Worse News, Dearer Money

What you can check: every Scoreboard close is Friday 2 October bar one, which is dated on its own line; every crash-gauge reading carries its own date, and the three that are carried or lagged are asterisked and say why; and every crash-gauge weight and threshold is printed with the arithmetic that reaches the score.

Lending for two years got two basis points dearer. Lending for thirty got fourteen.

The part of Friday’s employment release that moved our own machinery was not its headline. It was the revision table underneath, which the third section sets out, and which turns a soft month into a trend: a hiring pace no longer fast enough to absorb new workers.

So the raw material for a dovish week was there and our own rate model took it. The composite moved from 21.3, hike-leaning, to 9.7, hold, its largest single-week move. The rate implied by its policy-rule anchor fell from 4.90 per cent to 4.35 against a funds rate at 3.875, so the gap between where the rule says rates should be and where they are narrowed from 1.03 points to 0.48. A positive gap still means a cut is not yet rule-justified; a narrowing one means a cut coming into view.

Two of the model’s three evidence blocks moved the same way, the minimum it needs before the rate view may move. Inflation went dovish on August core personal consumption at 3.0 per cent against 3.3, services at 2.99; labour went dovish on the payroll trend; markets and financial conditions went the other way.

And an admission is owed, because the model reached the right band by the wrong road. Last week printed the pair of conditions that would flip the rate view dovish: payrolls below 100,000 and the two-year back under 4.45 per cent. Payrolls delivered. The two-year rose instead, to 4.83, and rose five basis points on Friday itself into the weak print. Half a two-part test fired and the composite moved anyway, carried by the inflation figure and the policy-rule anchor. So the reading is the band the model gives, hold with a mild hawkish lean, not a cut coming, and it is provisional: 88 per cent of the panel had fresh data and one inflation cell had none.

The strongest competing explanation is not growth or inflation. It is supply and flows. The Treasury funds a very large deficit at the long end, and month and quarter-end rebalancing lands in the same window. The signature of a flow-driven long end is what the curve shows: the twenty-year yielding more than the thirty-year. No economic story requires that, and a mismatch between who must buy which maturity produces it easily. That read has a separating test, published and dated. If the gap closes to zero or better by 16 October while the thirty-year holds at or above 5.35, the economics own this move. If it stays inverted by three basis points or more, the mechanics own a material part of it and this section is too confident by that part.

Where I could be wrong. The reading rests on the long end moving for cost-of-capital reasons, and two things would undo it. The revisions could be the story rather than the level: if 29,000 is itself revised up next month, this week’s labour data was noise, the model moved on a number that did not survive, and the bond market was right to ignore it. And the rate model may be leaning on one cell. Its dovish turn came mostly from the policy-rule anchor and a single inflation print, on a panel 88 per cent complete with one inflation cell empty. If that cell fills hot, the composite goes back towards hike-leaning and this week’s reading of it will have been a month early.

Four data producers, four days, three different countries described. Then every asset that depends on the cost of money fell at once.

At half past eight on Friday morning in Washington the Bureau of Labor Statistics posted the document it posts on the first Friday of most months, and for about forty minutes the market traded it the way the textbook says. Jobs light, yields down. Then the yields turned and by the close every maturity sat higher than on Thursday. The forty minutes is the interesting part: the reflex is still there and no longer decides the day.

Who pays. The Bureau’s own figures: 29,000 jobs in September and unemployment up to 4.2 per cent, with July and August cut by 60,000 between them and August now standing at 133,000. The three-month average is near 51,000. Set that beside what the Conference Board had published on Tuesday 29 September, a consumer confidence index of 81.9, down 6.7 points from a revised 88.6 and the lowest reading since 2014. And beside what the Bureau of Economic Analysis published on Wednesday 30 September, a personal saving rate that jumped to 4.1 per cent from 3.0. Households are earning less confidently and saving more of it, which is the oldest defensive move there is.

Capacity. Then on Thursday 1 October the Institute for Supply Management published its manufacturing survey, and the new-orders sub-index rose 1.6 points to 55.3, with the headline factory index at 54.5. Read that number carefully, because it is a headcount and not a quantity. The survey asks purchasing managers one question, whether their order books grew or shrank, and reports the net. Above fifty means more said grew than shrank. The move from 53.7 says that majority got bigger, and nothing at all about how much anybody ordered. Three of the four readings, from three different producers, one government and two private, and no single story holds all of them: weakening employment, collapsing confidence, and order books widening. A reader who wants one narrative this week will have to discard a measurement to get it, and the point of printing all three is that none of them should be discarded.

Forced flows. The bond market’s answer was to charge more at every maturity it charges by the decade. The two-year rose 2 basis points on the week to 4.83, the five-year 8 to 5.06, the ten-year 11 to 5.28, the twenty-year 13 to 5.67 and the thirty-year 14 to 5.63. The twenty-year is still yielding more than the thirty-year, by four basis points, narrowed from five. Long-dated government bonds remain the worst line on our board, now down 17.81 per cent for 2026, and the aggregate bond index is down 7.73.

Cost structure. What fell with them is the list that makes the week legible. Silver dropped 6.64 per cent to 59.98 dollars an ounce, gold 3.68 to 4,162.30, copper 3.04 to 6.49 a pound, natural gas 5.04 and the Baltic Dry freight index 8.11. In credit, high-yield fell 1.22 per cent and investment-grade 1.34. The FTSE 100 fell 2.18 per cent, the Swiss market 2.04, the Euro Stoxx 50 1.02. On our own measurement, high-yield credit, the FTSE and the aggregate bond index all moved at or beyond two standard deviations of their recent weekly range, on eight-week annualised volatility of 3.7, 7.7 and 3.3 per cent respectively, and every one of them moved down. The larger percentage falls were not the unusual ones: silver fell 6.64 per cent and natural gas 5.04, and on our own measurement both sit inside two standard deviations, because those series swing that much in an ordinary week.

Where value is captured. And American shares did almost nothing. The S&P 500 finished at 7,722.72, down 0.27 per cent, within one per cent of the 13 August high of 7,798.99. The Nasdaq 100 rose 0.65 per cent. The Russell 2000 fell 0.16. The only major equity index that clearly rose was Japan’s: the Nikkei 225 gained 2.93 per cent to 68,309, and is now up 31.80 per cent for the year, the best equity line on the board. A week in which bonds, metals, credit and European shares all fell together while American and Japanese equities held is not a risk-off week. It is a discount-rate week, and equity markets are the last to be asked.

Two dials changed band this week and they changed in opposite directions, and the one that deteriorated matters more than the one that improved.

The composite is 27.5 out of 100, up 2.5 from 25.0 after three editions unchanged. Two dials moved band and nearly cancelled: bond volatility green to amber adds 7.5, the energy dial amber to green takes away 5. Everything else held. So the score rose for a reason unconnected to what made the week, which is how a weighted count of eight conditions behaves.

What the gauge does not capture is the most interesting credit event since the last edition, because it measures prices and this was a contract. On 24 September, which falls inside the previous edition’s window and was not covered there, Oracle sent a force majeure notice to Blue Owl, its lender on Project Jupiter, a 2.45 gigawatt campus in New Mexico. Its shares fell on the day, by three per cent on CNBC’s count and about four on Bloomberg’s. The cause is worth more than the headline: a gas pipeline, not a change of mind about demand. The state land office refused rights of way in March and July. The operator rerouted across federal land and the in-service date moved from August to February. Oracle wants the right to defer payments if the campus is not open in 2028, not to walk away. This is reputable secondary reporting; the notice is unpublished.

On our own tracking of contracted artificial-intelligence obligations, this is the first to be publicly re-underwritten, and the binding constraint turns out to be a permit. Every bear case on this build-out has been written about demand: that the revenue will not arrive, that the models will commoditise, that the chips will be obsolete. The first thing actually to slip is a right of way across a piece of land. A thesis about whether the computing is worth it does not price a pipeline, and the people lending against these campuses are discovering that they have underwritten a construction schedule rather than a technology.

The money being raised to build artificial intelligence is now counted in the units national budgets are counted in. The thing it sells keeps getting cheaper.

Where value is captured. OpenAI is reported to be in talks to raise about 30 billion dollars at a valuation near 1.4 trillion, per TechCrunch on 29 September, alongside separate reporting of a compute agreement worth something over a billion dollars a month. Every figure there is reported rather than filed, so it is attributed and not adopted. What matters is not the valuation but that the commitments are increasingly to each other: a compute buyer and a compute seller signing a contract that is revenue on one side and cost on the other, inside a circle of about six companies.

Commoditisation. Against that, the price of using the output keeps falling. Each frontier release of the last eighteen months has arrived priced below the version it replaced while doing more, which is the signature of a product whose unit economics are improving and whose pricing power is not. The reading this letter has held since the summer follows: the returns here may well be real and may well land with the people using the tools instead of the people selling them. That is a question about where profit goes, not about whether the technology works.

Which brings this section to something a reader can do this week, and it comes straight out of the credit widening in the fourth section. High-yield borrowing costs moved sharply in five days. That is irrelevant to a company that funds itself and expensive for one that must refinance next year. The difference between those two is a piece of information you can extract from documents you already have access to, and extracting it is a four-minute job rather than a research project.

What happens to a geopolitical premium when nothing happens?

It deflates, and this week it did. Nothing material escalated or resolved at the Strait of Hormuz, the standing flashpoint this section has led on for three months. Crude sat in the low nineties and fell 1.41 per cent to 91.11 dollars, and the premium built into it over the summer has now largely gone. That is the whole Hormuz entry, and it is kept short because an absence of news is a fact about a market, not a reason to manufacture one.

Enforcement. The sovereign spread that moved is French. The gap between the French ten-year and the German, which prices the risk of lending to Paris rather than Berlin, is above 110 basis points, against a stated 54 billion euro consolidation with no agreed funding behind it, per a Deutsche Bank note we hold as a summary. The second-largest borrower in the single currency is repricing on domestic politics and not on the central bank. It happens to fall in the same week as the American move described in the second section, and the two have nothing to do with each other. That is a coincidence of timing rather than a cause, and a reader is better served knowing it than being handed one global story.

On 24 June 2025 a capsule came through the atmosphere, survived the hot part, got its signal back, and then disappeared in the last few minutes.

Everything had worked: the ascent, the separation, the stabilisation, the reentry, and the reacquisition of the radio signal after the blackout reentry always causes. That is the sequence which kills most first attempts and it had been completed. Then contact was lost minutes before splashdown and the vehicle was not recovered. On board were customer payloads, among them human ashes entrusted to a memorial-flight service. The Exploration Company published the loss itself, called the mission a partial success and a partial failure, which is an unusually exact description, and opened a root-cause investigation. No root cause has been published since.

The problem, stated plainly

Europe cannot bring anything back from orbit. It can send things up on its own rockets and has sent people up on other people’s, and it cannot return mass to the ground. It could once: the Automated Transfer Vehicle flew five missions to the International Space Station between 2008 and 2014, and the programme was allowed to lapse. Hélène Huby spent her career inside the organisations that built that lineage, at Airbus and ArianeGroup, where she worked on the European Service Module, the part of NASA’s Orion spacecraft Europe supplies. She knows which capability was given up because she was in the building when it went.

She is, in this letter’s shorthand, a Salvage Diver: someone who goes back down for what was abandoned. They can, because they were on the boat when it went over the side. No searching required. They know the depth, what it was worth, and which bolts were left loose. Huby’s founding team came from the same programmes, so the one accusation that cannot be made here is learning the domain in public.

And then the part that is genuinely unresolved

The company has sold ten missions to four destinations: the International Space Station, and three commercial stations called Axiom, Starlab and Vast. Three of those four do not exist yet, and none of the commercial stations has flown. The fourth is being retired, and the firm European contract requires the demonstration to dock there no later than the second quarter of 2029, the last practical window before it comes down.

So the schedule is not a commercial deadline to be renegotiated in a difficult quarter. It is set by the orbital decay of a destination. If the demonstration slips a year the only certified place to dock is gone and the 450 million euros of options have no mission to attach to. A book described as more than two billion dollars of contracts and commitments then reverts to being mostly commitments, from counterparties that are themselves pre-revenue. The company does not disclose how that two billion splits, so until it does the firm backlog is the 310 million euros the agency signed.

Underneath the schedule sits the harder problem, and it is June 2025’s. A reusable capsule that cannot be recovered is an expendable capsule carrying a reusable capsule’s cost base. The whole economic case, up to ten flights each, rests on getting it back, and the one attempt at the full sequence did not. The early warning signal is public: the next subscale flight. If that also fails at terminal descent or recovery, the problem is architectural rather than one piece of workmanship, and no amount of sovereign enthusiasm fixes an architecture.

Where the durable advantage actually sits

The framework this letter runs on private companies was applied before any of this was written and comes out at watchlist, on public information with no company document read. Its conclusion is the useful part: it validates the strategic reading and contradicts it as an investment at the entry price it must estimate, because on that estimate the deal clears no ordinary return hurdle even in its base case. The thesis can be right and the price still wrong.

Nine of the last twelve Fridays priced interruptible capacity below the 1.65 dollars an operator needs to cover costs, and the two most recent are the lowest in the series. The guaranteed rate has not followed it down: it is higher now than it was in July.

Dollars an hour to rent one previous-generation accelerator, Fridays, 17 July to 2 October, open-market medians from vast.ai. Interruptible: 1.333, 1.333, 1.867, 1.733, 1.467, 1.333, 1.333, 1.267, 1.467, 2.000, 1.067, 1.067. Guaranteed: 2.161, 1.923, 2.642, 2.059, 2.268, 2.267, 2.269, 2.896, 2.761, 2.856, 2.301, 2.403. The 1.65 line is an operator cost estimate this letter has not verified, so it frames the chart and scores nothing.

Condition one, memory contract prices falling: not met, and the latest published figure is still an increase. The condition needs contract prices to print down quarter on quarter. Down is the test, not decelerating. This is a quarterly series rather than a weekly one: conventional DRAM contract prices are projected to rise 10 to 15 per cent quarter on quarter in the fourth quarter, published by TrendForce on 30 September. The next print is due in January, and until it lands this condition cannot change.

The same series bears on a claim this letter made last month. In September we called the rate of increase decelerating hard, and on the stored series it is: the quarterly rise runs 95 per cent, then 63, then 18, then the 10 to 15 now projected. The level keeps climbing and the rate of climb is falling. Scope stated because it carries the argument: only the fourth-quarter figure holds a recorded source and date, the three earlier quarters being carried forward from our own prior record, so the latest print is the sourced one and the trend behind it is indicative. Micron’s results of 30 September agree on the level: fourth-quarter revenue of 54.23 billion dollars against 41.46 billion in the quarter immediately before it, with gross margin at 86.8 per cent against 84.6, on the company’s own release. No business posts that quarter on quarter into falling prices. The release says nothing about contract prices, so it corroborates the direction without being the series this condition is defined on.

Condition two, counterparty funding stress: not yet built. Neither met nor unmet. It stays blocked until five financier instruments are named and shown both genuinely liquid and genuinely dependent on these contracts. A basket failing either test produces a number worse than none. One candidate arrived this week and is logged as a candidate only: the project debt behind Oracle’s New Mexico campus, quoted by a sell-side desk at 89 to 91 cents in late September. That is the shape the basket needs. It is one instrument of five.

Reported, not scored. Friday’s interruptible rental rate for the previous generation of accelerator was 1.067 dollars an hour and the guaranteed rate 2.403, with 37 live offers behind the median, up from 23 while the interruptible price did not move at all. That rate has sat below the 1.65 dollars an operator needs to cover costs for a sustained period, and the level is not yet defended by a verified cost build, so it is reported and does not score. The newer generations come from a separate and far sparser poll: three readings in the last eight weeks, the most recent on 11 September, and none since. They point the other way. Between 18 May and 11 September the guaranteed rate went from 2.80 dollars an hour to 4.59 for one generation up, and from 2.12 to 6.79 for the one above that, while the generation on the chart fell. Not observed this week rather than estimated.

The book, beside the thesis and never merged into it. Of the seventeen names this gauge watches, four sit more than twenty per cent below their own ninety-day peak, and the worst is 28.3 per cent down. Those four are in foundry, memory and data-centre power equipment. Meanwhile three of the seventeen, the largest chip designer, the largest foundry and the sector index itself, closed Friday exactly at their ninety-day highs. A strategically interesting thesis and a bifurcating book are two different facts and this section prints both.

The binding constraint, and it moved. For most of this year it has been electrical power and the ability to connect it. Since the last edition it became narrower: not generation, not grid capacity, but the legal right to cross a particular piece of ground with a pipe. That put a 2.45 gigawatt campus’s planned 2028 opening at risk, in a notice dated 24 September, and no trigger on the original list covers it, because the list assumed the bottleneck stays where it was put. Capital attacks a bottleneck, capacity arrives, and the constraint moves somewhere nobody is watching. Permitting and rights of way are where it is now.

Three things to watch, each verifiable. Whether a second contracted campus invokes force majeure or renegotiates before the year ends. Whether the Oracle project debt trades below 85 cents, turning a repricing into a mark. And whether the next memory contract reset, due with December-quarter reporting, prints down quarter on quarter, which is the only thing that moves condition one.

What would re-tighten scarcity: a memory contract reset printing up again alongside the newer accelerator generations becoming consistently quotable at premiums to the previous one, which together would say the supply response is failing and the shortage is structural rather than cyclical.

Positions changed by this gauge since inception: zero. That line stays on this page every week until it is not zero.

Arte Moreno bought a baseball team for 184 million dollars in 2003 and agreed to sell it for four billion in September.

He is a billboard advertising man who bought the Los Angeles Angels from the Walt Disney Company and then presided for twenty-three years over a team that holds the longest active post-season drought in Major League Baseball. On the field, by any measure a supporter would use, the period was a failure. As an investment it returned about 14 per cent a year, compounded, for twenty-three years, which is my own arithmetic from the two prices. American shares did roughly eleven to twelve over the same period. He beat the index by owning something that never had to report a quarter.

The buyer is Stan Kroenke, who already owns the Los Angeles Rams, the Denver Nuggets, the Colorado Avalanche, the Colorado Rapids and Arsenal. Reporting puts the price at four billion dollars, attributed to unnamed sources rather than confirmed by either party, and it needs 23 of the other 29 owners to approve after a committee review, with completion expected in the first quarter of 2027. It would be a record for the sport, two and a half per cent above the San Diego Padres at 3.9 billion earlier this year.

What makes this a markets story this particular week is the discount rate. On Friday the American thirty-year government bond yielded 5.63 per cent, so four billion dollars put there instead would pay about 225 million a year, with no approval vote and no supporters to disappoint. The arithmetic the buyer accepts therefore writes down exactly. Matching that bond over ten years needs the franchise worth about 6.9 billion by 2036. Beating it by the four points a private-equity committee asks of a control deal needs about ten billion. An ordinary fifteen per cent return over five years needs 8.05 billion by 2031, a doubling, unlevered, with nothing distributed meanwhile.

Our framework, run before this section was written, reaches watchlist and not top-decile, and fails its own coverage test for the most interesting reason in the asset class: no disclosed revenue, no audited accounts, therefore no multiple. Nobody outside the two principals knows what was paid for what. Its base case is a 2031 value near 5.6 billion, 1.40 times the money and about seven per cent a year, or 133 basis points over that thirty-year bond, for an illiquid, unaudited, approval-gated asset. Its bear case is 3.4 billion and minus three per cent a year.

Why anyone pays it anyway

Because the supply side is the most rigid in finance: thirty memberships, no relegation, no expansion unless existing owners agree to dilute themselves, perpetual territorial rights, and no mechanism by which a membership lapses. The buyer pool is small, growing and almost insensitive to yield, because it consists of people acquiring a position and not a cash flow. Against the Padres six months earlier, four billion for the second-largest media market in the country is not even expensive.

Where the price is actually at risk

Not in the owners’ vote, which will probably pass. In the land. The City of Anaheim owns the ballpark, which opened in 1966, and the roughly 150 acres around it, and certain tenure runs only to 31 December 2032. The league’s own material says the buyer may seek to purchase and develop the site, drawing the comparison with the stadium he built in Los Angeles. A previous attempt to sell that land to the current owner for 320 million dollars was approved in 2020 and collapsed in 2022.

So a material part of four billion dollars is an option on real estate the buyer does not own, from a seller who walked away from the same deal once, with just over six years left on the lease. That risk appears in no franchise-value table, and it is why the clean observable signal here is not a price: it is whether a land negotiation with the city opens during 2027.

What generalises has nothing to do with baseball. Scarce assets are priced on who can still clear the number, not on their cash flows, and when money gets dearer the buyer pool does not usually get more price-sensitive. It gets smaller. The buyer here is not new money: he already owns teams in five leagues and is buying into a sixth, in a week when most other long-dated assets were marked down. That is what happens when the appeal was never the yield in the first place.

Two each, and the running total goes to fifty-one against twenty-five.

The framework and why it is scored this way are on the standing method page. This week’s evidence, one observation per pillar and four different observations between them.

Adoption speed, to the Displacer. A private-credit data provider reported this week that software dealmaking has moved substantially into private bilateral negotiations, with valuations bifurcating according to whether a company can show its artificial-intelligence position is a help or a threat. That is second-hand and the provider’s own framing, so it is attributed rather than adopted. But a sector whose valuations now split on that single question is a sector in which buyers believe the displacement is real enough to underwrite. Markets pricing a thing is weaker evidence than the thing happening, and it is still evidence.

Labour, to the Augmenter, on an absence, and weaker than usual for it. No named employer attributed a headcount reduction specifically to this technology in a filing this week. Under the rule this letter adopted in August, a pillar awarded on an absence carries a positive fact beside it where one exists. This week one nearly does. At its capital markets day on 30 September, BMW told investors it would cut about a fifth of its most senior management, roughly a hundred roles concentrated at its Munich headquarters, by the middle of 2027, and its finance director named the continued rollout of this technology’s agent applications as the mechanism. It was said at an investor event and not in a filing, so under the condition this ledger pre-registered before the week began it does not score, and the condition is not being loosened after the fact to admit it. It is the closest the falsifier has come in nineteen weeks, and a filing that repeats it takes the pillar outright. The pillar is still the weakest of the four.

Type of shock, to the Displacer. Global private-equity exit value reached about 462.9 billion dollars across 1,009 sales in the third quarter, the strongest quarter but one since the second quarter of 2021, while fundraising ran roughly ten per cent below last year. Taking money out at a near-record rate while raising less of it is capital leaving the private economy rather than being recycled into it, and the distribution of that cash goes to the people who already owned the assets. That is the type-of-shock pillar almost by definition: a gain concentrated in capital rather than in wages.

Compute constraints, to the Augmenter. Oil and gas private-equity deal activity is reported down about 65 per cent, attributed to uncertainty around the Iran conflict. Also second-hand. The relevance is not energy prices, it is that the capital which would expand the energy supply the build-out depends on is not being committed, and displacement requires the computing to exist before it can displace anything. A constraint on the arrival of compute is, mechanically, time bought for the humans currently doing the work.

Week 38: the Augmenter 2, the Displacer 2. Running total, the Augmenter 51, the Displacer 25. Nineteen weeks at four pillars each is seventy-six, and 51 plus 25 is 76, so the ledger reconciles.

What would flip the score: a named large employer attributing a headcount cut specifically to this technology in a filing takes the labour pillar outright and takes it on evidence rather than on an absence, which is the single cheapest and most decisive change available to the Displacer.

There is no new call this week, and the reason is worth more than a call would have been.

This section exists to put names and prices on the record and be scored for them, and it has produced no new name since edition 25, thirteen editions ago. Either nothing qualified or the machine looking is broken, and a reader can only tell which if both routes report every week whether or not they found anything. This week one route produced two candidates and killed both. The other did not run.

The one that is worth the story

AZZ Inc. (NYSE: AZZ) arrived in an unusual way: it was surfaced independently by two separate external research threads in the same week, which is normally a good sign. It is a metal-coatings and electrical-infrastructure company, and the pitch is the obvious one for 2026, that electricity transmission has become the bottleneck in everything and the companies that galvanise and build the hardware get paid for it.

It cleared the first gate. No kill criterion fired. The market-already-moved test was comfortable too: at 135.41 dollars the shares sat about 16.5 per cent below their high, so the tape had not taken it. The quality pre-screen said proceed with one red flag for insider selling, a financial-strength score of seven out of nine, momentum green. The technical detector read no edge at twenty points of 111, which is the expected reading for a name chosen on mispricing and is a dial that sets conviction, not a gate.

And then it died on the cleanest possible ground: there is nothing for the market to be missing. On its own earnings call of 22 April 2026, management itself presented the data-centre, poles, towers and substations story, and analysts on that call asked about data centres. The grid link is not a variant perception. It is the consensus, delivered by the company, to the people who write the research. A thesis that consists of agreeing with the earnings call is not a thesis.

Three further things sat against it, recorded because a candidate that fails one test can pass next time while the others stay true. The share of galvanizing volume going to transmission and distribution, the number the whole case turns on, is not disclosed. The price sits above the valuation the framework reaches, implying about seven per cent growth already paid for. And the metal-coatings margin is down about 260 basis points with zinc expensive. The full analysis came out at watchlist and was revised down by its own adversarial pass.

Re-examination point, pre-registered: second-quarter results on 14 October 2026. It converts only on two conditions, both stated now so neither can be invented later: a disclosed transmission-and-distribution share of volume, and a price at or below about 112 dollars.

The other one, briefly

Apollo (NYSE: APO) came from a separate research run after the Thursday screen found nothing. It was declined on 1 October on a fact in the company’s own quarterly filing. The net spread at its annuity business was 1.14 per cent against 1.22 a year earlier, with the cost of funds on new business up to 3.83 from 3.68, in a year when long yields rose. The thesis was that a long-end sell-off widens that spread. The filing says it narrowed. A thesis with the wrong sign is not a cheap thesis, it is a different one. There was also no variant perception, and the company is a principal lender into artificial-intelligence infrastructure, the exposure the fourth section argues is being re-underwritten. Re-examination: third-quarter results, 3 November 2026.

One process note, because it bears on how much these candidates should be trusted. The research tool that produced the Apollo name misstated figures for all six companies it returned. It is usable for names and for dated source references and not for numbers, and every figure above was taken from the filings instead.

And the part that is a finding rather than a report

The section going quiet is a reporting decision. The pipeline going quiet is not, and this is the pipeline.

The second route, the one screening a ranked universe for names before they move, did not run this week and has not run properly for seven weeks. The table it reads was last refreshed on 14 August, forty-nine days ago. The table it used to read was last scraped on 3 August and has produced no new entry in sixty days, and it returns duplicate rows, so forty of them held about seven distinct companies. A screen for companies before their move, working from a universe seven weeks stale, is not a quiet market. It is an instrument reported broken in six consecutive editions without acquiring an owner, which is why the target of one new name every three editions has been missed thirteen editions running.

“A value play at 5.2 per cent.”

That was Jim Bianco on 28 September, his first bullish call on American government bonds since 2020. Within two sessions the ten-year was at 5.29 per cent and the entry about nine basis points offside. It closed Friday at 5.28.

In the same window the research service publishing as MacroEdge called the rise in long yields parabolic and said the frightening move was still ahead, and put a date on it: the ten-year above its 28 September close of 5.24 on 30 October. Same instrument, same week, opposite conclusions, both on the record and dated. This section exists for weeks like this, and what it does with a disagreement of this quality is print both positions with the condition that settles each, then say how much either opinion should weigh.

The answer is: not very much yet, and the reason is arithmetic rather than disrespect. This letter scores outside forecasters’ dated claims. Bianco has four scored calls, three correct; MacroEdge has four, three correct. The gate at which a record starts weighting a view here is five, and neither has reached it. So the ledger does what it should when two credible people disagree and neither has a measurable record: it refuses to break the tie and forces the view to amber on the evidence instead of settling it by reputation.

The contrarian position here is not bullish or bearish on bonds. It is that the week’s most confident commentary, in both directions, was written about a move whose cause is still undetermined, and that the strongest views belong to the people whose records are too short to weigh. A market whose loudest voices are unscoreable is not short of information. It is short of accountability, which is why this letter publishes a line and a date every week and then scores it in public.

The jobs number landed quite small,
Which usually softens the call.
  But lenders read deeper,
  Declined to be cheaper,
And raised what it costs us all.

Four things land inside the next four weeks and each one settles a different part of this edition. The services survey around 5 October, which replaces a carried input in the rate model. September consumer prices around 15 October, which is the first inflation reading after the softest core figure of the cycle and the one that decides whether 3.0 per cent was a turn or a month. The twenty-to-thirty gap on 16 October, which is the published test of whether this week was economics or mechanics. And the central bank meeting on 27 and 28 October, into which our own model arrives at hold while the two-year arrives at 4.83. If inflation is soft and the gap closes, the long end was right about the cost and our model was right about the path, and both can be true. If inflation firms while the gap stays open, then the week was mechanics on top of a cost shock, the model is early, and the uncomfortable combination arrives: dear money, a weakening labour market, and no instrument in this edition pointing at relief.

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

The two ends of this table came back together, and not because the bottom recovered. Freight gave up nearly fifteen points of its lead and long-dated American government bonds fell to a new low for the year, so the distance between first and last narrowed from 97.9 points to 85.1.

26 assets ranked by year-to-date return · baselines locked 1 January 2026 · close of Friday 2 October 2026. The basket was fixed on 1 January and cannot change: twelve equity indices, four bond and credit funds, six commodities, two currencies and two cryptocurrencies. Five rows are abbreviations: MSCI ACWI (All Country World Index) is the broadest global share index, AGG the US aggregate bond market, LQD investment-grade corporate bonds, HYG high-yield, and TLT long-dated US government bonds. The single-company calls in Portfolio Watch are tracked separately. The masthead figure scores this year’s thesis on five pre-registered lines, not all twenty-six: shares, bonds, gold, crude and high-yield credit, two up and three down across a 51.9-point range. A wide range with split signs is the thesis working. The five priced commodity rows are front-month futures, not spot, so gold reads tens of dollars above a spot quote; the Baltic Dry is an index. Annualised volatility is how much each line typically swings in a year, judged from its last eight weeks: higher means bumpier, not worse. The lira reads under one per cent because it crawls by the same small amount each week, which is a managed rate and not a quiet one. One line has no Friday price: India’s exchange was shut on 2 October for Gandhi Jayanti, so the Nifty 50 is carried at its Thursday 1 October close and its week is a four-session week. The closure was confirmed against the exchange calendar we keep, not inferred from the other markets being open.

Every close is read from the same price record the table draws on, so the figures here cannot drift from the figures we hold. All closes are Friday 2 October 2026, except the Nifty 50, which carries its Thursday 1 October close because India’s exchange was shut. Emerging markets is the one derived line: the verified fund close of 67.67 multiplied by the ratio 28.367, locked at the start of the year and never recalculated, gives 1,919.60. The fund tracks the index and is not the index.