The cap broke, and it broke on the physicals.

The re-arm that stayed capped through mid July has fired, and the sequence matters more than the level: LME warehouse stocks fell to 278,275t on 22 July from 298,775t on the 6th — about a third below the January level near 420,000t and the lowest since 2022, with Shanghai stockpiles also drawing — and the front curve then flipped into a roughly $20 cash premium over 3 months, the specific trigger named last month as the signal for the war premium re-entering the prompt. EGA's Al Taweelah restart and rising Chinese and Indonesian output are still running and did not prevent it; the drawdown outran the restart. The benchmark at about $3,192/t, up 24% on the year, is the least informative part. What makes this the sector worth reading against the rest of the chain is the discrimination: where crude's price detached from physical flow this month, here inventories fell, the curve backwardated and the price rose together, so a reader taking the energy finding as a general rule that commodity prices no longer measure supply would misread this market. Corroborate the headline against the physical series; believe it where they agree. Downstream, both airframers posted strong quarters naming their remaining constraints — Airbus its engine volumes, Boeing its wings and engine related components — and neither named metal, while the billet premium that would show a metal constraint reaching aerospace was last printed on 29 May.

The tests landed. The tariff was a handover; the write down was the bill.

Both dated events closed and the one that mattered was not the one on the calendar. Section 122's 10% surcharge expired by operation of law at 12:01 eastern on 24 July and the Section 301 forced labour action took effect at the same moment — no gap, autos explicitly in scope, but two tier at 10% and 12.5% across roughly 60 economies covering about 99.4% of US imports, broader in coverage and materially lower in rate than the 25–50% range much commentary anticipated, and close to a wash for many importers. USMCA changed nothing at the border, with preferential origin treatment continuing at 0%. The durable feature is that the replacement carries no statutory expiry and no rate cap. Then the prints delivered a larger number: GM beat on $48bn of revenue and raised full year guidance for the second time this year while booking a $2.3bn special charge, taking cumulative EV restructuring to roughly $10.9bn; Ford beat on adjusted EPS and raised guidance while posting a $1.32bn net loss on $4.2bn of charges, mostly a $3.6bn battery joint venture exit; Tesla set a revenue record on its weakest automotive margin in three years. One pattern across three companies inside an industry retreat reported past $35bn — the operating business beating while the EV programme is written off, which is the hybrid over BEV mix shift arriving as capital rather than as share data. Both OEMs are absorbing tariff cost corporately rather than passing it through. The magnet axis is unchanged and remains the sharpest constraint here.

The margin split — and it split on the date, not the balance sheet.

The pass through held for a second quarter and the dispersion beneath it turned out to be substantially calendar rather than structure. Delta reported on 10 July and reaffirmed full year EPS; United reported on the 15th and guided to the upper end; American reported on the 23rd with record revenue of $16.7bn, up 16.3%, and adjusted EPS of $0.15 against a $0.03 consensus — then cut full year guidance to a range spanning a $0.65 loss to a $0.65 profit and fell around 8%, with Southwest and Alaska, also reporting late, likewise cutting. Jet fuel spot rose 78 cents a gallon between Delta's guidance date and American's; American's third-quarter fuel expectation rose more than $700m in that window and nearly $1.6bn for the rest of 2026, roughly $550m of the latter in the final week, and its chief executive said the outlook could have been 4 times better had they reported a few weeks earlier. A guidance range is a snapshot of a forward curve on one morning rather than a statement about a company, and the same caution applies to the fuel assumptions themselves — $3.15, $3.75, $3.93 and $4.05 are four numbers struck on four different dates. None of which makes American's thin cushion unreal at 2.7% operating margin with no refinery and no hedge; the correction is attribution. Revenue confirms pricing power is industry wide: premium unit revenue up more than 13%, managed corporate up 26%. The corridor hardened durably rather than spiking — the Gulf advisory was extended on 22 July to 31 August, unchanged in content, applying at all altitudes.

The wrong chemistry paid out — three profits in one day.

The mismatch this sector has carried since the spring got its clean test on 30 July and the squeezed input never reached the growing segment. All three Korean makers returned to operating profit for the first time since the third quarter of 2024, on storage rather than vehicles: LG Energy Solution on revenue of KRW 7.56tn and operating profit of KRW 113.3bn, with first half ESS revenue up 4.6 times to a high 20% share and capital expenditure cut 53%; Samsung SDI on KRW 3.7688tn and operating profit of KRW 203.8bn against a KRW 290.2bn loss a year earlier, its first profit in seven quarters against a consensus expecting a loss; SK On on KRW 2.95tn and KRW 821.8bn, an implied margin far above its peers that is recorded as reported rather than reasoned from. All three name North American data centre storage demand. The pivot is physical rather than accounting, which is what makes it a supply-chain finding — Samsung SDI is converting part of its Kokomo, Indiana EV lines to storage cells, SK On is weighing the same at Georgia, and LG is scaling output across Michigan and Tennessee with third-quarter shipments guided up more than 50%. One qualification survives the vindication: LG's operating profit includes KRW 241bn of US production incentives, so the manufacturing line is a loss of roughly KRW 128bn ex credit. Profitable at the reported line, credit dependent underneath — which makes policy risk on those incentives the largest identified vulnerability, and it would separate Samsung SDI from LG rather than hit both.

The bottleneck held. The two things expected to worsen both came in better.

The transformer crisis throttling the AI and energy build outs is unchanged for a fifth consecutive cycle, which is what a war indifferent structural constraint should look like: lead times around 128 weeks for large power transformers and 3 to 5 years for high power custom units, switchgear effectively sold out through 2028, and interconnection queues still pushing a new entrant's energisation into the 2030s — against sub 18 month deployment cycles, on equipment that is under 10% of data centre cost and 100% of the critical path. Every dated relief milestone, including Panasonic's De Soto conversion and General Motors' Tennessee stationary storage pivot, lands in 2027 at the earliest and mostly 2029, which is after the capex wave they are meant to serve. What changed is both forward risks, and both moved the other way. The tariff handover on 24 July came in at 10% and 12.5% across roughly 60 economies — permanent where its predecessor was temporary, but close to a wash on rate, with Section 232 on copper, steel and aluminium unaffected and still the durable layer striking the transformer's own windings. And the homebuilder reset built on Lennar's 15.6% gross margin did not generalise: PulteGroup posted 25.0%, up 60 basis points sequentially with orders up 6% and guidance reaffirmed, while D.R. Horton closed 4% more homes at a 13.3% pre-tax margin. A nine point spread across three builders in one quarter. The reset is real at Lennar and it is not the industry.

The pump rose and sentiment rose with it.

The forecast here was that a re-rising pump would reverse June's sentiment inflection, and both legs of it are wrong. The AAA national average went from $3.84 on 9 July to $3.94 on the 16th, jumped 15 cents to $4.09 on the 23rd and held there on the 30th, with most states above $4 and the national figure nearly a dollar above a year ago. Sentiment rose anyway: the Michigan index reached a final 55.2 for July from 49.5 in June, almost 12% higher and a second consecutive double digit gain, its highest since February, with current conditions up 14.9% and improvements broad based across income, education, wealth, age and party. The reconciliation is level against peak rather than weekly direction — the pump peaked at $4.56 on 21 May, so at $4.09 the household is paying roughly 47 cents less than at the worst point, and the survey's language about easing pressure is accurate against that reference even while the weekly print rises. The more damaging correction is that the survey states consumers remain focused on pocketbook issues while political and military developments stay in the background, which negates the conflict judged temporary premise the forecast rested on; interviews ran 23 June to 27 July, covering both the escalation and the 15 cent jump, so there is no timing defence. Year ahead inflation expectations eased to 4.2% from 4.6%. Long run expectations held at 3.3% — the Fed relevant variable did not move.

There was never a barrel and a still. There is one balance, and the price stopped measuring it.

The migration of stress from crude to products is withdrawn, because they were never separable: commercial crude drew 7.2M barrels to a cycle low of 404.5M for the week ending 24 July, roughly 6–7% under the 5 year average, while refineries ran at 97.2% of operable capacity. Maximum effort converts tank inventory into product faster than imports at 5.7 mb/d, 6.9% below the year ago 4 week average, can replace it — so there is no loose crude leg to hedge a tight product leg. The larger instruction for anything reading this as an input cost is to stop citing benchmark levels: Brent touched $101.01 intraday on 23 July, shed some $16 in three sessions on talks reporting, settled at $90.74 on the 29th, and fell on the two sessions immediately after Aramco shut Jazan and Abqaiq was struck. Use cracks, inventories, loadings and outage schedules instead. Beneath the instruments, three physical features carry. The absorber has a calendar — 5 months of shock ran through inventory drawn at 3.8 mb/d, and both state legs fail together, with the US reserve at 61% of drawdown design and a June solicitation that offered 40M barrels and awarded 500,000. The deferral has a due date: January–May maintenance outages averaged 470,000 bpd against 900,000 in 2024, and that work lands in an autumn window overlapping hurricane season in a fleet with no slack. And the gate count went from one to three, with the Caspian Pipeline Consortium halting Novorossiysk loadings on 30 July and Kazakh output falling to 1.63 mb/d.

The front end easing was a fortnight, not a regime — and war risk stopped being a number.

Two of the chain's three front doors now report that their headline signal has stopped measuring what it is used to assert, and this is the second. The FOMC held at 3.50–3.75% on 29 July for a fifth consecutive meeting by 9 votes to 3, with Hammack, Kashkari and Logan dissenting for a quarter-point increase — regional presidents rather than governors, so the inflation objection is arriving from outside the Board — under a shorter statement, no submitted individual projections and a chair who has withdrawn forward guidance. The answer was a bear steepener on a hawkish hold: the 30 year up more than nine basis points to about 5.193% while the 2 year fell four to 4.236%. There is no longer a path to hand downstream, only a dispersion, and it is wide — three dissents for a hike, a committee median of one, and a market pricing two, with September hold odds at 41.9% against about 24% before the meeting. The direction of funding cost risk has inverted back to up. Meanwhile the insurance gate did something worse than harden: it fragmented. Hormuz war risk runs 7.5–10% of hull value against 1–3% weeks earlier, while on one sea a Jeddah or Yanbu call prices near 0.1% against up to 3% for Saudi linked tonnage — thirtyfold apart by counterparty rather than by route. Lloyd's insurers have begun excluding any Saudi touchpoint from Red Sea cargo cover, potentially including prior port calls. That converts a cost line into an availability line, and availability cannot be hedged by paying more.

One survey, two directions — and the floor got cut by proclamation.

The decomposition is now visible inside a single instrument. DTN's 29 July retail tape has nitrogen falling for a seventh consecutive week — urea at roughly $0.74 per pound of nitrogen against $0.90 in late May, anhydrous down 11% to $967 and back under $1,000 for the first time in 17 weeks, UAN32 down 15% month on month to $465 — while in the same survey over the same weeks the phosphates rose for a sixth, MAP up 1% to $958 and DAP to $913, with potash flat at $494. A reversible premium and an irreversible floor separating inside one dataset. Two corrections follow. Retail nitrogen was called as re-firming on the wholesale lag and fell for 2 more weeks instead; the wholesale side did firm, with Profercy's index up nearly 12 points and North African spot urea stepping more than $50 a tonne, so the mechanism is intact and the timing was wrong. More consequentially, the phosphate floor described here as structural has been administratively cut for one buyer: a presidential proclamation of 29 June suspended anti-dumping and countervailing duties on Moroccan phosphate for up to 8 months, which the agriculture secretary puts at a 22% reduction in US phosphate cost. It has not reached the tape — DAP and MAP were still rising a month later — and it arrives into the autumn prepay window that sets 2027 plantings. At the shelf the fork resolved on pricing architecture: Coca Cola raised guidance on 5% volume growth while PepsiCo reported tightening shopper budgets and weaker US sales.

Copper has two prices, and the acid shortage is paying for one of them.

The three source acid shock is unchanged — Gulf sulphur re-stranded, China's ban on more than 40% of world output still in force, Russian damage as an independent third leg — and no single de-escalation relaxes it. What is new is a sign change in the transmission and a split in the metal. Copper has separated into two benchmarks measuring different objects: COMEX inventories reached a record 639,147t and have weighed on the US contract, while LME stocks fell to 276,775t, the lowest since March, and Shanghai stocks dropped 12.9% in a week to 69,610t, the lowest since February 2024. One price is measuring a pending Section 232 determination on refined cathodes and the stockpile built in anticipation of it; the other is measuring metal that is not there. Anyone buying off one exchange and selling priced off the other now carries a margin exposure with nothing to do with copper — and the rule is to identify which series measures the physical rather than to average them. Underneath, trade reporting records copper smelters leaning on byproduct sulphuric acid revenue to offset negative concentrate treatment charges, so the shortage compressing every seaborne acid buyer is simultaneously subsidising the smelters the concentrate collapse is bankrupting. The captive acid divide is a revenue asymmetry, not only a cost shield. One correction: the Indonesian quota rise recorded a fortnight ago was reported and never confirmed, and half year smelter consumption at 46.2% of quota suggests demand rather than the quota may be binding.

The second test passed, and the war finally named itself — in freight.

The 100% Section 232 duty on patented pharmaceuticals, their active ingredients and key starting materials took effect at 12:01 eastern on 31 July for the seventeen Annex III companies, with 29 September following for everyone else, on a ladder running from 0% for the thirteen Annex II companies through 10% for the UK and 15% for the EU, Japan, Korea and Switzerland. None of it turns on the Strait, which is what makes this sector the exception. The shielded tier call has now been tested twice and passed twice: Johnson & Johnson beat and guided to recoup tariff cost, and AstraZeneca posted core EPS of $2.63 against a $2.48 consensus with core operating margin up two points to 34%, held full year guidance and reaffirmed its 2030 ambition. The new datum is the channel. AstraZeneca's chief financial officer stated that the Iran war had pushed up logistics and distribution costs — the first named war cost in a pharma income statement, arriving through freight rather than through the API acid line watched for three builds. Cold chain pharmaceutical freight is high value, time critical and air and reefer dependent, so it is acutely exposed to exactly the war-risk and routing costs now fragmenting by counterparty. Beneath the policy layer, cost not disruption held for a third quarter: active US shortages rose to 227, with sole source products at 48% of new 2026 shortages and discontinuations at 170, the highest since 2019. That is a margin signature, and it describes precisely the generic tier that both frameworks leave unprotected — on a carve out that expires around April 2027.

The anchor holds; the flex never closed. The operator said the opposite.

The structural fact is untouched and it is the reason this sector is tracked at all: there is no chemical substitute for sulphur cross linking at street vehicle scale, and tyres sit at the bottom of the margin hierarchy, so they are outbid for constrained acid in every frame — mining uses captive supply, batteries pivot chemistry, agriculture shifts crop mix, tyres cannot. What is withdrawn is the escalating three pressure cost narrative built on top of it. Michelin reported on 27 July with half year segment operating income of €1.45bn at an 11.4% margin, up 7% at constant scope and currency, and free cash flow before acquisitions of €282m against minus €102m a year earlier — attributing the margin support to improved mix, replacement brand growth and lower raw material costs. Full year guidance confirmed. The tape corroborates the operator: rubber futures settled near 217 US cents a kilogramme in late July in a narrow range, with Thailand in peak tapping season and southern output recovering, and the trade commentary attributing part of the softness to cheaper oil reducing natural rubber's cost advantage over synthetic. That is the elastomer flex operating, not closing. Level and direction diverge — TSR20 is up roughly 28% on 12 months while Thai first half exports fell 13% — so supply is genuinely constrained year on year while the current direction is sideways to softer. The binding commercial variable this quarter is demand, not input cost, with Chinese vehicle sales down for a ninth straight month and EU anti-dumping duties having pulled dealer inventory forward before their July start.

The inputs held a fourth time. The trade got contractualised, and the ceiling is being bought down.

The three track input map held on its conflict exposed and China controlled legs for a fourth consecutive test. Track A stays constrained by the war's persistence rather than its resolution: Ras Laffan's damaged trains keep 33–40% of global helium behind a repair estimated at up to 5 years and gated by turbine availability rather than by any transit outcome, with Korea roughly 64% Qatar sourced. Track B never depended on the Strait — the tungsten hexafluoride cliff fired on schedule, with Kanto Denka and Central Glass, together some 30% of world capacity, permanently ceasing in July and contracts locked 70–90% higher, while the two clock rare earth truce runs to 10 and 27 November over a never suspended military end user floor. Track C is where the description was wrong. It was called untouched; it is being actively expanded. TSMC raised 2026 capital expenditure from $52–56bn to $60–64bn against a $58bn consensus and added $100bn of Arizona commitment, with CoWoS running from about 75,000 wafers a month at end 2025 toward 125,000–130,000 by end 2026 and the supply demand gap projected to halve. Track C is immune to the war, not immune to change — the variable that moves it is capital. And the pricing model itself is converting: SK hynix printed a 76% operating margin, the highest in the industry, missed a ₩84tn consensus on shipments pushed into the second half, and has signed multi year agreements with around ten customers. A spot priced shortage is being sold forward.

The gate stopped being a place and became a counterparty test.

The Red Sea is reopening to boxes and closing to Saudi barrels at the same time, over the same water, because the blockade selects by affiliation rather than geography. Maersk and Hapag-Lloyd moved the AE15 Gemini service back to trans-Suez on 6 July and Maersk returned MECL on the 9th with an eastbound Jeddah call from August — while Lloyd's underwriters began excluding vessels with any Saudi touchpoint from Red Sea war-risk cover, potentially including foreign flagged hulls that merely called at Saudi ports previously, with Ascot and Navium reported preparing to withdraw policies. The premium spread is the proof: roughly 0.1% of hull value for a Jeddah or Yanbu call, about 0.5% for a Bab el-Mandeb transit, and up to 3% for Saudi linked tonnage or calls at Jizan and Al Shuqaiq. Thirtyfold on one sea, and not a risk any routing decision manages. There are now three gates selecting three different ways — Hormuz administered by permit at 20–34 crossings a day against 120–147 pre-war, Bab el-Mandeb by affiliation, and the CPC terminal at Novorossiysk by nothing at all, halted twice in 12 days. The freight tape isolates the Hormuz toll precisely: TD3C rose 4 consecutive weeks to $382,397 a day while TD34 outside the Strait fell to $120,750, widening the ratio from 2.1 to 3.2 in 3 weeks. That is a toll, not a tanker bull market. One consequence worth flagging: the Greek owned Merbabu and three Bahri-operated tankers appear to have switched off tracking, so sovereign fleets are now going dark defensively — degrading the loading and transit series everyone has just been told to substitute for price.

The cliff resolved into a cotton purchase mechanism, and the relief does not arrive until September.

The handover executed with no gap and the inversion this sector faced — apparel sits outside Section 232, so a bare Section 122 lapse would have cut apparel cost — was closed within the same minute on 24 July. What replaced it has a shape nobody anticipated. The rate assignment is an enforcement judgement rather than a trade metric: 54 economies were found to have failed both to impose and to enforce a forced labour import prohibition and took 12.5%, while six — Canada, Ecuador, the EU, Indonesia, Mexico and Pakistan — failed only on enforcement and took 10%, placing Bangladesh 2.5 points below China, Vietnam and Thailand. The textile mechanism is not a reduced rate but a zero one, delivered through tariff rate quotas for Bangladesh, Cambodia, Indonesia and Malaysia keyed to each economy's purchases of US cotton and running an initial three years — and the Trade Representative has advised those quotas are not feasible now but will be by 1 September. So the four intended beneficiaries pay the full duty for roughly five weeks on volumes not yet defined, which is not a plannable input for a forward order book. Meanwhile USMCA and CAFTA-DR apparel are fully exempt, converting the nearshoring lane from a cost advantage into a categorical one. The domestic textile trade body has come out against the mechanism alongside the importers — the first time both sides have aligned. Bangladesh's paper win sits against March RMG exports down 19.35% and LDC graduation on 24 November.

The bet got scored, and it was never the bet.

The test set last month returned the opposite answer on both the outcome and the premise. Hilton reported Middle East and Africa RevPAR down about 29.5% against a guided decline near 50% — roughly 20 percentage points of expected damage that did not materialise — with system-wide RevPAR up 3.9% currency neutral, occupancy up a point to 74.9%, and full year guidance raised to 3.0–3.5% from 2–3%. Hyatt posted system wide RevPAR up 5.9%, international excluding the Middle East up 7.5%, and raised its own full year range while quantifying combined Middle East and Mexico headwinds at $25m of fees and maintaining EBITDA guidance. The premise was the larger error: the raised guidance does not assume Gulf recovery at all. Hilton now guides Middle East and Africa down high single to low double digits for the full year, with the regional drag on system wide growth at roughly 0.5 percentage points, and rests the raise on US strength, business transient demand and a World Cup that delivered group RevPAR growth above 13% in host cities. The regional damage is real, formally expected to last the year, and no longer moves the global chains' numbers. What survives and matters more is the redistribution, now confirmed at operator level: Hilton's pipeline reached a record 541,300 rooms with over 70% of signings international, Hyatt's a record 154,000. Capital is being committed to the absorbing markets, which is far harder to reverse than a booking pattern.

The concentration structure is unchanged and it still has no single resolution point. Sulphuric acid runs on three sovereign legs — a Chinese ban covering more than 40% of world output, Gulf re-stranding, and Russian damage — so no de-escalation relaxes it and it takes all three. Helium is gated by turbine availability on a repair estimated at up to 5 years, which no transit or political outcome touches. Shipping stays outside the percentage frame because it gates every other row's recovery timeline, and this month it changed kind rather than degree: underwriters now price the hull's trading history rather than the voyage, at 0.1% for a Jeddah call against 3% for Saudi linked tonnage on the same water, and cover withdrawal is not a risk that routing manages.

The vector that moved most is the instrument itself. Three rows now carry a number and a direction that disagree. Copper has two prices measuring two different things — a record COMEX stockpile pricing a pending tariff determination against London and Shanghai inventories pricing metal that is not there. Natural rubber is elevated in level and soft in direction, with the operator citing lower raw material costs against a table that had called the substitution flex closed. And aluminium is the counter case where inventories fell, the curve backwardated and the price followed, which is what corroboration looks like. Identify which series measures the physical. Never average them.