Two Repricings, One Referee.

The week in eight

• Crude closed the week at 89.31 dollars, up 8.3 percent on the week and close to 29 percent over four weeks. The Iran ceasefire has collapsed and the Strait of Hormuz, through which about a fifth of the world’s oil passes, is barely open. Petrol follows crude with a lag of two to three weeks, so the July inflation number will carry this move inside it.
• The oil shock has flipped the rate outlook. Our internal rate model, which spent June leaning towards a cut, swung this week to leaning gently towards a hike, and the two-year Treasury yield rose with it. The Federal Reserve decides on Wednesday.
• Beneath a calm stock market, the AI trade came apart. The S&P sits only about 2 percent below its record, but the market's momentum factor, the long-running strategy of owning its biggest winners, had its worst month since 2008, down around a quarter in July, and today those winners are overwhelmingly the artificial-intelligence names. Taiwan Semiconductor reported revenue up 36 percent and a record margin, and the shares fell more than 5 percent. When a blowout result is sold, the good news was already in the price.
• The crash gauge reads 25 of 100, unchanged. This is the first edition of the eight-dial version, which now watches the energy shock the old seven-dial version was blind to. Two dials are red: the shape of the yield curve and the new energy signal. The full working is in Appendix A6, so you can reproduce the number.
• Last week’s tell fired. We said that if crude held above 80 dollars and the two-year Treasury above 4.10 percent at Friday’s close, the higher-for-longer read was confirmed and the September cut should fade. Both legs fired: crude at 89, the two-year at 4.37. The cut is fading.
• Case study: the Jio gamble that wired a nation. Mukesh Ambani spent tens of billions building a mobile network across India and then gave it away for free, resetting the price of an entire market to a level only his balance sheet could survive.
• Accountability: a call that did not pay. Talen Energy reaches its thesis horizon this weekend. The nuclear-demand thesis proved out, but the stock went nowhere while the index rose, so the call cost roughly six points against simply holding the S&P. We score it honestly in On the Radar.

Markets in brief

The S&P 500 closed Friday 24 July at 7,411.98, down about 0.6 percent on the week but still up 8.3 percent for the year, a calm surface over a churning market. The 10-year Treasury yield, which sets the tone for mortgages and business loans, rose 16 basis points (hundredths of a percentage point) to 4.71 percent as the oil shock repriced the outlook, while crude itself finished at 89.31 dollars, up 41 percent for the year and now second on our Scoreboard. Gold firmed to about 4,068 dollars. The VIX, Wall Street’s fear gauge, eased to 18.58, calm on the surface even as the market churned beneath it.

The gauge reads 25, unchanged, but for the first time that number includes the oil move it would once have missed: the new energy dial is one of only two lights now showing red.

The full working, every signal, its weight, its threshold, this week’s reading and the score it earned, is printed in Appendix A6, so you can recompute this number yourself.

Hold into Wednesday, tilting hawkish: the tape has already moved into the oil-pass-through scenario, and the meeting is the referee.

Reconciliation: the market spent June pricing a September cut. Our model reads Hold, tilting hawkish, and this week it moved with the oil, its composite rising as the two-year Treasury trend turned up. It will not lurch: its rule requires two of its three blocks to agree before the view changes, and this week the inflation and market blocks moved hawkish together while the labour block stayed soft. We treat the market’s cut pricing as sentiment to be tested, not an input to follow.

Last week’s tell, scored first. We left a tell with two legs, both exact: if crude was still above 80 dollars and the two-year Treasury still above 4.10 percent at Friday’s close, the market was pricing a disinflation the oil market had already revoked, and the September cut should fade. Both legs fired. Crude closed at 89 dollars and the two-year at 4.37 percent. The rear-view read is confirmed: the cut that looked live in mid-July is fading, and this week our internal rate model swung with it, its composite moving from cut-leaning to hold, tilting hawkish.

Two repricings, one referee

Start with the barrel, because it is the loud number once you look for it. Crude has risen close to 29 percent in four weeks, from about 69 dollars in late June to 89 now, as the Iran ceasefire collapsed and the Strait of Hormuz, through which about a fifth of the world’s oil passes, was throttled to a trickle. An oil move of that size does something a month of central-bank speeches cannot: it feeds into inflation with a lag of a few months, which forces a central bank that fears the pass-through to keep interest rates higher for longer. That is the whole chain, oil to bond: higher crude, higher for longer, a higher two-year Treasury yield and lower long-bond prices, which is why long-dated Treasuries are the worst of the major government-bond and equity benchmarks this year. The reader hoping for an autumn rate cut should see that the barrel, not the Fed, may already have taken it.

The second repricing is quieter and sits inside the stock market. The S&P is only about 2 percent below its record and barely moved on the week, so the surface looks placid. Underneath, the most crowded trade in the market came apart: the momentum factor, the strategy of owning the market's biggest winners, had its worst month since 2008, down around a quarter in July, and today those winners are overwhelmingly the artificial-intelligence names. The clean proof landed on Wednesday, when Taiwan Semiconductor reported revenue up 36 percent, its highest-ever margin and profit up 77 percent, and the shares fell more than 5 percent. When a result that good is sold, the market is not paying for good news any more; it is repricing what it already owns. Both stories meet on one desk on Wednesday, when the Federal Reserve decides.

The rate read, governed by the model. The model’s dial moved this week from 4 to 14, still inside the Hold band but now tilting toward a hike, and what moved it was the oil pass-through and a two-year Treasury trend that turned up. The discipline matters as much as the direction: the model changes its view only when two of its three blocks, inflation, labour and markets, agree, and this week the inflation and market blocks moved hawkish together while the labour block, weak payrolls and rising claims, stayed soft. So the view is Hold, tilting hawkish, not a call for a hike. The Taylor Rule benchmark it anchors on (a mechanical formula for where the policy rate should sit given inflation and unemployment) still prescribes a rate about 1.33 percentage points above the funds rate: a cut is not yet rule-justified, however the market prices it.

The competing explanation, named. We have told this week’s equity move as an AI-concentration unwind, and we should say what else it could be: the same oil shock, taxing every business at once. The tell that separates the two readings is what got sold. An oil-driven selloff sorts markets by energy exposure; a positioning unwind sorts them by crowding. This week the sort ran by crowding, the crowded AI winners fell while the broad market held, which is why we read it as positioning first. If next week the energy-importing indices start falling while oil holds, the oil read takes over, and we will say so.

Where I could be wrong (the standing self-check): First, the model we anchor on is fed by June-vintage inflation data, and the oil shock is a July event, so if the July prints re-accelerate the model is late by construction; the tell is a two-year yield breaking above 4.45 percent, which has not happened but is close. Second, I may be reading the equity rotation too calmly. If the AI unwind stops being a rotation and becomes a broad de-risking, the tell is high-yield credit spreads widening through 350 basis points from today’s 277; that has not started, and until it does the “sorting, not selling” read holds.

2026 thesis check-in

Six editions on, the Repricing thesis is easy to show with a number rather than assert. Across our five reference asset classes, US shares are up around 8 percent this year, high-yield credit up around 1, the aggregate bond index down around 5, gold down around 6, and crude oil up around 41. Best to worst, that is a spread of roughly 48 percentage points, with three of the five up and two down, which on our own tightened standard, a spread above 35 points with at least two moving each way, is a strong confirmation. The honest caveat, and it matters: most of that width is the oil shock. Strip crude out and the remaining four run from plus 8 to minus 6, a spread of about 14 points, which is only moderate. The gaps between asset classes are real and wide, exactly as the thesis said they would be; this week, one gap is doing most of the work.

The barrel wrote the week, again. Crude added 8.3 percent to close at 89.31 dollars, its second consecutive weekly jump, and is now up close to 29 percent in a month. Nothing else on the board moved like it, and nothing else mattered as much, because a barrel at 89 dollars is not just an energy story, it is a rate story and an inflation story wearing an energy costume. The Iran ceasefire has collapsed and the Strait of Hormuz is barely open; crude did give back about 3 percent on Friday as stalled peace talks took a little risk premium out.

Silver was the quiet second act. It rose 4.7 percent against gold’s 1.4, running roughly three times its larger cousin and snapping back the ratio that had gone the other way the week before. Copper added 1.6 percent. The industrial metals are telling a firmer-growth story than the precious ones, and gold, down 6 percent on the year, is having a poor 2026 by its own recent standards.

The largest currency move was the rand. The South African rand weakened 2.5 percent against the dollar on the week, and most of that on Friday alone, the biggest verified move on our board. It is worth a reader’s pause: a move that size in a week is usually either a country-specific wobble or a broad emerging-market tremor, and this one looks more like the latter, the dollar firming as the rate-cut hopes that had softened it drained away.

The equity indices barely moved, and that is the story. The S&P slipped 0.6 percent, the Nasdaq 100 fell 1.6, small caps eased 1.1. But the calm is an average of two violent halves: the crowded artificial-intelligence winners fell 20 to 25 percent from their peaks over the month while the rest of the market held, so the index netted out to almost nothing. Taiwan Semiconductor is the tell of the tape: it reported one of the best quarters a chipmaker has ever printed, revenue up 36 percent and a record gross margin, and its shares fell more than 5 percent. When results that good are met with selling, the good news was already in the price.

Bonds fell, crypto slid. Long-dated Treasuries, the TLT fund, lost about 1.5 percent as the oil shock pushed yields up, leaving them the worst of the major government-bond and equity benchmarks this year, though natural gas and the two big cryptocurrencies have fallen further still. Bitcoin fell to about 64,000 dollars and Ethereum to about 1,860, both down on the week, risk-off along with the AI names they have come to track.

The standing qualifier. The gauge reads eight dials now, including, from this week, the energy shock it was previously blind to, so the instrument is watching the one stress that defines the week. Two dials are red, the dis-inverted yield curve and the new energy signal; the other six are calm. Two of the eight, factory orders and insider selling, are carried from earlier dates, and we flag them rather than let you find them. The honest read of a 25 is not that nothing is wrong, but that what is wrong, the oil shock feeding inflation, is a slow-burning risk to the rate outlook rather than a fast one to the market’s plumbing.

On 15 July, a lab founded by the former chief technology officer of OpenAI released a model called Inkling and, in the technical notes almost nobody reads, admitted two things most companies would bury. Its architecture, it said, “largely follows” a Chinese model, DeepSeek-V3. And it had trained the model partly on synthetic data generated by another Chinese model, Kimi K2.5. An American frontier lab took the Chinese blueprint and used Chinese-model output to bootstrap its own. Then it said the quiet part out loud a second time: its next model “will train entirely on internal data rather than using outputs from external models.” A shortcut to the start line, openly labelled as one.

Here is why that matters for anyone trying to value the trillion-dollar bet on artificial intelligence. The mistake is not optimism about the technology. It is pricing the tap and the bottle as one asset. The commodity layer of AI, the raw capability to answer a question, is racing towards free, and increasingly it is Chinese and open: cheap open models now run roughly half the enterprise queries on one large marketplace, up from under 5 percent a year ago, and a benchmark task that costs a couple of cents on a Chinese model costs closer to a dollar-eighty on the American flagship, a spread of about ninety to one for near-identical work. The frontier layer, the small set of jobs where a wrong answer is expensive and you pay for trust, is a completely different business, still Western and still defended. Owning “AI” as a single trade means owning both and being paid cleanly for neither. Open weights are the challenger’s oldest weapon, the same one that turned the web browser from a product you bought into something the world expects for free.

(A reader in the technology industry forwarded the material behind this note; the framing rests on the lab’s own release notes and a cost comparison from Artificial Analysis via Bloomberg.)

The Strait of Hormuz, the 21-mile-wide sea lane through which roughly a fifth of the world’s oil passes, is effectively closed to normal traffic. The ceasefire between Iran and its adversaries collapsed on 8 July; a naval blockade was reimposed on the 13th; the legal authorisation that had allowed some Iranian crude to keep flowing expired on the 17th. Tanker transits through the strait have fallen to a trickle, two or three a day where there were dozens, and the cost of chartering the largest oil tankers has spiked to around 470,000 dollars a day, a level the market last saw in the panics of 2019 and 2022. This is the single largest reason the barrel closed the week above 89 dollars.

The discipline we hold on this story is to watch the insurance market, not the headlines, because war-risk premiums and force-majeure clauses move before prices do, and both have moved. The non-consensus point worth giving a reader is this: the more sophisticated view among the desks we track is not that oil is heading to 110 or 120 dollars in a demand-destroying spike, but that a partial, grinding closure that keeps crude in an 85-to-90 dollar band is something else entirely, a margin tailwind for oil producers rather than a catastrophe. The world is also drawing down its strategic reserves to fill the gap, which cushions the price but spends a buffer that cannot be spent twice. The barrel is the marginal source now, and the reserve is thinning.

Beyond the strait, the more durable geopolitical story is the one running underneath the AI trade: the erosion of the assumption that the technological frontier is American. Chinese open-source models now run about half the enterprise queries on major marketplaces, and this week an American lab conceded it built on a Chinese blueprint. That is not a market you can chart in a day, but it is a slow reallocation of a kind of power, and it belongs on this page as much as any tanker.

The cross-check that keeps the story honest. A genuine supply catastrophe would show up in the long end of the bond market and in credit spreads. Neither has ruptured: the 30-year sits around 5.17 percent and high-yield spreads are still calm at 277 basis points. The market is treating the strait as a priced friction rather than a rupture, which is also exactly what the pricing would look like the week before it was wrong, which is why the insurance clauses, not the indices, remain the tell. For a reader, the strait matters because it is the cleanest current example of how a faraway conflict reaches an ordinary budget: not through the news, but through the price of filling a car and, three months later, through the interest rate on a loan. The 1979 lesson is the one to keep in mind: the oil spike is Act One, and the central bank’s response to the inflation it causes is Act Two, and Act Two is usually harder on markets than Act One.

The founder, in one archetype: the Pack Alpha. Ambani does not compete for a market. He coordinates overwhelming force and takes the whole thing at once, and the Jio launch of 2016 is the purest example on record: not a product that won share, but a move that reset the price of an entire industry to a level only his balance sheet could survive.

The problem, first. In the summer of 2016, India was one of the most expensive places on earth to use mobile data and one of the least connected. A gigabyte of data cost the better part of 200 rupees; hundreds of millions of Indians had never been online at all. The country that would soon have the world’s largest pool of potential internet users was being priced out of the internet.

The bet. Mukesh Ambani, chairman of Reliance Industries and the wealthiest man in Asia, poured more than 20 billion dollars, with the full network and spectrum bill often put north of 30, into building a nationwide 4G network, and then did something that looked less like a business plan than a declaration of war. On 5 September 2016, Reliance Jio launched with free voice calls, free data and free messaging, valid for what would stretch to seven months. He gave away the thing he had spent a fortune to build.

The moment of real uncertainty, and the personal stake. The bet was total, and it was personal. Reliance is a family empire, and Ambani was spending money built by his father, in a business his father had never entered, on a proposition that had to work at a scale nobody had attempted. If the free blitz did not convert into paying customers fast enough, or if the incumbents could match it and outlast him, the loss would be counted in tens of billions and in the family’s own standing. The incumbents assumed no one could sustain free at that scale. The genuine question was not whether Jio could sign users; it was whether it could ever charge them, and whether Reliance could absorb the losses long enough to find out.

The resolution. The answer came faster than anyone forecast. Jio signed 16 million users in its first month, the quickest ramp any mobile operator had ever managed, anywhere. It crossed 50 million in 83 days, a milestone that had taken the older carriers more than a decade each, and passed 100 million by February 2017. When the free period ended, most of them stayed and paid. Data prices across India collapsed by roughly an order of magnitude; the country went from a connectivity laggard to the world’s largest consumer of mobile data, burning through more than an exabyte a month. Two incumbents merged to survive; others did not survive at all. And in 2020 the prize revealed itself: global investors, from Meta to Google to the world’s largest private-equity firms, poured some 20 billion dollars into Jio Platforms, validating a bet that four years earlier had looked like recklessness.

What it teaches. The lesson outlives the deal. Ambani understood that in a network business the asset is not the customer you charge today; it is the standard you set that everyone else must then meet. He did not buy market share. He reset the price of the entire market to a level his balance sheet could survive and his rivals could not, and then he owned the pipe through which a nation’s digital life would flow. The free giveaway was never the strategy. It was the price of admission to a market he intended to define.

A note on the numbers: the launch date, the subscriber ramp (16 million in the first month, 50 million in 83 days, 100 million by February 2017), the investment figure and the data-price collapse are drawn from the public historical record of the 2016 to 2020 Jio launch and verified against contemporaneous reporting. This profile frames a strategic case study; it is not a live recommendation on Reliance shares.

While everyone watched the artificial-intelligence trade come apart this month, the year’s biggest and least-discussed reversal was already complete. Boring old value shares, the banks, the energy producers, the industrial names that spent a decade being left behind, have beaten glamorous growth shares by roughly sixteen to eighteen percentage points so far in 2026, one of the widest gaps in years. Value is up on the year; large-company growth is essentially flat. And almost nobody is talking about it, which is precisely what makes it a Contrarian Corner.

The plain-English reason is interest rates. A growth share is a promise: pay me now for profits that arrive years from now. A value share is a receipt: here are the earnings and the dividend today. When money is cheap, the market pays up for the promise; when money is dear, it prefers the receipt in hand. With long-term interest rates above 5 percent, the receipt has been winning all year, and this year’s value leadership is, in one line, the sound of higher-for-longer finally being believed.

Two honest caveats, because the number is doing real work. First, name the measure: this is large-company US value versus large-company US growth, the Russell 1000 versions of each; a different lens, small companies, global markets, or a purer book-value measure, could show a narrower or a wider gap. Second, “rates did it” is the clean story but it should not run bare: a meaningful slice of value’s lead this year is simply that its heaviest sectors, energy in particular, have had a spectacular run on the very oil shock this edition is about, so some of the “value beats growth” trade is really “energy beats everything” wearing a value label.

The trade that has already worked is rarely the trade in front of you.

Here is the part that cuts against instinct. The obvious trade now, watching AI wobble, is to sell growth and buy value. But a gap this wide is usually a mostly-closed opportunity, not an open one. The lesson value’s quiet sixteen points teaches is not “buy value now”; it is that the market rewards receipts over promises exactly when money is expensive, and the moment to have acted on that was when it was least comfortable to, months ago, when it looked like a mistake.

This week’s fear is that energy is expensive again, and the instinct that follows is an old one: that we are hostage to the barrel, as we were in 1979. So here is a number worth holding against it. In 1980, the world burned roughly twice as much energy to produce a dollar of output as it does today. The amount of energy it takes to generate a dollar of world GDP has fallen by more than half in four decades, and it falls a little more every year.

That decline is not a slogan; it is the slow compounding of a thousand unglamorous efficiencies, better engines, better insulation, lighter materials, the quiet substitution of electrons for molecules across the economy. It means an oil shock, however painful, lands on an economy far better armoured against it than the one that gave us the queues and the stagflation of the 1970s. The same rise in crude that would have been a body blow to the economy of 1979 is, to the economy of 2026, a tax, real and regressive and worth taking seriously, but survivable in a way it once was not.

None of this makes the current shock harmless. A tax on energy still falls hardest on the households that can least afford it, and the inflation it can cause is exactly what has our rate model on alert. The Long View is never an argument for complacency; it is an argument for proportion. The fear that an oil spike must mean a repeat of the 1970s is understandable, because the 1970s taught a generation to fear exactly this. But the machine the shock is hitting has changed. It uses less of the thing that is getting expensive, and it can find substitutes faster than it ever could before. The barrel is dearer this month. The amount of prosperity we wring from each barrel has never been higher, and that number does not go back down.

The world extracts far more prosperity from each unit of energy than it did in 1980, which is why this shock is a tax and not a catastrophe.

Each week the WMP scores the central economic argument about artificial intelligence as a contest between two forces. The Displacer is the case that AI substitutes for human labour, concentrating the gains in capital and eventually destroying the spending power the economy runs on. The Augmenter is the case that AI raises human productivity, expands output, and spreads the gains broadly. Both sides agree AI is powerful; they disagree about whether it is a demand shock or a supply shock.

• Pillar 1, adoption speed, the Displacer. This is the Displacer’s clearest point in weeks: Chinese open-source models leapt to roughly half of enterprise queries on a major marketplace, from under 5 percent a year ago, and one American lab conceded it bootstrapped its new model on another model’s synthetic output, the nearest thing to artificial intelligence training artificial intelligence we have logged, even as the defended frontier stays Western.
• Pillar 2, labour, the Augmenter. No named white-collar layoff wave citing AI as the cause surfaced this week; June’s weak payrolls read as a rates-and-tariffs macro softening, not substitution, and software roles remain resilient.
• Pillar 3, type of shock, the Augmenter, narrowly. Retail sales held positive at plus 0.2 percent, so the income that sustains demand is bending rather than breaking, and the demand-destruction read stays with the Displacer as a risk, not yet a fact. The honest footnote: the oil shock is a regressive real-income tax, and the July petrol bill will test this pillar hard next month.
• Pillar 4, compute cost, the Augmenter. Long rates above 5 percent, hyperscaler free cash flow turning negative and bond-cover ratios under two times are the compute-cost ceiling made concrete: the build-out is being throttled by the price of money, exactly the Augmenter’s argument.

What would flip the score: a negative July retail-sales print would hand the Displacer the economic-shock pillar; a named AI-attributed white-collar layoff wave would hand it labour; and a fall in long rates that cheapens the build-out would harden the Augmenter’s hold on the compute ceiling instead.

The ERDR (Equity Return for Debt Risk) framework tracks twelve income strategies that aim to earn an equity-like return for taking on debt-like risk. This is an even week, so the twelve-strategy dashboard runs in full, each strategy refreshed with its current yield, its spread over investment-grade credit, and any change in thesis. A Deep Dive on a single strategy runs on odd weeks.

The one macro cross-current worth naming this week is the same rise in long-term interest rates that is hurting long bonds: it is lifting the yields available across most of the income complex, from senior secured loans to private credit to preferred shares. Higher-for-longer is a headwind for bond prices and a tailwind for new income. The strategy to watch is private credit direct lending, where the AI-financing story in the Bubble scan is starting to matter: as data-centre operators borrow against fast-depreciating chips, the quality of the collateral in parts of the direct-lending market deserves a closer look than the headline yield suggests.

The same rise in long-term rates that makes long bonds the worst line on the Scoreboard is quietly lifting the yield on almost every income strategy below.

Each strategy is explained in full when it is the week’s Deep Dive.

The read this week: the oil shock left income markets remarkably alone, high-yield spreads at 277 basis points, the long end near flat, so the dashboard’s moves are notes rather than signals. The two standing cautions are unchanged. The leveraged-loan and private-credit complex (strategies 6 and 11) still prices more dispersion in the market than the rating agencies admit. And the energy-linked royalty strategies just received a windfall week that should be read as volatility, not run-rate. No thesis changed this week.

What I am watching and why, not a recommendation to buy or sell. The scoring and ledger material lives here, in its own block, and does not lead the edition. One call is active this week on a fresh catalyst; one reaches its thesis horizon and is scored in public.

USA Rare Earth (NASDAQ: USAR): back on the radar on a real catalyst

USA Rare Earth returns to the main radar this week on a discrete, dated event: on 19 to 20 July the company announced a chief-executive transition, with a new chief taking over on 1 October as the current one retires, alongside the closing of its Serra Verde deal by the end of August. That is a company-specific catalyst inside the seven-day window, which is what earns the section its place.

USA Rare Earth (NASDAQ: USAR) · twelve months · entry $21.00 (Wk 18) · close $14.15 (24 Jul) · thesis-horizon date 16 August.

What the market is missing. USAR has been priced almost entirely as a casualty. The shares sit at 14.15 dollars against our 21.00 dollar entry in May, down roughly a third, dragged down with the whole rare-earths complex after its speculative first quarter deflated. But the thesis was never a quarter’s price action; it was that Western governments would pay, structurally and for years, to build a rare-earths supply chain that does not run through China, and that policy arc has if anything hardened. A leadership change and a closing acquisition are the unglamorous execution steps a market fixated on the drawdown tends to ignore. The historical parallel is the early years of any government-backed strategic industry, where the equity trades on sentiment long before it trades on the policy that ultimately underwrites it, and the sentiment low and the policy peak rarely coincide.

What would change the thesis, pre-registered and unchanged: a regulatory block of the Serra Verde combination, or the strategic-minerals policy support visibly weakening. The thesis-horizon date stays 16 August; a published goalpost does not move. The four-week price driver is rare-earth sentiment and broad risk appetite, currently adverse, and this entry proceeds into that headwind with eyes open.

Talen Energy: scored at its thesis horizon, and it did not pay

Talen Energy (NASDAQ: TLN) · twelve months · flagged in April at $361.01 · closed its window at $359.90, down 0.3 percent while the S&P returned about 5.7 percent.

Talen reaches its thesis-horizon date this weekend, and we score it in public, as we said we would. We flagged Talen in April at 361 dollars on the thesis that nuclear-power demand from artificial-intelligence data centres would re-rate the owners of existing nuclear capacity. Three months on, the demand thesis has proved out, the Susquehanna-to-Amazon nuclear expansion has been enlarged and extended, but the stock has not: it closed the window at 359.90 dollars, essentially flat, down 0.3 percent. Over the same three months the S&P 500 returned about 5.7 percent. So the honest verdict is uncomfortable, and we state it plainly: the analysis of the sector was right and the call still cost you roughly six points against simply holding the index. A correct thesis the stock ignored is a modest miss, not a win, and we log it as one, at minus one on the five-point scale.

The theses have been sound and the entries have not paid, and a run of correct-but-lagging calls is its own lesson about the difference between being right and making money.

The running record, with denominators. With Talen closed, our thesis-horizon record stands at three of three calls directionally correct, Bloom Energy, Freeport and now Talen, but all three trailed their benchmarks, by roughly one, thirteen and six points respectively. That is the honest shape of the record so far. Zero double-misses. These are the calls logged in our call record since Week 13. From this edition the four-week tactical marks are tracked internally and no longer published week to week, because this is an investor’s letter, not a trader’s post-mortem; the full ledger, both horizons and every denominator, is published in the Quarterly Reckoning. The other open calls, Venture Global, MP Materials, Arista, YPF and Mitsubishi UFJ, are tracked in Portfolio Watch.

Spare a thought this week for the central banker’s most thankless job: being handed, two days before you must decide, a fresh oil shock and a jobs market that cannot make up its mind, and being expected to look serene about it. The Federal Reserve meets on Wednesday knowing that whatever it does, the inflation reading that would tell it whether it was right lands two days after the meeting. It is the economic equivalent of being asked to mark your own homework, then being shown the answer sheet only once the exam has already been posted. There is a small dignity in doing a hard job in the wrong order and still being expected to look calm about it, and this week the Fed has it.

Three things to watch next week, and they all point at Wednesday. First, the Federal Reserve’s decision and, more than the decision, the two-year Treasury yield in the hour after it. Second, the July core PCE inflation reading on the 31st, which lands two days after the meeting and will tell us whether the oil shock is bleeding into prices or being absorbed. Third, the barrel itself: does crude hold above 85 dollars, or do the stalled peace talks give the shock back? If all three move the same way, oil holding, the two-year rising, core firming, the September cut is gone and the higher-for-longer world is confirmed. If all three break the other way, this was a tremor and the cut lands in the autumn. If they diverge, oil high while core cools, we get the genuinely hard tape, where the Fed must choose between the inflation it fears and the growth it is starting to lose. We will be here to score it.

Until next week. Stay curious and stay hedged.
Anthony Rosenthal

Crude and the Baltic Dry are not just the year’s best performers, they are the wildest, swinging two to three times as hard as the S&P; the gap between best and worst is 83 points, the widest of 2026, and one shock, oil, is doing most of the widening.

26 assets ranked by year-to-date return · baselines locked 1 January 2026 · close of Friday 24 July 2026. The basket is a fixed set, chosen on 1 January and unchangeable during the year. MSCI ACWI joins this week, tracked from a 1 January baseline like every other line, bringing the basket to 26: twelve equity indices, four bond and credit funds, six commodities, two currencies and two cryptocurrencies. Nothing else is added, dropped or substituted mid-year, which is the only way a year-to-date table means anything. The single-company calls in Portfolio Watch are tracked separately. Annualised volatility shows how much each line typically swings in a year, judged from its last eight weeks: higher means bumpier, not worse, and a dash means we do not yet have enough weeks to measure it honestly.

Every close here is drawn from the same price record the table itself reads from, so the numbers on this page and the numbers in our record cannot drift apart. All closes are Friday 24 July 2026. MSCI EM is derived from the EEM ETF close (63.33) multiplied by the locked index ratio (28.367); Baltic Dry is the Baltic Exchange BDI as published by Hellenic Shipping News (24 Jul, 2,743). Every year-to-date figure is recomputed from the locked 1 January baselines rather than carried forward, so an error cannot compound week to week.

Every company that has appeared in On the Radar is tracked here until its formal score date. A company moves from On the Radar to this appendix when there is no fresh catalyst that week, the analytical call is intact, but there is nothing new to add. Every close below is the verified close we recorded on the date shown beside it. Two names have left the table this week: Bloom Energy and Freeport-McMoRan closed their thesis windows and received their final verdicts in On the Radar.

This week: five active calls tracked here, with USA Rare Earth featured on the radar and Talen closed at its thesis horizon this weekend. Four of the five here are in the money (Arista, MUFG, Venture Global and YPF) and one is behind, with the loser named as loudly as the winners. MP Materials, down nearly 39 percent, is the book’s biggest loss, and USA Rare Earth sits on the same repriced thesis. The marks come quickly now: Venture Global on 2 August, MP on 9 August, USA Rare Earth on 16 August and Arista on 28 August.

You do not need to read it. It is here so that you can.