June's CPI lands at 8:30 this morning with the softest headline inflation in two years, and the ceasefire that produced that softness has already been replaced. At 4:00 PM the Hormuz blockade reinstates with a 20% toll. Three of the strongest pieces in today's brief run the same machinery underneath. Airlines are legally obliged to buy a carbon unit that only a sovereign can mint, and a sovereign must subtract it from its own climate pledge to mint one. Senior housing has the demographic wave it was promised and no legal way to hire the people who staff the beds. Hormuz transit now carries a price set by a government rather than by the cost of the passage. BECAUSE the binding input in each case is issued by permission rather than produced by a market, capital cannot clear the shortage: the bid rises, the quantity does not move, and the price overshoots and stays there. Watch the 10Y after the print. If it ignores the soft number and holds above 4.55%, the bond market has already traded through the reference month, and this morning's disinflation is a measurement rather than a trend.
Iran's Revolutionary Guard struck two oil supertankers inside the Strait of Hormuz and fired missiles and drones at US military assets in Kuwait and Bahrain, while US Central Command ran a third consecutive night of strikes on Iranian coastal surveillance and missile sites. The toll and blockade you will read about in Geopolitics below now land into an active exchange of fire rather than a standoff.
Hormuz traffic is emptying ahead of the 4:00 PM deadline. Ship tracker MarineTraffic counted 57 transits from Friday through Sunday, down more than half from the prior week, against roughly 130 vessels a day before the war. The Department of Energy says 8.5 million barrels moved through on Monday under US military escort, which it called consistent with the recent average. The oil move itself is priced in Markets and Macro below.
Asia: Tokyo, Hong Kong and Seoul all closed higher. Europe: futures firmer. US futures are mixed into the 8:30 print, with the VIX at 17.2 after a two point jump, second quarter results from JPMorgan, Bank of America, Wells Fargo, Citigroup and Goldman Sachs before the bell, and Fed chair Kevin Warsh testifying at 11:00 AM ET.
Crypto data provided by CoinGecko
The softest inflation print in two years lands at 8:30 this morning. Its cause expires at 4:00 this afternoon. The Bureau of Labor Statistics publishes June CPI at 8:30 AM ET. Consensus is headline -0.1% month over month, pulling the annual rate from 4.2% to roughly 3.9%, with core holding near 2.9% year over year. The softness is real and it has one driver: US gasoline prices fell roughly 10% in June, and they fell because the Hormuz ceasefire reopened the shipping lanes and pulled Brent from $82 to $66. The ceasefire collapsed on July 8. Brent has since rallied back to $83.30, a 9.6% single day gain on Monday alone, and trades near $85 this morning. At 4:00 PM today the US Navy reinstates the blockade with a 20% toll on all cargo. So this is not a print that merely describes a world that has moved on. It describes the single event that has already been undone, and it will be read as permission to ease on the morning the war premium is reinstated by order. Which means the number is not the read. The reaction is. If yields fall on the soft print, the market is trading a June that no longer exists. If the 10Y ignores it and holds above 4.55%, the bond market has already filed the reference month as history and is trading this afternoon instead. The disinflation the print will show has been repaid in full and on the tape: $82 to $66 and back above $85, a full round trip and then some, which the data has not been told about. A price series does not lag. It reports. The lag lives entirely in the people who trade it as news.
Gold fell 2% to $4,020 on Monday while Brent rallied 9.6%, its biggest single day gain in more than six years, and in a geopolitical crisis those two assets are supposed to move together. The divergence is not noise. Gold falling while oil climbs is the market pricing Hormuz as an inflation shock, not a safe-haven event: higher rate expectations are overwhelming the geopolitical bid. The comparable is March 2022, when gold spiked on the Ukraine invasion and faded inside six weeks as the hiking cycle repriced real yields higher. Take the market at its word: it has decided that a naval blockade of the waterway carrying a fifth of the world's oil is a cost-of-living problem, not a financial-system problem, and it is funding that view by selling the one asset that pays out if it is wrong. The market is not hedging this. It is budgeting for it. Those are different bets, and only one of them survives an escalation.
The 10-year Treasury yield pushed to 4.59% on Monday, its highest in nearly two months, rising into an energy shock that should, by textbook logic, slow growth. Rates climbing alongside oil says term premium, not growth expectations, is the driver. The market spent six weeks pricing a Warsh pivot toward easing and is now unwinding that trade, with bond traders lifting bets on a hike at the July 28 to 29 meeting. This is the rate regime reasserting itself. The yield curve un-inverted in June, and a steepening curve with rising long rates and an energy shock is the late-cycle pattern that has historically preceded either a policy error or a growth surprise. It does not precede a clean soft landing.
Starbucks is cutting $10 million of software, and the market repriced four mega-caps on it. Per Bloomberg, the company is building in-house replacements for one Microsoft inventory system and one IBM maintenance tool, with Oracle's POS a third target. That is a 2.5% line-item cut against a $400 million software budget. IBM fell about 4%, Salesforce 4%, ServiceNow 3.5%, erasing tens of billions of enterprise-software market value on a $10 million cut at one company: a reaction three orders of magnitude larger than the cash flow in dispute. What changed is not the disintermediation. It is the reaction function. Investors have stopped pricing SaaS cash flows and started pricing the option that AI flips build-versus-buy, which means any anecdote that rhymes now moves the sector regardless of magnitude. Amazon and Whole Foods, June 2017: grocers lost roughly $22 billion of market value in a single day, and the disintermediation never arrived. The repricing is real. That is not the same as the thesis being right.
The on-chain credit business is running a negative spread, and its growth metric is hiding it. Grove's Q1 2026 financials, published by Token Terminal, show average capital deployed of $2.0 billion, up 19.3% on the quarter, and a book ending at $2.6 billion. Those are the numbers the real-world-asset narrative is built on. The income statement underneath: revenue $19.4 million, expenses $26.9 million, gross profit negative $7.4 million. The implied run-rate yield is roughly 3.8% against an implied funding cost near 5.3% (both figures derived by Token Terminal from reported data, not disclosed by Grove), which is a 150-basis-point negative carry on a $2 billion book. Size that and the picture sharpens: 150 basis points on $2.0 billion is about $30 million a year of bleed, which is precisely what the reported quarterly loss annualizes to, against a revenue line running near $78 million. The book does not have a pricing problem. It has no price. Expenses grew 156.5% while revenue grew 25.8%, which is the arithmetic of buying assets rather than underwriting them. That is the tell worth carrying: when a lending book's growth rate and its spread point in opposite directions, the growth rate is a marketing expense. Structured investment vehicles ran this exact trade into 2007. Borrow short, lend long, report the book. When the carry inverted, Citigroup brought $49 billion of SIV assets back onto its own balance sheet that December, and the vehicles that had looked like a business turned out to be a funding position. Capital deployed is a number the operator chooses. The spread is the one the market chooses.
The fastest-growing stablecoin holds its reserves at BlackRock, Goldman, and JPMorgan. Read that, not the growth rate. USDGO, issued by federally chartered Anchorage Digital Bank, grew supply 96% in a month toward $1 billion, now fourth among Solana stablecoins at 6% of the chain's $14.9 billion total. The reserves tell the story: BlackRock's BUIDL, Goldman's STBXX, JPMorgan's JLTXX. A dollar-token backed by TradFi money funds is a float business: the holder earns nothing and the issuer splits the short rate with the reserve manager. At 4%-plus, that is the most profitable float available, which is why supply doubles monthly. American Express ran the purest version for a century in travelers' checks: the customer prepaid, Amex invested the balance, and when rates fell the product died. The unpriced risk is not a depeg. It is a cut. When the Fed eases, issuers must share the yield to keep the float, and a spread business becomes a price war. The float is not the moat. The rate is.
Apple is suing OpenAI for trade secret theft, and the lawsuit names the scarcest input in AI: hardware-aware systems engineers. Filed July 10 in Northern California federal court, the complaint alleges Tang Tan, OpenAI's hardware chief and a 24-year Apple veteran, directed employees interviewing at OpenAI to share confidential hardware specifications. A second engineer, Chang Liu, allegedly kept a laptop with confidential documents after leaving for OpenAI. Apple is seeking an injunction barring OpenAI from possessing or using any Apple trade secrets. The structural story: the pool of people who can design AI-native consumer hardware is measured in hundreds, and the two largest AI product companies are litigating over it. The last comparable, Waymo v. Uber in 2017, settled for $245 million plus equity and permanently changed how Silicon Valley handled engineer departures. The constraint being priced is not intellectual property. It is the labor supply for a capability class that no school produces, and you cannot hire your way out of a shortage that no school produces.
OpenAI is buying its way into the customer relationship, because the model alone is no longer enough. The OpenAI Deployment Company, launched in May with $4 billion, agreed July 8 to acquire Northslope, an applied AI firm founded by former Palantir engineers. The deal adds hundreds of forward-deployed engineers who embed inside customer organizations to build AI systems. The strategic admission: model quality is converging, with Chinese open-weight models performing at 80 to 90% of frontier capability at a fraction of the cost, so the moat is migrating from the model to the last mile. That is the same migration that turned enterprise software from a product business into a services business in the 2000s, and the same one that made Accenture, Deloitte, and McKinsey the largest implementation partners in tech. OpenAI just told those firms it is coming for their margin. When the product commoditizes, the company that owns the installation owns the customer, and that is a worse business with a better moat.
Financial regulators designated AI infrastructure as systemically important this week, treating cloud providers the way they treat clearinghouses. The ECB gave every significant European bank until October 31 to prove it can absorb an AI-powered disruption, and the UK placed AWS, Google Cloud, Microsoft, and Oracle under a supervisory regime reserved for firms whose failure would break the financial system. The logic is concentration: 4 cloud providers host the compute layer for most global financial services, and a single outage cascades across banks, insurers, and payment processors simultaneously. Designation carries mandatory resilience testing and incident reporting, raising the cost of financial AI workloads and shifting economics toward on-premises alternatives. Watch the October 31 deadline: the first compliance reports will name which banks are single-sourced and which have built redundancy, a disclosure that has never existed. A utility is not a company that got big. It is a company the state has decided it cannot allow to fail, and the 4 hyperscalers just crossed that line without asking to.
Japan is building its first centralized intelligence agency since World War II, and Western allies are staffing it. The new agency launches this month with 700 personnel after legislation passed in May. Officials from the US, Germany, and Australia are advising on cyber defense, counter-espionage, and foreign-investment screening. The frame is not the agency but what it completes. Japan's 2% GDP defense target, counterstrike missiles, submarine fleet, and semiconductor export controls were hardware without software. A military that cannot see is a military that can be surprised. The line with a price is the last one. Japan is building capacity for the kind of inbound-investment and export-licence regime that, in the American case, repriced entire supply chains, and the first place it binds is semiconductor materials, where Japanese firms hold chokepoints in photoresists, wafers, and tools no buyer can route around. A state that can finally see is a state that can finally say no. That capability will be felt in licences long before it is felt in ships.
Trump reinstated the US blockade of Iranian ships in the Strait of Hormuz on Monday and demanded a 20% toll on all other cargo passing through the waterway. The blockade takes effect today at 4:00 PM ET. The toll would apply to a passage carrying roughly 20% of the world's oil and a third of its LNG, and its legality under UNCLOS is already contested. Watch two things. Whether allied navies enforce or publicly dissent, which decides whether this is a US action or a coalition posture. And whether shippers begin routing around the Cape of Good Hope, which adds 10 to 14 days and roughly $1 million per voyage, a cost that turns rational once the toll exceeds it. If rerouting volumes cross 15% of current Hormuz traffic within 30 days, the toll has stopped being a military posture and become a tariff. A price collected once is a precedent. Collected twice, it is a line item, and line items do not get removed when the shooting stops.
A naturally occurring bacterium from the intestines of Japanese tree frogs completely eliminated colorectal tumors in mice with a single intravenous dose. Ewingella americana, a facultative anaerobe that thrives in the oxygen-deprived cores of tumors, outperformed both immune checkpoint inhibitors and chemotherapy in the study, published in Gut Microbes by researchers at the Japan Advanced Institute of Science and Technology. Mice re-challenged with cancer cells formed no new tumors, suggesting durable immune memory. The mechanism is dual: the bacterium attacks tumor cells directly while recruiting T cells, B cells, and neutrophils to the site. (JAIST, Gut Microbes, July 2026)
Extreme deep-sea pressure squeezes dissolved organic nutrients out of sinking marine snow, and the leaked compounds feed microbial communities 30 times faster than the particles themselves would. Researchers at the University of Essex found that as organic clumps sink through 2 to 6 kilometers of water column, hydrostatic pressure forces out as much as 50% of their carbon and up to 63% of their nitrogen, creating a nutrient plume that deep-sea microbes metabolize within two days. The finding rewrites the standard model of deep-ocean carbon cycling, which assumed most nutrients reached the seafloor intact. (University of Essex, ScienceDaily, July 11 2026)
Earth's greatest mass extinction did not kill at random, and a new model finally explains its choosiness. Researchers publishing in the Proceedings of the National Academy of Sciences reconstructed the metabolic oxygen balance of marine animal groups and found that survival 252 million years ago tracked a single variable: whether an animal's oxygen demand, which warming raises, could still be met by an ocean whose oxygen supply warming lowers. The model reproduces the pattern in the rock. Brachiopods and crinoids, the slow filter feeders that carpeted the Paleozoic seafloor, were nearly annihilated, while roughly half of the mollusks, the mobile clams and snails carrying faster metabolisms, came through. The reason a modern beach is covered in clam and snail shells rather than brachiopod shells is a physiological threshold crossed a quarter of a billion years ago. (Stanford, PNAS, July 2026)
Astronomers using ESA's Euclid space telescope uncovered 31 of the oldest known quasars, including two from when the universe was only about 670 million years old, roughly 5% of its current age. Each core shone with the light of a trillion suns, powered by supermassive black holes that should not have had time to form under standard models. Thirty-one objects from a single survey suggest the problem is not rare outliers but a population the field has been systematically missing. (ESA/Euclid, published July 7 2026)
Every airline on earth has been ordered to buy a compliance instrument that only a foreign government can call into existence, and that government has to subtract it from its own climate pledge to do it. Demand for the first compliance period runs to 236 million units. Supply is about 40 million. The window closes on December 31.
CORSIA is the global scheme, adopted by ICAO and now covering 130 participating states, that forces airlines to offset the growth in their international emissions above a fixed baseline. Its First Phase runs 2024 to 2026, and per IATA's own June 2026 fact sheet the numbers are not close: the official 2024 offsetting requirement was 55.6 million tonnes, 2025 is estimated at 58.8 to 81.5 million tonnes, and total First Phase demand for eligible units lands between 170 and 236 million. Against that, the current supply of CORSIA Eligible Emissions Units is, in IATA's words, "nearing 40Mt." Here is the part almost nobody outside the carbon desks has looked at, and it is the whole story: this is not a project-development shortage that capital can fix. A unit only becomes CORSIA-eligible when the country where the project sits issues a Letter of Authorization and performs a "corresponding adjustment," which means formally subtracting that emission reduction from its own Paris pledge so it cannot be counted twice. The host country must give away a piece of its own national target to manufacture an asset an airline can buy. IATA names the bottleneck plainly: host countries' "lack of awareness about how to implement or their reluctance to establish the required authorization infrastructure." Supply is gated by sovereign consent, and sovereigns have a direct interest in withholding it. Worse, IATA notes the thin eligible pool is already "attracting demand from sectors other than air transport due to their high environmental integrity," so airlines are being outbid for the few units that exist. And the clock does not stop: from 2027, CORSIA widens from volunteer state-pairs to all international flights, dropping a second, larger book of demand on top of a First Phase that is already five-to-one short. If authorized supply is still running far below the obligation as the 2027 compliance cycle opens, expect the cost of flying internationally to acquire a new, permanent, politically-priced line item, and expect it to fall hardest on the international network carriers, Delta (DAL), United (UAL), American (AAL), IAG and Lufthansa, while the domestically weighted carriers, Southwest (LUV) and Alaska (ALK), largely escape it because CORSIA prices international flying only. The relief valve is the other compliance path: burning CORSIA-eligible fuel reduces the offsetting requirement directly, so a scarce, expensive unit makes a sustainable-aviation-fuel contract look cheap by comparison, which pushes airline compliance dollars toward the fuel producers, Neste (NTOIY), Darling Ingredients (DAR, through its Diamond Green Diesel venture with Valero, VLO), and Gevo (GEVO). Watch: the ICAO CORSIA Central Registry's published eligible-unit supply and the running count of host-country Letters of Authorization, through the First Phase close on December 31, 2026 and into the 2027 cancellation cycle. If authorized supply is still under roughly 100 million tonnes against a 170 to 236 million tonne obligation at the moment the 2027 scope expansion lands, the shortage is a design feature rather than a lag, and the clearing price of aviation compliance will be set by sovereign reluctance, not by the cost of abatement.
Undercoverage warrant: the numbers above are lifted from primary sources, IATA's June 2026 CORSIA fact sheet and ICAO's published Sectoral Growth Factor, not from research. A six-month consensus search for sell-side or wire coverage of this thesis returns only carbon-market trade press (Fastmarkets, AlliedOffsets, Carbonwise) and price forecasts from carbon-data shops; no major desk has published an airline-sector thesis on the Letter-of-Authorization supply gate, and 2026 wire coverage of airlines is fares, capacity and fuel. The specific mechanism no one has assembled: eligible supply is constrained not by a shortage of carbon projects but by a sovereign's unwillingness to weaken its own NDC, which is why money cannot solve it and why the gap survives into a 2027 scope expansion.
The most crowded demographic trade in American real estate assumes the beds get filled. The industry's own trade associations are in Congress right now asking for a visa category that does not exist, because the people who fill those beds cannot legally be hired. And the exposure to that does not run along sector lines. It runs along the lease.
The demand half of this story is consensus and correctly priced. More than two million Americans turn 80 in 2026; the 80-plus population grows by over four million by 2030, to 18.8 million. Meanwhile senior housing construction starts have collapsed to the lowest level since 2009, with roughly 2,500 units started in the fourth quarter of 2025, and occupancy is climbing hard: Welltower's operating-portfolio occupancy hit 87.3% in Q1 2026, up from 85.1% a year earlier, with the industry underwriting a path through 90% this year toward the low 90s by 2028. Every desk owns that story. What none of them price is that a senior housing unit is not an asset, it is a service, and the service requires a person to show up. Three in ten direct-care workers in the United States are immigrants, and more than a quarter are foreign born. The 2025 and 2026 revocations of Temporary Protected Status and humanitarian parole pulled legally authorized workers out of facilities where they had worked for years, and University of Pennsylvania researchers estimate deportations could shrink the direct-care workforce by roughly 395,000 by 2028, including about 275,000 immigrant-held positions, which is the same timetable as the occupancy forecast. Underneath that sits a slow structural drain: the caregiver support ratio, women aged 45 to 64 per adult over 80, falls from 3.4 to 1.7 between 2020 and 2040. The tell that this constraint is binding rather than rhetorical is in the legislative record. H.R. 5494, the Essential Workers for Economic Advancement Act, introduced September 18, 2025, would create a new H-2C visa specifically because long-term care has no visa program that permits year-round temporary hiring at all, and a second bill, the Careworker Visa Act, was introduced June 9, 2026. Neither has passed. When a sector's own trade groups tell Congress, in writing, that the labor input for their growth cannot legally be obtained, the growth forecast is a labor forecast wearing a real-estate costume. But the tradeable object is not the sector, and this is where the crowd is positioned wrong. It is the lease structure, because the lease decides who owns the payroll. Welltower (WELL) is driving its senior housing operating portfolio, the RIDEA structure in which the landlord takes the operating margin directly, toward roughly 72% of cash flow, and Ventas (VTR) has that segment above half of net operating income. In that structure, every dollar of wage inflation and every unstaffable unit lands on the REIT's own income statement. The triple-net owners, Omega Healthcare (OHI), LTC Properties (LTC), National Health Investors (NHI) and Sabra (SBRA), collect a contractual rent and feel nothing until an operator is impaired enough to stop paying it, which is a later and slower exposure. Brookdale (BKD), the pure operator, sits at the sharp end. And if facilities cannot hire permanently they rent the labor instead, at a premium, which would hand the contract-staffing agencies, AMN Healthcare (AMN) and Cross Country Healthcare (CCRN), the rate re-acceleration nobody is modeling after their brutal post-pandemic downcycle. Watch: same-store operating margin in the Welltower and Ventas seniors housing operating portfolios, quarterly through 2026 and 2027, set against their occupancy prints, and the BLS direct-care employment series against the NIC MAP occupancy series. If occupancy crosses 90% while that operating margin stays flat or falls, the ceiling is labor rather than buildings. Two landlords, one wave, one wage bill, and opposite outcomes decided by a contract signed years before anyone knew there would be a labor problem. The landlord who took the operating margin bought the payroll along with it. The landlord who took the rent cheque did not, and will not, until the operator stops writing it.
Undercoverage warrant: the topic (the 80-plus demand wave) is well covered and I am not claiming otherwise; the undercovered object is the mechanism. The primary sources are non-consensus and academic or legislative rather than sell-side: the University of Pennsylvania deportation-impact estimate, KFF and PHI workforce-composition data, and the text of H.R. 5494 and the Careworker Visa Act. A six-month consensus search returns the demand thesis in force (Bisnow's "Wall Street piling into senior housing as massive shortage looms," REIT initiations, trade press) and returns labor shortage mentioned as a generic operating headwind, but returns nothing that assembles the workforce forecast and the occupancy forecast into a single ceiling claim, and nothing that splits the exposure by RIDEA-versus-triple-net lease structure. That split is the tradeable object and it is absent from the coverage.
Restart Price. When a competitor is taking your industry with subsidised supply, everyone asks the same question: how long can they fund it? The question that actually decides the outcome is the one nobody computes: what will it cost to switch your side back on when they stop, and is anyone paying that bill? Financing is recallable. Capacity is not.
Michael Pettis stated the mechanism yesterday: a phone maker becomes a major carmaker in three years not because of scale but because of "near-unlimited financing at very accommodative terms... with little concern about eventually hitting hard budget constraints." That is Kornai's soft budget constraint, and the receipts are on the tape. China's macro leverage ratio hit 302.3% of GDP at the end of 2025, up 11.8 points in a single year, per NIFD. Meanwhile BYD's Q1 profit fell 55% and Xiaomi's EV division swung back into the red, even as BYD's overseas shipments grew more than 50% and it raised its 2026 export target from 1.3m to 1.5m vehicles. Volume expands while profit contracts. Under a hard budget constraint that is impossible; under a soft one it is the design. The input being consumed is credit, not margin.
Which is why the bear case on China is right, and why its consolation prize is imaginary. Japan ran this exact play from the 1970s, dominated every sector Tokyo blessed, and reversed after 1990-91, because debt capacity is always the binding constraint. So the clock runs out. But exit and re-entry are asymmetric. Dixit (1989): sunk costs open a band where a firm exits and does not return even after the price has fully recovered. Here the asymmetry is literal. Slovalco is spending €100m to restart 75,000 of the 175,000 tonnes it switched off in 2022: four years and roughly €1,300 a tonne merely to turn it back on. Speira stopped waiting and converted its Rheinwerk potline to recycling. ExxonMobil is permanently closing an 830,000-tonne ethylene cracker in Scotland; seven European crackers have shut or been slated since April 2024. None of that un-demolishes when Chinese supply reverses. The aggressor's collapse is not the defender's recovery. Those are two different events, and only one of them is on a timer.
The call: by June 30, 2027, European primary aluminium and ethylene capacity is LOWER than today, not higher. At most one idled European primary smelter returns to nameplate, and the closure list gets longer, even as the credit funding its executioner shrinks. And the tradeable line is not the sector, it is the STATE: capacity mothballed with a costed restart re-rates when China stops; capacity severed or converted never does. Ask any industrial you own for its restart price. Most cannot tell you, and that is the answer.
Where this might be wrong. Strongest objection first, and it is a good one: capacity does come back, but on a new cost structure, not a recovered price. The US Gulf Coast rebuilt the petrochemical industry everyone had written off, on shale gas. Which means this thesis is, underneath, a bet that European energy costs stay high; give Europe cheap power and the restart price gets paid and the crackers get rebuilt. Watch European industrial power prices, not Chinese subsidies. Second: irreversibility is a price, not a wall. Alcoa completed the San Ciprián restart on April 7, 2026. A violent Chinese reversal would fund restarts that a gradual one cannot, and the debt-capacity mechanism I am leaning on implies the gradual path, which is convenient for me and should be treated with suspicion. Third, the objection I least want to hear: Beijing may impose the hard budget constraint itself. Its campaign against below-cost "involution" pricing is precisely that, and if it works, the window of destruction is years shorter than I assume and the wait-them-out posture is vindicated on schedule. And the engine itself is falsifiable, not just the conclusion: if Chinese vehicle exports fall year-over-year for two consecutive quarters while credit to the sector is still expanding, then the binding constraint was demand all along, not credit, and I have misidentified the mechanism, in which case the timetable and the destruction estimate both go with it. Dated falsification: if by June 30, 2027 two or more idled European primary smelters have committed to full restarts AND a new cracker above 400,000 tonnes is sanctioned in Europe, the restart price is being paid, and the framework is wrong.
"...looking on the shelves for one book, [the reader] is attracted by the kindred ones next to it... and finds himself involved in a new trend of thought."
— Gertrud Bing, describing Aby Warburg's law of the good neighbour
Aby Warburg spent his life and his family's banking money building a library in Hamburg, and then he did something to it that librarians still find slightly insane. He refused to shelve the books by author, by title, or by any fixed classification. He shelved them by problem, and he re-shelved them constantly, because the problems moved. He even put the call number on the front cover rather than the spine, so that finding a specific book you already knew about would cost you a little extra effort, and looking would cost you nothing.
He called it the law of the good neighbour: the book you know you need is usually not the book you actually need. The one you need is the one standing next to it, the one you would never have thought to ask for, and would therefore never have found by searching.
This inverts what we instinctively try to improve. When thinking feels stuck, we look harder: better sources, sharper questions, a more precise search. Warburg assumes precision is exactly what fails, because a precise question can only return an answer inside the frame that produced it. What breaks the frame is proximity, the thing you did not go looking for, close enough that your eye falls on it while reaching for something else.
Which means the leverage is not in your effort. It is in your surroundings. Most people's environments are optimised for retrieval: everything filed, nothing adjacent, the search bar doing all the work. That is a room designed to give you back exactly what you already know you want, and nothing else, forever.
Today's practice: name the one piece of work that actually matters this week. Then take the book, the paper, the half-read document you keep meaning to get to and keep deferring, and physically put it where that work happens. On the desk. Open. Within reach. Not to read it. To be next to it. You are not adding a task. You are moving a shelf.
In 1990, two psychologists at Michigan, Daphna Oyserman and Hazel Markus, sat down with 238 adolescents between the ages of 13 and 16 who differed in one measurable way: how much trouble they were in. Some had no record at all. Some were in custody. The researchers did not ask about self-esteem, which had been the field's favourite variable for a decade and had never predicted much of anything. They asked each teenager to describe who they expected to become next year, and who they were afraid of becoming.
The hoped-for selves came back almost identical across every group. Nearly all of them wanted to get a job, finish school, be respected. Aspiration was not the variable. What separated the groups was the pairing. The youths who were not in trouble tended to report a feared self that sat directly opposite an expected self in the same territory: expecting to graduate, afraid of dropping out. Expecting to work, afraid of being unemployed. Oyserman and Markus called this balance. The most delinquent youths were the ones whose fears and expectations lived in different rooms, or whose feared selves had no countervailing expectation at all.
The finding runs against almost everything we are taught about motivation. The standard advice is to make the positive vision vivid and to banish the negative one, on the theory that fear is corrosive and confidence is fuel. Markus and Nurius, who introduced the concept in 1986, found something structurally different. A possible self is not a fantasy. It is a working cognitive object, a specific, detailed representation of a future person, and it only produces behaviour when it is paired with its opposite in the same domain. The hoped-for self supplies the direction. The feared self supplies the reason today matters. Alone, either one is inert. The hope becomes a daydream you can enjoy without moving. The fear becomes an anxiety you can only escape by not thinking about it.
The failure mode is worth naming precisely, because it is the one most people are running. An abstract positive self ("successful," "healthy," "a good parent") paired with nothing produces no behaviour at all, and it feels wonderful. It is the most pleasant way to make no progress that has ever been invented. The correction is not more positivity and it is not more fear. It is matching: for each future you want, construct the specific, same-domain version of the future you are avoiding, in equal detail, and hold them together.
The decision tool: when a goal you genuinely hold is producing no behaviour, stop working on the goal. The vision is almost never the missing piece. Ask instead what specific, concrete, same-domain failure you are trying to avoid, and check whether you have actually built it, in the same resolution you built the success. If you cannot describe the failure as vividly as you can describe the win, you do not have a plan that will move you. You have half a plan, and it is the enjoyable half.
In 1976 two control theorists, Bruce Francis and W. M. Wonham, published a result in Automatica that reads like philosophy and is in fact a theorem. A feedback controller can drive an error to zero and hold it there only if the controller physically contains a model of the dynamics of the disturbance it is fighting. Not a model of the system. A model of the disturbance. The proof of it is sitting inside every thermostat and every cruise control: the reason those controllers carry an integrator, the "I" in PID, is that an integrator is a model of a constant, persistent load, a draft under the door or a hill in the road. Strip the integrator out and turn the gain up as high as you like, and the car still sags below the set speed on the hill, forever, with the gap shrinking but never closing. Now change the disturbance from a constant into a rhythm, a periodic wobble, and the integrator is suddenly the wrong model; nothing cancels the wobble until you build an oscillator of that same frequency into the controller itself. Fifty years on, the principle is still being extended to nonlinear and time-varying systems, and it has never been softened. The controller must carry, inside its own machinery, a replica of the thing it means to nullify.
The famous cybernetic slogan, Conant and Ashby's "every good regulator must be a model of the system," is the blurry ancestor of this, and the blur hides the useful part. The requirement is not a model of the world. It is a model of the specific disturbance, which means a controller can be exquisitely equipped for one shape of shock and completely helpless against another, no matter how much capacity it is carrying. That inverts the way we instinctively buy robustness. We buy it as a quantity: more buffer, more margin, more speed, more people, faster escalation. The theorem says robustness to a class of shock is not a quantity at all, it is a representation, and gain is not a representation. It also explains a pattern we routinely mislabel as resilience: the organization that absorbed one crisis handles a repeat of that crisis beautifully and is flattened by the next unfamiliar one. It did not get tougher. It acquired exactly one model.
So when the same kind of shock keeps getting through, stop adding gain and start asking what shape it is. Concretely, this week: take the failure that keeps recurring, and before you add another buffer, another reviewer, another escalation path, write down its dynamics. Is it a constant drift? A one-time step? A cycle running on somebody else's clock? Then put something inside your process whose only job is to generate that shape and subtract it. If you cannot name the shape, no amount of effort will cancel it, and the effort will only make the system feel busy, which is precisely what an over-gained controller does in the moments before it starts to oscillate. The same architecture runs far outside engineering. Immune memory can neutralize only the pathogens whose signature it has stored, which is why a vaccine is a model of a disturbance rather than a stronger defense. The cerebellum cancels the sensory consequences of your own movements by running a forward model of them, which is why you cannot tickle yourself. And an institution rebuilt around one crisis carries that one model, and only that one, into the next.