil-in-transit contribution (positive = cargoes at sea draining into ports, negative = floating storage absorbing supply). Delivered supply = production + net observed inventory change (−129 Mb Mar, −74 Apr revised, −143 May, +21 Jun, −69 Jul — of which onshore only −6, the IEA releases winding down). April's components are inferred (hatched): the IEA never re-published the split after revising April's total, but the tanker-tracked on-water figure (+53 Mb, May OMR) survives, so onshore is the residual (+4.2 draw, −1.8 absorbed) — mixed vintages, treat as estimate. May has no published water figure, so one gray net bar. June detail: oil-on-water +117 Mb, ~35 Mb exiting Hormuz within a week per Kpler, while consumers kept draining tanks (+3.2 incl. 44 Mb government). All figures are estimates subject to revision.

Every US cache, not just the SPR: net draw since the pre-war peak (Mb)

Round-2 stress test: how long would the buffers last this time?

So what — a third of the runway is gone — round one was fought with a full tank; the same shock now starts from two-thirds of it, which is why the scenario prices sit higher than the round-one prints they are calibrated against.

Scenario modeling. Usable buffer = SPR after the 100–150 Mb cavern-condition haircut and the 43.2 Mb still-committed release (EIA weekly print of Aug 28), plus other-IEA government stocks and US commercial excess; months measure endurance at the 2.1 mb/d release ceiling — price closes the rest.

Methodology & caveats (components, judgment calls)

A stricter definition that also reclassifies the 43.2 Mb still-committed SPR leg out of "usable" gives a larger decline. Other-IEA government stocks are approximated from IEA aggregates (~800 Mb Feb → ~637 Mb now). Counting US commercial stocks above their 2015–25 floors is an author judgment call — excluding it gives ~17 vs ~12 months. Release contribution is capped at the 2.1 mb/d maximum pace observed in April; the shut-in range 8–11 mb/d spans the round-1 June average but sits below the observed peaks.

What America actually burns: 20 mb/d, and why the total barely moves

Is demand being ground down? 2026 against its own calendar, fuel by fuel

So what — the overshoot has unwound — the flat-looking total above hides a turn. Measured against what the same weeks of 2021–25 did, the country opened the year consuming far above normal, gave all of it back by midsummer, and is now running below. The fuel mix says who: gasoline and propane are giving up barrels, the petrochemical bucket has fallen back to its calendar, and jet alone is still flying above it.

Weekly deviation from the 2021–25 same-week average (kb/d) — above the line, running hot; below it, giving up barrels

How to readEach bar is one week of 2026: segments above the zero line are fuels running hot against their 2021–25 same-week average, segments below are fuels giving up barrels, and the diamonds are the total — the segments sum to it by identity. Everything is smoothed over four trailing weeks.

Table view (weekly deviation from the 2021–25 same-week average, kb/d, 4-week smoothed)

Week	Gasoline	Distillate	Jet	Propane	Other oils & resid	Total (smoothed)	Total (raw week)

Methodology & caveats (why a five-year average flatters jet, and what the smoothing hides)

A five-year average mixes trend and shock. Jet fuel and the "other oils" bucket are still growing year over year, so they would sit above a 2021–25 average in any year, war or no war. The signal in this panel is the change across 2026 — from far above the calendar to below it — not the level of any single bar. Anchored on last year alone the conclusion holds: the latest eight weeks run  kb/d below the same weeks of 2025.

Every segment is a published EIA series — the six product-supplied components sum to the published total to EIA's own rounding, so the panel carries no derived leftover bar. "Other oils & resid" (other oils plus residual fuel oil) is the least interpretable of them: it is the big NGL and petrochemical-feedstock bucket, and ethane or butane disappearance moves with crackers and blending, not with anyone's commute. Deviations are smoothed over four trailing weeks because single prints of this survey are noise; the first three weeks of January therefore fall away, and the newest bar leans hardest on the newest single print.

The US crude balance: what fills the tanks, and where it comes from

So what — the wrong ocean — imports are a third of US crude supply, so the exposure question is worth asking. The answer is that the barrels come from this hemisphere: Canada alone is two-thirds of them and arrives overland. A Red Sea or Hormuz disruption reaches the US through the world price, which is what every other panel on this site is about — not through the import mix.

The weekly balance (kb/d) — supply above, use below

How to readThe stacked terms, EIA's adjustment included, sum by identity to the diamond — each week's observed change in crude stocks, commercial plus SPR.

Table view (the weekly balance, every term)

Week	Production	Imports	Exports	Refinery runs	= Stock change	of which EIA adjustment

Why an unaccounted bar is drawn. EIA publishes a weekly crude supply adjustment because its inputs come from different places: production is modelled, imports from customs, runs and stocks from separate refiner and terminal surveys, and condensate blended into the crude stream is not fully reported. In 2026 it averages +0.6 mb/d and reaches +1.9 in its widest week, so a reader adding the terms by hand would find roughly one to two million barrels a day missing if it were hidden. It is a residual, not a flow.

And a whole stream sits outside this balance. Refined product exports run 7.5 mb/d — larger than crude exports at 3.6 — but they leave after refining, not before it, so they belong to no crude balance. The middle panel draws them. Together the US exports about 11 mb/d of crude and products, against 6.2 of crude imported: on a net basis the country is a large exporter that nonetheless buys 6 mb/d of the heavy grades its refineries are built for.

Methodology & caveats (the adjustment term, two cadences, and Cushing)

The identity is production + imports − exports − refinery runs = the change in stocks (commercial plus SPR). It closes to within EIA's own weekly balancing item, which averages 484 kb/d here — about 2.5% of the supply base — and which is drawn as its own term rather than folded into the others, because a balance chart that silently absorbs its residual asserts a precision the weekly survey does not have.

Two cadences, and they must not be quoted as one. The balance is weekly; imports by origin are monthly and publish about two months behind (origin month 2026-05 against balance week 2026-07-31). Lane assignment is by the route a cargo must use, not by observed voyages: Canadian and Mexican barrels are overwhelmingly but not exclusively overland — some Canadian crude reaches the West Coast by sea. Iraq and the UAE have no by-origin series at this endpoint, so they sit inside the 9.6% unnamed residual, which makes the Hormuz share a floor rather than an exact figure.

On Cushing. Cushing is landlocked — an Oklahoma pipeline hub some 500 miles from the coast, so no import has ever arrived there by ship. In the week to Aug 14 it fell 1.31 Mb while the Gulf Coast, where imports do land, built 8.4 Mb with seaborne imports down 0.7 mb/d on the week — what Cushing holds is set by pipeline nominations out of the coast and the mid-continent, not by ships. One week of PADD-level data is noisy enough that we would not build a claim on it — the durable point is the geography above, not the single print (the stock panels).

Do prices predict US crude exports? Yes — with a lag

So what — the cargo pipeline — higher Brent widens the WTI–Brent spread, pulling more US barrels onto the water. The response is not instant: it takes about four weeks for a price signal to become a loaded tanker, which is why the price already printed says something about next month's exports.

Monthly US crude exports against Brent one month earlier — and the month the fit implies next

How to readEach dot is a month: exports on the vertical, the Brent average of the month before on the horizontal. The dashed line is the 2026 fit; the hollow marker at the end of it is what that fit implies for next month, read at the latest four-week Brent average — a prediction from prices already printed, not from anything still to happen.

Methodology & caveats

The positive sign is the WTI–Brent arbitrage: higher Brent widens the spread, which is the economic incentive to load US barrels for export. The lag is physical — cargo scheduling, terminal queuing, and voyage time bake in a delay from price signal to loaded tanker. On weekly data the correlation between Brent and exports peaks at a 4-week lag (r = 0.64 in 2026, against 0.37 paired same-week); the monthly scatter is drawn because weekly export data is noisy (customs and survey timing). The level fit suggests about 265 kb/d of extra exports per $10/bbl of Brent, with that delay.

Forecast implication.

US stocks vs the 5-year range (Mb) — where winter starts from

So what — three caches below range, crude back inside — every barrel of this is the cushion the winter draws down from. Cushing, gasoline and distillate start below their own five-year ranges — but two consecutive August builds have put commercial crude back inside its band, essentially at its same-week average, so the crude leg of the deficit story is over for now. Propane is the exception in the other direction: far above its range, and made from gas processing rather than crude, which is why it can build while the others draw.

How unusual each one is, on one scale

How to readEach bar is that cache's percent distance from its own 2021–25 same-week average — deviations, not levels.

Each one against its own five-year range

Gasoline and especially distillate enter the winter season at or below their 5-yr bands, and Cushing is 3.3 Mb above operational tank bottoms — the starting conditions behind the risk case. Each panel is on its own scale, in Mb; the ranked chart above is the one place the five are comparable. Propane is the exception and worth reading separately: it is made from natural gas processing, not crude refining, so it does not compete for the barrels the other four draw on — and it accounted for roughly two-thirds of the commercial stock build that halted the national draw in July (every US cache).

The fuel ladder: every fuel in $/MMBtu, ranked — and the Brent price at which oil stops beating each (late July 2026; individual prints Jul 7–22)

So what — gas-to-coal is the only fast lever ≈1.0–1.2 mb/d-e can shift from gas to coal within months — the one big fast lever — while gas-to-oil switching pushes the other way, adding oil demand.

How to readBars are Brent-equivalents in $/bbl, not native prices — a bar ending left of the dashed Brent-spot line marks a fuel already cheaper per Btu.

Table view (native price, $/MMBtu, as-of and source per fuel)

Fuel	Native price	$/MMBtu	Oil beats it below (Brent, $/bbl)	As-of	Source

Methodology & caveats (conversions, what parity ignores)

The lever, sized: 55–65 bcm/yr of IEA gas-to-coal potential, with caps and bans already lifted across Asia and ~1.8 bcm realized by June. Gas-to-oil switching adds up to a few hundred kb/d of oil demand while Brent sits below the ~$102 JKM crossover. LNG cannot fill in from elsewhere: the non-Gulf ramp is already spent (+27 bcm Mar–Jun ≈ 0.5 mb/d-e, ~¾ of the Gulf hole) and the next real increment — the 2027–28 wave, ~+50 bcm/yr ≈ 0.9 mb/d-e — arrives after this war, not during it. Pure conversion arithmetic on sourced price prints: crude 5.8 MMBtu/bbl, VLSFO 38.5 and MGO 43.5 MMBtu/t, coal 23.8 MMBtu/t (6,000 kcal/kg NAR, the ICE Newcastle/API2 spec — not the 25.1 of the older 6,330 GAR basis). Btu parity ignores plant-efficiency spreads (a 50%-efficient CCGT vs a 35% coal or oil steam unit shifts power-sector thresholds in gas's favor), carbon (EUA €75 adds ~$3.2/MMBtu to coal, ~$1.6 to gas in the EU), and logistics adders (LNG bunkering +$1–1.5/MMBtu). Product versions of the same test: Singapore VLSFO $638.50/t ≈ $16.6/MMBtu (≈ the front-month crude line), MGO $905.50/t ≈ $20.8 — burning residual fuel oil beats spot LNG by a hair; burning diesel does not. Sources and access dates: linked per print in the table view above.

Wells back in the black: WTI breakevens vs the price (survey averages, $/bbl)

So what — price stopped being the constraint ≈+0.1 mb/d in weeks (DUCs) · +0.25–0.5 mb/d by end-2027 — at $100 or at $150 — price stopped being the constraint at ~$70.

How to readEach range spans group averages — regional and firm-size splits, not firm-level dispersion; the dashed line is the latest FRED WTI close.

Table view (every region and firm-size split, Dallas Fed Q1-26 / KC Fed Q2-26)

Threshold	Region / group	WTI ($/bbl)	Note

The "back in the black" margin is the wrong margin: everything already produces above ~$50, so there is no shut-in inventory waiting on price. The margin that matters is new drilling ($62–70), crossed in March — what follows is what producers did with it. KC Fed's Tenth District puts profitable drilling at $66, inside the Dallas Fed band and from a different survey, so the two corroborate each other; it is left off the chart for that reason rather than drawn as a fourth row.

Methodology & caveats (opex floors, strippers)

What binds instead of price: capital discipline, record-low DUCs, and a strip capped by the 2027-surplus outlook. Nothing is shut in waiting for higher prices — everything produces above ~$50 — so "back in the black" buys ~0 extra barrels. Dallas Fed: "All respondents can cover operating expenses for existing wells at current prices." WTI bottomed at $55.44 (Dec 16, 2025) and spent only ~3 weeks below $58 in 2025–26 — still far above every opex floor; the ~420k stripper wells producing ~915 kb/d stayed on throughout, so the reactivation channel is worth tens of kb/d at most. New-drilling breakevens are up 30–40% from 2020 on cost re-inflation.

…and how many barrels the price signal actually buys (incremental US supply, mb/d)

How to readEach bar is one published estimate's range — survey rows are producers' own modal answers, model rows Rystad scenarios; hover a row for its source.

The observed 2026 record matches the low rows: WTI spent 43 of 51 post-escalation days above $90 while the satellite rig count fell — rigs turned only when the strip validated ~$75–80 (+45 rigs in six weeks from mid-June). Above ~$70 the cost curve stops binding at all; capital allocation, service capacity and lead time bind instead.

Methodology & caveats (survey basis, rigs, cash flows)

Survey rows are producers' own modal answers (Dallas Fed, 120–125 firms); model rows are Rystad scenarios. Kayrros satellite rig count fell 534 → 462 Jan–Apr (Baker Hughes flat ~543–545) — 2026 budgets were built at $55–60 and the spike was read as temporary; the turn came at the KC Fed's $83 "substantial increase" threshold with the classic 4–6-month lag (588 rigs Jul 17, +44 y/y). The DUC cushion is at a record low (4,972 in April). At $100 avg WTI, ~$63 bn of extra 2026 shale FCF (Rystad) is earmarked for payouts and debt — Dallas Fed Q2 "expected capex next year" index: 0.0. Non-US short cycle is slower still: oil sands ~1 mb/d over ~7 years, Vaca Muerta infrastructure-paced; Rystad's average new-project breakeven is $47.

OPEC+: quota theater vs deliverable barrels (mb/d)

So what — the quotas moved, the barrels did not 0.02 mb/d (what's actually spare) to ~1.5–2 mb/d (only if it can reach water) — higher prices change OPEC+'s press releases, not its deliveries.

How to readEach row is a different measure — estimator, date and definition all differ — on one mb/d axis; never blend or sum rows, hover for each basis.

The two smallest rows are the point: the world's shock absorber sat inside the chokepoint that failed — EIA surplus capacity (Table 3d) collapses to 0.02 mb/d for Q2–Q4 2026. What can move outside the strait is Petroline/Yanbu plus ADCOP, and in 2026 that was measured, not estimated: Yanbu went 0.97 → 4.2 mb/d and Fujairah ~0.5 → ~1.8, so ~4.5 mb/d of diversion happened — and the UAE took ADCOP out of OPEC on May 1.

Methodology & caveats (estimators, logistics)

Why nothing arrives at the margin: the cadence is pre-planned, the spare sits behind the strait, and the fast rungs (Petroline surge, ADCOP) were pulled in March — until transit clears, at which point the 2027 surplus, not scarcity, becomes the problem. EIA surplus capacity ran 3.0–3.6 mb/d a year earlier (3.9 in early 2025) and recovers to ~2.4 only in 2027; the IEA's March estimate of effective spare — a different estimator, shown separately, never blended — fell 3.6 mb/d in one month to 0.32, the lowest on record. Petroline ran a record ~7 mb/d throughput with ~5 claimed exportable. The binding constraint is the terminal, not chartering: Kpler measures Yanbu at 4.2 (Apr) and 4.1 (Jun) against a 4.5–4.7 sustainable ceiling. Quarterly series in the table below.

Table view (EIA STEO surplus capacity by quarter, 2025–2027)

Quarter	OPEC surplus capacity (mb/d)	Status

The reaction function: what OPEC+ actually did at each price

Table view (meeting by meeting: tranche, Brent context, outcome)

Meeting	For	Tranche (kb/d)	Brent context	What happened

Six consecutive pre-planned hikes straight through a $126.41 intraday peak and back to $72: OPEC+ sells insurance against physical shortage, never against risk premium — and 2026's shortage is one it cannot serve. The Aug 2 meeting approved the sixth (+188 kb/d for September), completing the 1.65 mb/d voluntary-cut rollback, and signalled quotas steady through year-end. What to watch: the Sep 6 meeting and whether the next layer — the 2.0 mb/d Oct-2022 group cut — gets touched; all of it is paper until the strait clears.

Methodology & caveats (historical templates, forward calendar)

Same reaction function as March 2022 (a <15-minute meeting approving the pre-planned +400 kb/d at $120+ Brent). The historical templates — 2004–08 (produced flat-out, no cushion, price found $147) and 2011 Libya (Saudi +700 kb/d while the IEA bridged with 60 Mb) — both required barrels that could reach water; this time consumer stocks did the work again (252 Mb delivered by Jun 12). The September tranche approved Aug 2 completed the 1.65 mb/d unwind, with the 2.0 mb/d Oct-2022 group cut flagged from as early as October. The flip side caps the upside: IEA's August balances still imply a ~5 mb/d surplus in 2027 (supply rebounding +8.3 to 110.3 against demand recovering +2.4) if transit normalizes — which pins Dec-27 futures near $77.04 and, via the strip, keeps the shale response asleep.

Open everything? The frontier scorecard — and every lever on one chart

So what — the experiment already ran ≈0 mb/d before 2030 from frontier acreage — the experiment already ran, mid-war — ANWR's June 5 lease sale, held with Brent in the $90s–100s, drew $3.7M and zero majors.

The frontier scorecard, lever by lever (sales, permits, DPA, federal share)

Lever	What happened when it was pulled	Barrels, and when

The revealed preference is precise: industry paid a record $164.7M for NPR-A acreage next to existing roads and pipe in March, and eleven weeks later just $3.7M for ANWR-adjacent frontier. Where policy did deliver barrels in months — the Venezuela licenses — the capacity already existed and was merely switched off, the general rule of every row below.

Methodology & caveats (timelines, license detail)

The sale ran under mandated-sale law and 28-day permitting. Even a funded ANWR is first oil ~10 years after leasing, peaking ~880 kb/d in the 2040s (EIA's own model); with >7,500 approved federal permits already sitting unused, acreage was never the constraint — capital, whipsaw risk and the strip are. Even the NPR-A barrels are a 2030s story: Willow, fully supported and half-built, runs FID Dec 2023 → first oil early 2029 → 180 kb/d. The Venezuela licenses (GL 49/50, Feb 13 2026; amended to GL 50A Feb 18) restored output to ~1.07 mb/d by May.

Time to barrels, next four years: the cumulative track — and the buffer dropping out

How to readLeft to right is the next four years of cumulative supply and relief; the hatched OPEC-spare bar is drawn but never counted — it joins only if those barrels can reach water.

The demand waterfall: where 1.4 mb/d of 2026 demand went (change vs the pre-war forecast, mb/d)

So what — the cheap half is already spent −1.37 mb/d on the 2026 average — half of it invisible (idle crackers), and the cheap half is already spent — the elastic channels fired first; if scenario A persists past the SPR cliff, the next ~3–8 mb/d (the scenario-A uncovered gap once the release leg ends, mid 6.4 — author scenario) must come from deeper rationing or recession, which is why the fitted price curve steepens as buffers empty.

Table view (the forecast erosion, vintage by vintage — and the OPEC/EIA counterpoints)

Vintage	2026 demand (mb/d)	Print

Where it landed: the same loss, four lived experiences

Red = rationed by decree or absence, amber = industry shut on margin, purple = the state stopped absorbing it — subsidy or stabilisation buffer cut, capped or lifted, so the price arrived late and all at once, green = merely trimmed by price; each tile states its own basis and as-of date.

The purple tiles are the second round's signature: governments that held the pump price through the spring with a subsidy or a stabilisation fund, and cut, capped or lifted it once the strait shut again. Each tile links its sources; the same grid on the Overview carries the distributional argument and the lifelines it ran through.

Three mechanisms, one label — and what snaps back

So what — most of it reverts, the EV pull-forward does not Most of it reverts when supply does; ~0.5–0.9 mb/d never comes back — the IEA already pencils demand +8 mb/d off the May low by October and +2.0 mb/d in 2027. What's permanent: the EV pull-forward (China's fleet alone displaces 540 kb/d of 2026 gasoline) and the two-year path netting ~0.7–0.9 mb/d below the pre-war trend. What's not even destruction: China's import collapse (−3.6 mb/d Feb→Apr) was mostly stock-financed deferral — most of it returns as the 2027 restocking bid, minus the slice EVs, LNG trucks, and the construction slide retire each year (decomposed in the China section of the Supply tab).

The distributional reading is the ugly one: the market cleared by poverty and idle factories, not rich-world inconvenience. Only ~25–30% of global gasoline/gasoil demand sees full market price pass-through (IEA) — the rest adjusts through subsidy budgets, FX crises, and queues.

Second-order: the fertilizer → food transmission (the shock that lags a season)

So what — an affordability squeeze, not yet a food crisis An affordability squeeze so far, not a food-price crisis — the test window is Oct 2026–2027 harvests — urea doubled then halved, and FAO's food index sits at 130 vs 160 at the 2022 peak.

How to readThe line stitches mixed prints — World Bank averages, Indian tenders, regional quotes — all in $/t; hover each point for its own basis and source.

Table view (supply-side anchors, each with its linked source)

Anchor	Value	As-of	Basis

Why it retraced: prices halved because China reopened urea exports, SABIC opened a Red Sea route at Yanbu, and the northern-hemisphere season ended — every one of those reliefs is fragile. The thing to watch for is a 2022-style food-export ban cascade; so far the 2026 cascade is only in inputs.

Methodology & caveats (supply shares, scenarios, pass-through)

The arc: $400 → >$850 at the April peak (India's record tender hit $935/t, ~86% above its February basis per AMIS), then halved as China reopened exports and Saudi rerouted via the Red Sea. Fertilizer works on planting calendars, so if round two runs through the autumn, India's rabi top-dressing (Nov–Jan), Brazil's September window (~30% covered) and East Africa's next planting take the yield hit into 2027. The Hormuz strait carries 32–36% of world urea trade (CRU puts it near 40% including Iran; a 45% figure in circulation is an outlier) — and the shock has since migrated from nitrogen to phosphate: urea is back below its February level while DAP and TSP, though both off a June peak, are still well above it — and the rock benchmark that sat unmoved through the whole doubling has now begun to reprice (the nutrient split). QAFCO's 5.6 Mt/yr plant (14% of traded urea) shut March 4 and Iran — the world's #3 urea exporter — idled all seven complexes. The China quotas are only 1.5–1.6 Mt (Jun–Aug, $660–670/t floor) and close after August; Yanbu is the same Red Sea chokepoint the tanker war threatens; the farmdoc extended-conflict scenario has US urea peaking near $996/st in October — right at the fall-application and rabi windows. Pass-through template: IMF put a 10% fertilizer rise at ~7% on cereal prices after one quarter. Each anchor links its source in the table above; two low-confidence secondary figures are excluded.

The bypass got its own blockade — and the barrels turned north instead of stopping

So what — a detour, not a loss Yanbu kept loading through July; Bab el-Mandeb transits fell 22–30% and Suez-routed Saudi crude rose 106% — but the northern lane is capacity-short of the volume it now has to carry.

On 20 July the Houthis declared a maritime embargo on Saudi Arabia — not a closure of the strait. The rule is written on the port call, not the position: a vessel that loads at a Saudi port may be targeted "in any location within the operational reach of the Yemeni Armed Forces." That distinction decides everything that followed. Their reach is southern — every corroborated strike sits near 17°N, and the missiles fired at Yanbu on 25 July were intercepted — so a cargo is safe at the Yanbu berth and hunted on the way out through Bab el-Mandeb.

So the barrels stopped turning left. Kpler puts Sidi Kerir loadings up 250 kb/d to 1.3 mb/d in the week to 27 July — its highest weekly level since the 2.2 mb/d of the opening week of the US–Iran war, apart from a brief 1.38 in late March — and Saudi crude reaching Asia via Suez up 106% to 1.06 mb/d. Aramco began offering spot cargoes ex-Sidi Kerir. Traffic through Bab el-Mandeb fell 22–30% on seven-day means (Windward −22%, IMF PortWatch −30%), with tankers down harder at 36–39%; the widely-quoted −56% is a real Kpler count but compares two single days in a series that swings between 20 and 46 with no embargo at all. The corridor did not close — 20 vessels still transited on the worst day.

The constraint on the northern lane is arithmetic, not threat. SUMED is rated 2.5 mb/d but has never been tested at capacity: its historical peak throughput is 1.77 mb/d, set in 2016. Against that, roughly 3.8 mb/d needs a home. And because every laden tanker sent north has to be matched by tanker capacity sailing back in, Kpler puts the full replacement programme at 3.9–5.1 southbound ballast transits a day — against 3.3 a day, the busiest stretch Kpler has ever recorded, in August–September 2022. The bypass has a bypass, and it is smaller than the flow.

Methodology & caveats (why the Yanbu volume estimates disagree by 1.4 mb/d)

Trackers conflict sharply on the week of 20 July and must not be averaged: Kpler and AXSMarine and Signal Ocean put Yanbu at 2.4–3.0 mb/d (−29% to −43% against Kpler's 4.23 the prior week), Windward at 3.09 — though Windward measures against a 5.16 mb/d baseline, 22% above Kpler's, so its "−40%" and Kpler's "−29%" are not comparable declines — and Vortexa at 3.8 mb/d and "broadly stable." The divergence has a stated mechanical cause — Vortexa observed four VLCCs, a Suezmax and an Aframax loading with transponders off, about a third of the week's volume, and Windward counted 11 dark tankers at Yanbu berths on 26 July plus 43 AIS gaps in the Arabian Gulf for 15–22 July, the highest reading in four years — that last figure is the wrong sea, so it is context for dark-fleet behaviour rather than evidence about Yanbu. AIS-based counts therefore understate by an unknown margin and the true figure plausibly sits at the Vortexa end. A separate and larger dispute: Wood Mackenzie puts the Red Sea bypass at ~2.39 mb/d in June (−41% from March), irreconcilable with Kpler's 4.1 for the same month. Both are named-provider estimates; both are published here; neither is reconciled. Aramco has declared no force majeure and confirmed no impact on crude loadings. The August OSP cut (Arab Light to Asia −$11) predates the embargo by two weeks and cannot be read as a response to it.

Six Hormuz numbers, one strait — which is which

So what — the denominators are not interchangeable Every figure below is right on its own basis and none of them can be compared to another without conversion — the two that get mixed up most often are ~20 mb/d of total oil transit and >15 mb/d on the exports basis, which differ by whether products and NGLs are counted.

Two conversions worth having: the ~20 total-transit basis is ≈25% of world seaborne oil trade, while the 17% figure on the ledger below is the share of that trade held captive — transit minus what the bypasses move. And the closure duration depends on which start you pick: de facto 28 Feb → 17 Jun is 109 days, the declared 2 Mar → 17 Jun is 107.

Every pipeline out of the Gulf was built for crude — share of each cargo with a route that avoids the strait

So what — every pipe out was built for crude Hormuz alone holds 17% of world seaborne oil trade captive; the Red Sea embargo takes another 5 points on top — and the one demonstrated non-crude bypass, Bahrain's aluminium trucked to Jeddah, lands inside that same blockade zone. Everything is scaled to world supply of its own cargo, so the rows are comparable.

How to readEach bar is a share of that cargo's own world supply — rows compare with each other; they never sum.

Table view (every row's route, fate and caveat)

Cargo	Hormuz flow	Route out	Basis	Fate	The route, or why there isn't one

Methodology & caveats (what the percentages are shares of)

The x-axis is a share of each cargo's own flow, not of the 19.87 mb/d oil total — the rows are not slices of one pie and the bars must not be summed. Only the eight rows with a published world-supply share are drawn. The other five are named in the line under the chart and carried in full in the table view — a row with no denominator gets a name, not a bar. Crude appears twice on purpose: the IEA's 3.5–5.5 mb/d is untested nameplate against 14.95 mb/d of crude transit, while the second crude bar is the measured outcome: Yanbu went from 0.97 mb/d in 2025 to 4.2 in April 2026 and Fujairah to ~1.8, so about 4.5 mb/d of diversion actually happened — inside the IEA's band. The two bars therefore read as capacity against realized — and the realized number is the sharper fact: the bypass ran at nameplate and still left ~70% of the Hormuz crude flow with no route at any price, while every non-crude cargo stayed at zero. (The corridor explorer shows the same barrels moving.) LPG/NGL is marked non-additive: those volumes already sit inside the 4.93 mb/d products line, and the 300 kb/d shown is a pipeline capacity, not a flow. Sulphur is drawn as an unmeasured band rather than a number because the Etihad Rail route to Fujairah demonstrably exists — which refutes the "Gulf sulphur has no bypass" claim — but its 2026 utilisation is not published. Aluminium's 40–60% is road freight for one producer and does not generalise to bulk cargo. Oman's ports (Sohar, Duqm, Salalah) are outside the strait; sources that code Oman as fully exposed are wrong.

Brent gave back most of its war premium. The cargo with no bypass did not.

So what — the shock is seaborne — and US natural gas, the same molecule on the far side of an ocean, never had a war premium at all.

How to readEvery line is a monthly average indexed to its own February 2026 (= 100) — IMF Primary Commodity Prices via FRED, and for Brent the FRED daily spot averaged to months — one basis, one base. The last point on the European gas and aluminium lines is the World Bank Pink Sheet's August print, a month ahead of the IMF series and labelled on the chart.

Table view (levels in native units, and the share of the premium given back)

Series	Feb 2026 (base)	Peak	Latest	vs base	Premium given back	Basis

Methodology & caveats (proxies, coal's spec, and the propane trap)

Every non-crude series is IMF Primary Commodity Prices via FRED; Brent is the FRED daily spot the price tabs quote, averaged to months, so the panel shares one monthly-average basis. Where the World Bank Pink Sheet has printed a month the IMF series has not yet reached and the two agree within 3% over the past year — European gas (within 1.9%) and aluminium (within 0.3%) — that month is drawn from the Pink Sheet, labelled on the chart and named in the hovercard, and gives way to the IMF figure when it prints; the retracement figures stay on the common window. Coal and Henry Hub have Pink Sheet counterparts that do not agree that closely (coal by 7.9% on a different spec, Henry Hub by 82.9%), so they wait for the IMF print. Brent's monthly mean agrees with the Pink Sheet within 2.7% but needs no lead: the daily series behind it runs ahead of both, so the Brent line already carries August from its own mean. The Pink Sheet's LNG series is Japan's import contract price, a different measurement from the IMF spot marker, so the Asian LNG line is never spliced and runs a month behind European gas. Brent is that daily series as a monthly mean, never the daily close itself — mixing a monthly average with a daily close inside an indexed comparison manufactures divergence, which is why Brent's April here is 165, a month's mean, while the price tabs' single-day peak is $138.21. JKM and TTF are not on this chart: FRED carries neither, so PNGASJPUSDM ("LNG, Asia") and PNGASEUUSDM ("Natural gas, Europe") stand in as free proxies and are labelled as such. Coal is indexed only. FRED's series is IMF's 12,000 Btu/lb (26.45 MMBtu/t) spec, not the 6,000 kcal NAR (23.8 MMBtu/t) spec behind the fuel ladder; converting it here would overstate it ~11% and put the two panels on different bases. Propane is a basis trap: on the daily window from 5 January it lags Brent by 28 points, while on monthly averages from the February base it ran within forty points of Brent through the spring and briefly led it in June, when Brent's monthly mean fell faster than propane's — the sign of the comparison depended on the window you chose. Since July propane lags on both windows, but neither figure means anything without naming its window.

Europe's Qatari LNG went to exactly zero in April. Its Russian LNG did not.

So what — exposure ≠ pain — Europe's molecular exposure to Qatar was single-digit and it vanished in one month; the pain since has been pure price transmission.

How to readEurostat LNG specifically — not pipeline gas — in million m³/month; the two Qatari reporting codes are both drawn, and their sum is the Qatari total.

Table view (partner split by month, mio m³)

Month	Qatar (QA)	Qatar via other	Qatar total	Qatar share	Russia	United States	All partners

Methodology & caveats (the partner trap)

Eurostat nrg_ti_gasm, siec=G3200 — which is liquefied natural gas specifically, not total gas, so the Russian series here is Russian LNG and not TurkStream pipeline volume. The partner trap: Eurostat's QA code does not capture every Qatari cargo — Poland's land under "Other Near and Middle East Asian countries". Reading QA alone undercounts Europe's Qatari share by roughly 2 points, so both are charted and the Qatari total is their sum. That aggregate may contain non-Qatari origins; it is attributed here because it goes to exactly 0.0 in the same month QA does. The IEA independently puts Hormuz LNG at "just around 7% of Europe's total LNG inflows", which brackets the 6.3–8.1% computed here. Volumes are million cubic metres of gas, not tonnes of LNG.

Nitrogen round-tripped. Phosphate didn't — and phosphate's driver is Gulf sulphur.

So what — the lagged channel — the food test window is the Oct 2026 and 2027 harvests, and it is still open.

How to readFive World Bank monthly benchmarks in $/t, each on its own pricing basis (hover a line for it) — so turns compare exactly, levels only loosely.

Table view (each benchmark with its own pricing basis)

Benchmark	Feb 2026	Peak	Jun 2026	vs Feb	Pricing basis

Methodology & caveats (why the bases can't be combined)

World Bank Pink Sheet monthly, and every column has a different basis: urea is fob Middle East, DAP is fob US Gulf, TSP is import US Gulf, phosphate rock is an administered contract quote fob North Africa, potash is Brazil CFR granular. So they cannot be combined into an input-cost decomposition: World Bank DAP prices off Tampa sulphur and Tampa ammonia, not the Gulf inputs this tab is about, and any waterfall built against Qatari sulphur would fit by coincidence. Flat rock was never abundance — it is an administered benchmark that did not reprice, which is precisely what isolated processing cost as the driver through the doubling, and roughly half of world traded sulphur is a by-product of Gulf sour-gas processing with no supply elasticity and no route out. That window has now closed: rock moved in June and again in July, so from here a phosphate move can no longer be read as processing cost alone. Note also that the FAO headline food index has risen mainly on vegetable oils (+10.2% since February), not cereals (+1.4%) — a different channel than fertilizer.

One percentage, every cargo: war risk as the universal multiplier

So what — priced once, paid by every cargo — insurance is not one more affected market; it is a single percentage applied simultaneously to crude, LNG, urea, wheat and washing machines.

How to readEvery print is one broker's assessment, in % of hull value per voyage; the two panels share that unit but not an axis — the lanes run an order of magnitude below the strait.

Table view (every print with its source and confidence)

Reading	As-of	% of hull value	Conf.	Source	Note

Methodology & caveats (the baseline dispute, and the $21m that isn't insurance)

The multiple is a range because one broker's own deck disputes itself. Howden Re's table (p.11) reads “~0.10–0.125%” pre-conflict while its prose (p.12) reads “from 0.25% of the insured vessel value pre-war” — and the table's own next row implies 0.25% too (“~$250,000/transit ($100M vessel)”). The two pages credit different sources. Marsh independently puts pre-war at 0.25%, so the honest floor-to-ceiling is 0.10–0.25% and the rise off it is 30×–100×. Al Jazeera's “1–3%” is not a pre-war baseline and is not used as one: it is Marsh's level “weeks ago” as quoted by Platts on 22 July — i.e. during the June–July truce — and it sits on the chart at that date (ledger). The 19 March and 26 March prints are not a decline — they are different books and different vessel classes a week apart; do not read a falling line between them. And the $21m is not insurance. The same source reports "about $21m" as the insurance cost for a 270,000-tonne tanker, but 270,000 × $77.96/t = $21.05m — that figure is the all-in freight bill at the quoted rate, not a charge on top of it. Insurance is a component of the $77.96 and the source does not break it out, so this card quotes the freight premium over the five-year norm and never adds the two. Conversion at 7.33 bbl/tonne (Arab Light runs ~7.6, so the $/bbl figures are marginally conservative).

Gas-fired metal: the one industry hit from both sides at once

So what — destroyed, not delayed — a frozen potline is destroyed capital, which is why this line recovers in years while crude recovers in weeks.

How to readIAI monthly production in tonnes/day, against the 17,800 t/day 2025 pre-war average.

Table view (monthly output, revisions, and the events)

Month	t/day	First print	Monthly kt	vs pre-war	Note

Methodology & caveats (revisions, and a baseline that reconciles)

International Aluminium Institute monthly statistics, via two IAI releases fetched 2026-07-25. March has two real prints — IAI first reported 15,963 t/day and later revised to 15,000: one source revising itself, not two sources disagreeing. The baseline reconciles: 17,800 t/day × 365 = 6.50 Mt/yr, which is IAI's own "around 6.5 million tonnes in 2025, or 9% of global capacity". The double exposure is the point: these smelters are gas-fired, so the LNG shock is their power cost, while the strait is their route — and their inbound alumina and petroleum coke transit it too. Gulf producers shipped 860,500 tonnes of primary aluminium to the United States in 2025.

Rank the strait's cargo by irreplaceability, not tonnage

So what — oil sits at the bottom of this board — sorted by captivity × time-to-restore, the cargo this whole site is built around sits at the bottom of the board.

How to readRows rank by captivity × months to restore; the time axis is logarithmic, and every clock except Ras Laffan's is an author scenario.

Table view (each clock and the basis for it)

Cargo	Captive	Months to restore	Fate	Basis for the clock

Methodology & caveats (the clocks are judgement; helium's two clocks; why this isn't a scatter)

The captivity figures are derived, the clocks are argued. Captivity comes straight from the bypass ledger. Every restoration clock, by contrast, is an author scenario with a stated basis per row, except Ras Laffan's 3–5 years, which is QatarEnergy's own published repair estimate. Treat the whole horizontal axis as scenario. Sulphur carries "captivity unmeasured" rather than a number, because its ledger row is an unmeasured band — printing its derived 0% here would assert a full bypass that nobody has established. Captivity is not the sort key it appears to be: eight of the thirteen cargoes sit at exactly 100% captive, so that column separates almost nothing on its own. Rows are ordered by captivity × duration, which means the ranking you read is very largely the clock. Helium carries two marks because it is two events. Qatar is ~33% of world helium structurally, but the 3–5 year repair clock belongs only to the ~17% of Ras Laffan capacity that was damaged; the rest sat on the strait's clock, since the complex halted on 2 March for want of an export route, sixteen days before it was hit. Multiplying the structural share by the repair clock would overstate the event roughly twofold. And for helium delay is destruction: it must be transported within 45 days of liquefaction or it boils off, so even a short closure destroys inventory rather than postponing it.

Gulf drinking water is a manufactured good — and the factory is gas-fired

So what — the LNG premium is the water bill — roughly three-quarters of GCC desalination is cogeneration bolted to a power plant, so the LNG shock is not adjacent to the water supply; it is its input cost.

Table view (both denominators, capacity, and whether the country was struck)

Country	Drinking water	Total demand	Capacity	Renewable fresh water	Struck	Note

Methodology & caveats (the two denominators invert — do not quote one as the other)

Both columns are from CSIS, 19 March 2026, and they measure different things. Drinking-water share is the share of potable supply from desalination. Total-demand share includes agriculture and industry, which in the Gulf run largely on fossil groundwater — so the two columns invert: Saudi Arabia is 70% desalinated for drinking but only 18.1% of total demand, while the UAE is 42% for drinking and 52.1% of total demand. Saudi Arabia's low total-demand share is not evidence of water security; it reflects irrigation drawing on aquifers that do not recharge. Bahrain's "over 90%" is charted at 90 as a conservative floor. Oman's coast is largely outside the strait on the Gulf of Oman, so its exposure runs through shared fuel supply rather than the chokepoint — the same geography the bypass ledger applies.

Thirteen strikes on things civilians drink, breathe and log into

So what — fragility, mostly not yet harm — the buffers were threatened and never actually drawn down, which is the honest finding and a different one from the headlines.

Table view (the full dated ledger, with attribution and confidence)

Date	Lifeline	Country	Asset	Fate	Conf.	Outcome for users	Attribution	What happened

Methodology & caveats (contested attribution; what is not a strike)

This is not the outage timeline. The Supply tab's timeline tracks assets that produce the barrel; every row here is an asset that consumes energy to keep civilians alive, and no row appears on both. Confidence is a column because attribution is contested. High means the operator or host government confirmed damage to its own asset — which is why the AWS and Kuwaiti rows are high and most others are not. The Qeshm Island strike is the extreme case: the damage is corroborated (30 villages lost supply, still out of service a month later) but every named party denies it — Iran accused the US, the US and Israel denied it, the UAE denied Israeli reports it had done it, and Trump denied knowledge. It is carried at medium with attribution recorded as disputed. Two rows are not strikes at all. Iran's 87-day internet shutdown was a government act — NetBlocks confirmed it was ordered, not broken — and the Barakah and Jebel Ali rows are near-misses carried because the concentration of what was nearly hit (43 desalination units on one site; a 5.6 GW reactor complex) is itself the finding. Iran's own Assaluyeh water claim is carried at low confidence: Israel confirmed striking the petrochemical complex, not the water units.

The megawatts that aren't there — and what they'd have run

So what — already realized — this is not a forecast of hardship; it is four-to-six-hour daily blackouts in Iranian industrial zones and full-day stoppages at Iran Khodro and Saipa.

How to readBands span low-to-high in GW; Iran's is one source's three cases — the dot is its realistic case, not independent estimates converging.

Table view (deficits and comparators, kept separate)

Row	MW	Kind	Basis	Note

Methodology & caveats (three scenarios, one source; what Iraq's row is not)

Iran's three figures are one source, not three. The 10 GW optimistic, 13 GW realistic and 18 GW worst cases all come from Iran's own parliamentary research centre, so they are charted as a single range with the realistic case as the midpoint. Showing them as three rows would read as independent estimates converging, which they are not — and the source is the deficit-holder's own government, which has incentives in both directions. Iraq's row is not Iraq's total deficit, which is much larger and long predates this war. It is only the generation lost to the cut in Iranian gas: deliveries fell to 5–7 mcm/d after strikes affecting South Pars and were still 15–20 mcm/d from early May, against roughly 50 mcm/d needed to run Iraq's gas plants properly. The GCC interconnector that was supposed to cover the gap slipped to at least end-August 2026. The three comparator rows are capacities, not deficits, drawn at reduced opacity and marked so they are never summed with the deficit rows — they exist because "13,000 MW" means nothing until it is 2.3 Barakahs.

One region's war took global air travel negative

So what — visible in the world total — IATA published the ex-Middle East figure precisely because one region had become large enough to invert the global sign.

How to readBars diverge from zero as year-on-year % change, each over its own period; colour marks a metric family — compare within one, never across.

Table view (every reading, its metric family and its source)

Reading	Change	Metric	Period	Source	Note

Methodology & caveats (three metrics that aren't comparable; IATA's own two numbers; what the closure cost)

Overflight counts are the wrong measurement and are not charted. They count aircraft passing overhead and barely move with whether anyone is travelling: a throughput metric standing in for a demand metric. What is charted is demand, and it turns out to be far larger than the routing story. Three metric families are on one axis and they are not comparable. RPK is IATA demand by region of carrier registration, so a Gulf carrier flying Bangkok–London counts as Middle East traffic; passengers is throughput at one hub, including transfer passengers who never enter the country; flights is third-party schedule analysis, not carrier-reported traffic — a cut schedule is a decision, an RPK fall is an outcome. They are coloured separately. IATA published two different numbers. The 28 May release carries −48.1% for Middle East carrier RPK in its data table and quotes Willie Walsh saying 46.6% in the same document; the likeliest reading is total-market versus international-only, but IATA does not say, so the table figure is charted and the quote is kept beside it. This was demand destruction, not a trimmed schedule: Middle East capacity fell 38.4% while traffic fell 48.1%, so the load factor dropped 13.1 points to 70.1% — fewer aircraft and emptier ones. The airspace closure is the mechanism, not the measure. EASA's conflict-zone bulletin prohibited Iranian, Iraqi and Lebanese airspace through at least 31 August 2026 (the notice in force at the July review), against the 1,000–1,400 daily overflights Iran handled before the war; a rerouted widebody burns an extra 2–3 hours at $6,000–7,500 per flight-hour ($12,000–22,500 a flight, priced off the same barrel), and Iraq's ~800 aircraft/day at ~$450 each is $360,000/day of sovereign fees, or $15,000/hour. The 2025 twelve-day war supplies the only clean measurement of where the traffic goes: Saudi overflights exactly doubled, 700 to 1,400 a day — 2025 evidence, labelled as such. A full-year 2026 Middle East contraction of about 11.4% appears only in secondary write-ups of IATA's revised outlook, so it is not charted.

The clock this tab runs on: every dated print and deadline from here to the midterms

Scenario — author narrative Four deadlines land inside one month — the War Powers 60-day expiry ~6 Sep, the committed SPR leg running dry at ~243 Mb (September 25 on the announced pace), fiscal year end, and the withdrawal deadline ~6 Oct — all inside the sixty days before the midterms — which is why a second MOU is the most likely path to lower prices, and why Iran, not Washington, holds the stronger clock. The prints between the deadlines are what grade the rest of this tab.

How to readThe dates are statutory or scheduled facts, and each carries the expectation this site already holds for it; the reasoning between them is judgment, not model output. The full ledger, with bands and grades as prints land, is on the overview.

The legal deadline is a forcing function, not a constraint — Congress has never used the War Powers Resolution to end a campaign, and in July the Senate killed the binding joint resolution 47–49 while the House passed a concurrent one 214–208 that compels nothing. What the deadline actually does is manufacture a decision on escalate-or-deal at the moment the oil buffer runs out, right before an election.

Both corridors at once — where the barrels go, and how much

Interactive — measured presets, modelled combinations Hormuz and Bab el-Mandeb are not independent. Petroline moves barrels out of the Hormuz denominator and into the Bab el-Mandeb one, so the same Saudi barrel cannot be charged against both straits — and closing both costs more than closing each — set the two corridors and read the interaction term below. The four scenarios that actually happened carry Kpler's published splits; every other combination is output from the allocation model and is badged as such.

Where — real geography; routes on the basemap, width proportional to volume

How much — widths are mb/d throughout; the same volumes as the map above

Table view (routes and volumes in the selected scenario)

Route	Fate	mb/d	% of capacity	Basis

Supply: where the barrels go in each scenario (mb/d)

So what — what the buffers can't reach — the release ceiling is fixed at 2.1 mb/d whatever the shut-in is, so every extra barrel lost lands entirely on the uncovered bar. That bar is the one price has to close, and it is what separates the two scenarios.

How to readEach bar's full length is that scenario's headline shut-in, mb/d — the three legend segments partition it; rows are rival scenarios and never sum.

Prices: three regimes, in order of buffer depletion (Brent, $/bbl)

So what — the buffer is the regime — the rows are not a forecast path but three pricing regimes, and what moves between them is not the size of the shut-in. It is whether there are still barrels in a reserve to meet it.

How to readEach bar spans the row's lo–hi band, dot at the central case, Brent $/bbl; the labelled verticals are published anchors, not scenario output.

How to tell which scenario you're in — each discriminator graded against a stated rule

Tracker — observed readings against author-set thresholds

How to readEach row names the public series that best measures it and the threshold rule that turns the reading into a verdict; the chip is the rule's answer, not a judgment call. Most rules grade the latest print; Brent grades the path — the share of recent sessions closing inside each scenario's bands — because one close inside a band says nothing about a run that has spent most of its sessions under it. Neither means the reading admits both signatures or fails both; no data means no public series exists. The verdict links to the chart the reading comes from.

The path so far, against the chapter it is in

Tracker — observed Brent against the author narrative

Scenario — author narrative Buffered $94–137 through year-end → rationing $147–203 as usable SPR exhausts Feb–Mar-27 (the full OECD buffer, ~13 months at the 2.1 mb/d release cap from mid-Jul-26, runs to ~Aug-27) → the fork in late-27: B at $182–272, or reopening into the deferred glut (~$65 base) — the prose is judgment; every number comes from a panel elsewhere on this site (price bands from the panel above, SPR dates, demand tranches, the supply trickle from the time-to-barrels waterfall).

How to readThe walk-forward of scenario A persisting — not a forecast, and not the base case (the market prices 37% odds of persistence into Q4 as of the Sep 1 strip — calibrated against Goldman's $115 case for ~2 mb/d of persistent losses, so an upper reading for this scenario's 8–11 mb/d shut-in; 19–64% across anchors).

The four-year lesson compressed into one strip: the first six months are paid for by buffers, the second six by decree, the third by recession — and the fork at the end is the only place a supply answer reappears.