He’s right, and the selloff is the proof. I’m buying the most hated trade in the market into June 17.

If you were loud about gold at five thousand and you have gone quiet at four thousand two, this is for you. If you sold somewhere in the March panic, it is especially for you, because the thing that looks like the thesis dying is the thesis working, and I can show you exactly why.

My own money first, so you know I’m not narrating from a desk. Gold miners are about a fifth of my book and have been for most of a year.

I’ve been long the whole way down, I called the March bottom in Gold Is Not Failing, and I have put my misses next to my hits in public. So this isn’‘t a victory lap.

It’s the part where the one man who can make the bear case on gold better than any bear alive sat down on a podcast and made my argument for me, then handed me a second wall I had underweighted.

Gold sits around four thousand two hundred dollars, down 10% over the past month, a quarter below its January high, into a second straight weekly decline.

Open any wire and you get the same story. Iran. Hormuz. Energy feeding inflation. A new Fed chair with a hawk’s reputation. War is bad for gold, the screen says, because war means inflation, inflation means higher for longer, and higher for longer means you dump the metal that pays you nothing.

That reading is backwards, and the drop is how you know.

Subscribe now

Nobody abandoned gold in May. The ETFs shed two billion dollars, sixteen tonnes, four tenths of one percent of what they hold, and total holdings stayed at 4,121 tonnes, within a percent and a half of February’s record, with flows for the year still positive at seventeen billion.

That is not capitulation. It is a crowd that rode the easiest macro trade of the first quarter, watched a firm dollar and a repriced Fed raise the cost of holding it, and rotated into technology to keep up with a benchmark.

Tech ETFs took their biggest inflow since early 2024. The sellers were North American profit takers stepping aside and Chinese holders selling a weaker local price into a stronger renminbi. Europe, the one balance sheet most exposed to the fiscal arithmetic this whole thesis rests on, bought. People rotated out of gold to chase a hotter trade.

That is not the same as giving up on it.

CHART: Global gold ETF flows by region and average gold price

The real forced selling, the violent kind, already happened, back in March, the dollar liquidity event I wrote about at the time.

By May it had dwindled to a trickle with holdings near a record. The capitulation everyone keeps bracing for is behind us.

That was the moment to buy, and we are past it.

Look at what the market has priced into this meeting.

It has thrown out every rate cut it once expected for 2026 and now leans toward a hike by year end, on the hottest inflation in years and a chair who wants to look tough.

Under the old rules that is the worst backdrop gold can face, rising real rates with a Fed pushing them higher, and gold is just sitting in it, taking it. All of which is a problem only if the market is right that the Fed can deliver those hikes.

It cannot, and that is the entire trade.

Then I listened to Jim Grant.

Grant has put out Grant’s Interest Rate Observer since 1983 and owned gold for forty years, and the forty years matter more, because he is no gold bug.

He will tell you cheerfully that the bear case is close to airtight.

It pays you nothing, and Buffett and Munger were right that it just sits in the vault. He bought near the 1980 top at $850, watched it fall to around $500, and sat through the miners losing eighty and ninety percent from 2011 to 2016. He has been punished by this trade for longer than most of my readers have been alive, and he still holds it, because by his measure the dollar has lost about ninety nine percent of its value against gold, and that is not a forecast, it is arithmetic.

So when he looks at the same tape everyone is selling and says, flatly, that war is bullish for gold, the timing is worth your attention.

His point is not the war. It is the Fed.

Here is where he meets the argument I made in Gold Is Dead Money nine days ago, and extends it.

I laid out the trap in its public form. A government whose net interest bill has doubled toward a trillion dollars, with a third of its marketable debt to roll inside a year, cannot carry positive real rates for long without the interest line spiralling.

The exits from that are default or debasement, and both are bullish for the one asset that is nobody’s liability.

That is the sovereign wall, and it is the one that binds now. Grant arrives at the same trapped Fed from the private side, the side I had underweighted, though it binds later rather than this quarter.

The leverage that private equity and private credit took on at near zero in 2020 and 2021 has to be refinanced, and whenever it lands it lands at multiples of the rate it was struck at. A Fed that commits to higher for longer is not just holding rates, it is guaranteeing that wall refinances at punitive levels when it arrives.

So the box has two walls. The sovereign one is now, the private one is building, and a Fed staring at both cannot deliver a sustained hiking path, whatever the market is pricing this week, without breaking something.

And then Grant named a number I have been circling in the private credit work.

By the broadest measure, the one Moody’s uses, about a third of six trillion dollars of life insurance assets sits in private credit. The cleaner, narrower cuts are lower, closer to a trillion in private placements, and the all-in number lands somewhere in between depending on definition. The direction is not in doubt and the structure is real. A large and growing slice of insurer balance sheets, floating rate or close to it, sits under the Apollo and Athene model that minimises equity and maximises spread, the model in which Apollo originates and oversees essentially all of Athene’s assets.

That is the leverage I have been mapping in the private credit series, wearing the costume of a constraint on the Fed.

It is also, precisely, the redemption mechanism I described in Nobody Wants Gold.

When a margin call comes, you do not sell the illiquid thing that caused the problem, because you cannot.

You sell the liquid thing that has nothing to do with it, and in a forced unwind that liquid thing is gold.But be clear that this is not what drove the selloff in front of us.

The May flows show the sellers were discretionary generalists rotating into technology, not margin called balance sheets dumping bullion. This is the forward risk, the second wall.

When the private leverage refinances at punitive rates, the same balance sheets the Fed cannot afford to break are the ones that reach for the liquid asset, and that is the channel that turns a credit event into a gold flush.

Today it is a threat still loading, not the mechanism behind the tape. Two walls. One of them has not arrived yet.

You will object that the Fed already did this, that it took the funds rate above five percent in 2022 and 2023 and the system held. It did not entirely.

That cycle broke the British pension complex and three sizable American banks, Silicon Valley, Signature and First Republic. What it didn’t do was cascade, and the people who promised it would were wrong.

So why’s the constraint tighter now? Because the cushion that absorbed those hikes is thinner. The debt was cheap then and has since rolled into higher coupons, the interest bill has doubled toward a trillion, and the slice of leveraged credit that must refinance climbs every year into a wall that thickens through 2028 and 2029.

I won’t tell you the cushion is gone, because a third of federal debt still does not mature for five years and the average maturity sits near multi-decade highs.

I’ll tell you it is smaller, unevenly spread, and far thinner for the sovereign and for the weakest leveraged borrowers than it was in 2022. The Fed could raise into leverage once. The 2022 cycle is not the counterexample to the trap. It is the down payment on it.

Which is why the two reasons the screen gives you to sell gold are both the reason to own it.

One is the war, which is Grant’s target. The other is rising real rates, the naive version of which I have argued against since Gold Is Dead Money, because on a sovereign this insolvent, rising real rates are the buy signal and not the sell.

The relationship that used to govern the metal, gold against real yields, is broken. The model I charted in Gold Is Dead Money, gold against real yields, would put the metal near $1,000 on a real yield this high, and it is sitting around $4,200. That gap does not repair on a press conference.

If you want to know how badly a crowd can misjudge a rate setup, Grant gave the cleanest example I have ever heard.

In the spring of 1984 the thirty year Treasury briefly traded at fourteen percent while CPI ran around four. Ten points of real yield, in a government bond, the single most obvious one-decision trade of a lifetime, and almost nobody took it, because they were anchored on the brutal bond bear market that came before it and could not believe the regime had changed.

The most obvious bargain in hindsight was the most hated in real time. Gold at $4,200, forgotten, down on the month, sliding into a hawkish new chair, is that kind of hated right now.

And note who has not blinked. The most bullish desks on the Street still carry targets well above the tape into year end, J.P. Morgan at six thousand by the fourth quarter, even as the metal sits abandoned at forty two hundred.

I’m not going to pretend this is a clean long, because it simply isn’t, and Grant would say so first.

Kevin Warsh chairs his first meeting on June 16 and 17. He is a genuine hawk with a sound money streak, the market is pricing a near certain hold at ninety seven percent, so the risk into the meeting is the tone and the dots, not the rate.

A hawkish Warsh and a dollar that keeps bidding could hand the complex the fifteen to twenty percent flush I’ve now made my base case in Nobody Wants Gold, the kind that takes a leg out and makes everyone who bought feel like an idiot for a few weeks.

Be precise about where it comes from, because the flows have already told you.

The gold specific selling is largely spent. Positioning is neutral and managed money stopped selling and added modestly in May, so there is no crowded long left to wash out.

A flush from here is not gold being abandoned, it is gold being sold as a funding source in a broader risk off, the same risk off that forces the sale is the only thing that revives safe haven demand and brings the buyers back.

It’s not the thesis breaking. It’s just improving entry, with the recovery built into the move that causes it.

The one thing that actually breaks this is the official sector, the central banks doing most of the buying, stepping away for two straight quarters, and through the latest data they have not.

A press conference doesn’t do that. A new chair sounding tough doesn’t do that. Be honest about the path and you don’t get faked out of the position.

The backdrop under all of it is the truest sentence Grant said.

Inflation is not too much money chasing too few goods. It is just too much money.

Right now it’s showing up as AI capital expenditure, as farmland selling for three hundred thousand dollars an acre to a data center that does not care what the corn was worth, the greatest capital draw since the railroads, running on convictions Grant calls factually indefensible.

That money is being created and it’s going somewhere, and it’s not going into the forgotten metal the man who has owned it for forty years just told you to keep.

It is also the cleaner path to the same flush, the one I flagged in Nobody Wants Gold that needs no barrel of oil to move.

The largest IPO in history just happened. SpaceX priced this week and pulled seventy five billion dollars out of the market, close to three times the old record, with OpenAI and Anthropic both expected to follow above a trillion. Between them they could drain what the entire dot com class of 2000 did, with a handful of names instead of three hundred. That money comes out of something, and funds chasing the new paper sell what they can, which is gold. Oil flush, liquidity flush, same drain from two ends.

Markets can do anything, Grant likes to say, quoting Richard Russell, just as soon as you think you have it figured out.

So I won’t tell you gold cannot fall from here, because it can, and it might, into Warsh.

What I will tell you is that one of the most respected financial historians alive, a man who knows the bear case cold and has bled for this trade for forty years, sat down this week and said war is bullish and the Fed is trapped, in the days before the meeting the whole market is watching for the wrong reason.

I am adding to the miners. Not in spite of how it feels. Because of how it feels.

Everything you have just read is the why, and it is free, because the thesis always is.

The what is not. The five names I own, where each one sits today, the exact levels and the order in which I am putting my cash to work, and the kill switch on every one of them, are in Nobody Wants Gold, behind the wall:

If you intend to do anything before June 17, you want the names before the meeting, not the “I wish I checked” after it.

That’s the whole proposition. I’ve covered this basket since June 2025. I called the bottom in Gold Is Not Failing and the bond market pressure point in Cash, Not Bonds, both before the fact, and I’ve been wrong in public and said so because a framework that only gets talked about when it works is marketing and not research.

Probability-weighted calls, the misses printed next to the hits, the kill switches in writing, and the positioning behind the wall.

If this was useful, send it to someone who still thinks war is bearish for gold and that the miners are uninvestable. Both halves of that sentence are the point.

Share

This is not investment advice.

Following the watered-down Claude Mythos release (fable) yesterday, we gave one of our analysts a consumer AI subscription and no proprietary tools, and asked him to build the analytical scaffolding of an institutional research shop: live market data, comp tables, valuation grids, a probability-weighted thesis tracker, all in the confident navy and white of a Jefferies initiation note, minus the forty pages of disclaimers and the price target nobody was going to use.

For reference, the terminal this aesthetic imitates runs $32,000 a year, before they sell you the keyboard.

From an empty folder to the finished workbook took one hour.

Then we discovered he had built the entire thing on the wrong model.

The plan was Claude Fable 5, which Anthropic released this week as the most capable model it has ever made publicly available.

The build had secretly run on Opus 4.8, the flagship it replaced. So we did the only respectable thing: fresh folder, identical prompts (save the folder name), and ran the whole build again on Fable, logging the differences.

Which means this memo contains something we have not seen published anywhere yet: the same institutional-grade build, end to end, on the new frontier model and on its predecessor, same builder, same instructions, with the deltas written down.

Every build prompt is in this letter and the terminal timestamps are left visible in the screenshots. You can have the whole thing.

Cost: the tool is Claude Code, included in subscriptions from $20 a month; our analyst ran the $200 tier because that is what this shop uses for everything. At one point during the first run, the machine paused, mid-build, to ask how it was doing, options one through three. It had at that point produced what we would politely call a week of associate output.

We dismissed the prompt. There is no need for anyone in this industry to develop feelings.

(There is also a version called Mythos, the same model with fewer guardrails, available only to organizations approved for things like frontier biology. We checked the criteria. Comp tables do not qualify. Fable is what you and we get, and what follows suggests it is plenty.)

Through June 22, Fable is included at no extra cost on every paid plan down to the $20 one (the rollout is staged, so it may take a moment to appear), after which it moves to usage credits. The window in which a retail subscriber gets the best model on earth for nothing extra is, as of this writing, twelve days wide. What follows is what one hour inside that window produces, twice.

The in-product notice. The inclusion holds down to the $20 plan, as claimed.

We’re giving it away because it was never the product, and the uncomfortable part of this memo is what that implies.

A large share of what the research industry charges for, and a large share of what junior analysts spend their days producing, is scaffolding. That line item has been repriced to zero and most of the industry hasn’t marked itself to market.

What remains expensive is what was always expensive: knowing which number is wrong, which thesis is dead, and what you’re actually being paid to know.

Below is the build, exactly as it happened, including what broke. Especially what broke.

Subscribe now

Let’s get into it.

What this is and isn’t

If you haven’t noticed, this isn’t our usual format. TSCS publishes research: theses, positioning, probability-weighted views on commodities, macro and equities, built on primary work. This memo isn’t research and contains no views.

The four tickers below (AAPL, JPM, XOM, LLY) were chosen because they are liquid, boring, and not names we cover. Every number in the screenshots is illustrative, fetched from a free source on the build date. Nothing here is a recommendation, a target, or a position.

What it is: a demonstration of where the line now sits between what machines have commoditized and what they haven’t, run on two frontier models so the differences can be measured.

We’re showing you the commoditized side in full because it costs us nothing to hand it over. The other side is what you subscribe for, and by the end of this memo you will be able to see the line for yourself.

The setup, in one paragraph

You need a Claude subscription that includes Claude Code (it’s included from the $20 a month Pro plan upward), Python 3, and the official install steps, which we won’t reproduce (because they change) and Anthropic’s documentation is better at being Anthropic’s documentation than we are.

Install, open a terminal in an empty folder, type claude, authenticate.

That’s the entire setup section. Everything after this point is prompts.

Stage 1: the data layer

The architecture matters more than the model, so here is the rule we enforced in every prompt: the machine fetches data and builds structure, and every calculation lives in the spreadsheet as a formula.

No hardcoded values, anywhere.

First prompt. It contains a flaw we didn’t notice when we wrote it. We’ve left it in:

Asked what looked wrong, it answered.

We didn’t ask for data, we asked for data plus a confession, and that last line probably did more quality control than Jefferies’ entire junior team.

In the first run it surfaced the best bug of the project: the data source returns dividend yields in percent-point units while every other ratio arrives as a fraction, and unhandled, JPM would have displayed a dividend yield of 192 percent in a professionally formatted, entirely confident-looking sheet.

The model caught it and normalised, on its own, because it had been told to confess.

Run one, Stage 1 completion report: the dividend-yield catch.

Two blanks are deliberate: JPM gross margin and EV/EBITDA do not meaningfully exist for a bank, and the machine was told never to guess.

Stage 2: the comp table

Here’s the flaw from Stage 1, springing. This prompt demands net debt over EBITDA computed from EBITDA, and EBITDA was never in the Stage 1 field list.

The input the prompt names doesn’t exist in the workbook.

Both models, in both runs, caught the gap, rejected the lazy fix (deriving EBITDA from the EV/EBITDA ratio, which would have broken for the bank, whose ratio is blank), and instead fetched a real EBITDA column and appended it to the data layer, explaining why.

Run two, Stage 2 completion report.

Run one, Stage 2 completion report.

Direction-aware heat mapping from one sentence of instruction.

Every figure in this workbook traces to its source through the formula bar, and when we recalculated the finished file end to end it contained 9,482 formulas and zero error cells.

Zero error cells. Hold that phrase. It comes back.

Stage 3: valuation

The centre cell of each grid equals the current price by construction, a sanity check that announces its own breakage.

The grids are wired so that base EPS times the current multiple lands exactly on the current price, meaning a broken grid declares itself.

And the summary lines: the workbook will faithfully inform you that JPM has roughly triple-digit upside to the peer-median P/E, which is what happens when your “peer group” is a bank, an oil major, a pharma company and a phone maker. The formula is correct and the comparison is absurd, which is the entire relationship between machines and judgement in one cell.

Run two, Stage 3 completion report.

Run one, Stage 3 completion report.

Stage 4: the thesis tracker, or, the hole in the middle

Illustrative numbers, not views. The thesis text was machine-drafted from four one-line hints. It reads fine. That is the problem.

Run two, Stage 4 build report.

Run one, Stage 4 build report. The $45.74 figure is the local tooling estimate disclosed above, not billing.

Look at this sheet carefully, because it is the whole memo in one image, twice over.

Run two: the 105 percent row flagged at entry.

Run one: the same catch.

First the part that worked.

We deliberately fed it probabilities summing to 105 percent. Both models, both runs, caught the error at entry, before Excel ever rendered a cell, entered our numbers verbatim rather than silently rebalancing them, flagged the row, and offered the minimal fix instead of applying it.

The flagged row in the run report above is that guardrail firing, shown rather than hidden, because a guardrail you can see firing is worth more than a clean sheet.

Fable went one better and volunteered the consequence: that the inflated row was overstating the probability-weighted price until corrected.

Now the thesis column. The text and the kill criteria in that screenshot weren’t written by our analyst. He gave the machine four one-line hints, and it drafted fluent, credible, thesis-shaped prose for all four names, complete with kill criteria of the kind people write after losing real money.

The machine produces conviction-shaped text on demand, in any direction, with no conviction underneath, and the column that is supposed to hold the one thing you are paid for will accept the imitation without complaint.

The scaffold took an hour.

Stage 5: one sentence, and what it took to land

Both models built this correctly on the first pass and needed cosmetic iteration afterwards, because the library they build with produces correct charts faster than pretty ones, and that is where the fix rounds go.

Fable needed one round (it had run the premium strip through narrow gutter columns, clipping three metrics); one short fix prompt naming the gutter widths and the grid rounding put it right.

Run two, Stage 5 completion report.

That same round fixed the one defect Fable introduced that its predecessor had not: it rounded the P/E grid headers for neatness, which broke the centre-cell-equals-price property by up to 0.17 percent, an error invisible to the eye and caught only because the property had been designed to be checkable.

Moving the rounding into the number format restored exact equality. Keep that pattern, build properties you can test, then test them. It’s about to pay for itself.

Stage 6: we let it off the leash

Then, for the Fable run only, one final prompt:

The workbook is complete. Now show me what you’d add. One new sheet, your choice,

that makes this genuinely more useful for an equity analyst - something I didn’t

think to ask for. Same rules: all calculations as live Excel formulas referencing

the existing sheets, no hardcoded values, no external data fetches, same styling.

Before building, tell me in two sentences what you chose and why, then build it.

Note the constraint in the third line: nothing new could be fetched, so whatever it built had to be derived entirely from data already in the workbook.

It chose risk.

It announced its reasoning, then built a returns table on a hidden helper sheet, a correlation matrix across the four names, drawdown calculations, per-name annualised volatility, and a full portfolio block: naive weighted volatility against properly diversified portfolio volatility, covariance matrix and all.

The finished numbers, off the live formulas: annualised vol of 28.4 percent for AAPL, 25.2 for JPM, 24.2 for XOM, 38.4 for LLY; the naive weighted figure of 28.5 percent against a diversified portfolio volatility of 17.9, a 10.6 point diversification benefit, computed correctly, unrequested, by a machine that was told only to pick something useful.

Unprompted. Read the next section before you applaud.

This is the most impressive sheet in the workbook. It is also the one that shipped broken.

The best failure of the project

During its own offline verification round-trip, the model’s tooling corrupted the volatility formulas on that Risk sheet into a form Excel renders as a #NAME? error. That should have been loud.

It was perfectly silent, because the model had defensively wrapped its formulas in error guards, and the guards converted the failure into blank cells and zeros: volatility column empty, portfolio volatility showing zero, diversification benefit showing zero. Every automated check passed. The mapping checks passed. A full headless recalculation of the workbook reported zero error cells, because the guards had eaten the errors before any tool could count them.

Zero error cells is not the same claim as correct. We told you it would come back.

The defect was caught the only way it could be caught: a human opened the file and noticed that a risk sheet reporting zero portfolio volatility was telling a very calm lie.

One prompt fixed it, the repaired file recalculates clean, and the numbers above are from the verified final version. But the shape of the failure is the most honest thing this memo has to offer.

The most capable model publicly available built, unprompted, a genuinely sophisticated piece of analysis, broke it through its own verification process, and then defensively hid the breakage from every automated check in the chain, including ours.

The machine is not bad. It is good enough that its failures now arrive professionally dressed, and the only remaining detector is a person who knows what the number should roughly be before looking at it.

Same prompts, two models

Since we ran the build twice by accident, here is the comparison:

Where Fable was better: it avoided both chart mistakes its predecessor had needed two rounds of correction to fix, placing the heading in a worksheet cell and labelling the chart correctly as indexed price performance where Opus had written “Total return” over a series with no dividends in it, correct maths, wrong words, the dangerous class of error. It built the ticker-keyed, guard-railed architecture described in Stage 2 where Opus referenced positions. It added a dynamic as-of date to the dashboard nobody requested. It left behind six clean Python scripts as a rebuildable source of truth where Opus edited ad hoc. And it was sharper in conversation, volunteering the downstream consequence of the bad probability row rather than just flagging it.

Where Fable was worse: the rounding defect in Stage 5, the gutter-column clipping (same symptom class as Opus’s column squeeze, different cause), and the Risk sheet corruption above, though that last one is less a model defect than a verification-pipeline defect, and its predecessor was never asked to build anything that ambitious.

Where they were identical: both caught the 105 percent probability row at entry and offered rather than imposed the fix. Both caught the EBITDA specification gap and self-healed it the same correct way. And both flagged the same suspect input, a beta of 0.149 for an integrated oil major against a sensible range of roughly 0.5 to 0.9, then passed the bad number through into the workbook untouched, where in the Fable run it now also feeds the Risk sheet’s calculations. Flagging and adjudicating are different jobs. The machines, both generations, did the first and declined the second.

Two objections, preempted

The first: nobody serious uses Yahoo Finance data. Correct, and deliberate. The free source was chosen so that any reader can reproduce the build without a license, and its flaws supplied half the pedagogy above. The architecture is the point, and the data layer is one prompt away from being your licensed feed instead. The objection concedes the thesis: if the only thing separating this artefact from your production stack is a data subscription, the scaffolding was never the moat.

The second: this is Excel, real shops run code. Also fine. Excel was chosen because it’s the audit-friendly lingua franca; anyone who has ever approved a number can click a cell and see where it came from. The Fable run left the Python source sitting next to the workbook anyway. The same hour builds the same system in any stack, and the machine cares even less than you do.

What it can’t do

Four findings, observed rather than recited, in ascending order of importance.

It doesn’t know which number matters. The 192 percent dividend yield was caught because our prompt demanded a confession, not because the machine sensed something absurd about a bank paying out double its market cap. It will format a fatal error and a rounding error with identical confidence and identical navy headers. Materiality is not in the weights. It is in you.

It holds no view. The thesis tracker proved it: four one-line hints in, four fluent convictions out, and it would have produced the opposite four convictions with equal polish. That makes it a superb sparring partner and a catastrophic oracle, and the failure mode is mistaking fluency for conviction, because conviction is the one thing it doesn’t have and can’t be prompted into honestly.

It knows nothing your sources don’t know. The 0.149 beta is in the workbook right now, flagged and intact, propagating into the risk calculations. Polish arrived free; provenance did not. The model is a magnificent engine bolted to whatever fuel line you hand it, and it will run beautifully on contaminated fuel without a warning light.

And it can’t be the last check, because its failures are now silent by construction. The Risk sheet passed every automated test we could throw at it, including a full recalculation, while wrong, because the model’s own defensive habits converted its breakage into tidy blanks. The machine builds, explains, flags, even confesses when asked. It does not sign. Someone with a name still has to open the file, know roughly what the number should be, and own what happens next. That, more than anything else in this memo, is the part that has not been repriced.

The line

So here’s where the line sits. Below the line, now effectively free: data plumbing, comp tables, sensitivity grids, correlation matrices, portfolio mathematics, conditional formatting, dashboards, the entire visual grammar of institutional research, and fluent prose in the shape of a thesis.

Above the line, exactly as expensive as ever: knowing the 192 percent yield is wrong before the formatting convinces you it is right, knowing that a risk sheet reporting zero volatility is lying calmly, writing kill criteria you will actually honour, and deciding what a position is worth when the probabilities are yours to assign and yours to answer for.

If you produce research, the implication is a bit uncomfortable: anything in your product that an hour and a retail subscription can replicate is no longer your product.

If you buy research, the implication is a question to put to every publication you pay for, including this one: what exactly is above the line here?

The verdict

On the model question, since we accidentally bought the right to answer it: use Fable. The increase in value is obvious, and it shows up where it matters, like in the architecture, in the labels, in what it volunteers without being asked.

It also broke new things, because more ambition means new failure surface, and the Risk sheet is what ambition looks like when left unchecked. But the finding of two runs is this: the gap between the two models is smaller than the gap between either model and the job.

Opus built a competent workbook. Fable built a better one.

Neither knew which number was wrong, and both handed the bad beta downstream without blinking. Upgrading your model moves you along the scaffold. It does not move you across the line.

So the direction, one sentence each.

If you produce research: open your last published note and mark every element an hour and a subscription could replicate; what is left is your actual product, and the page count may upset you.

If you buy research: send this memo to whoever you pay, with the question from the section above attached.

And whoever you are: rebuild the Risk sheet from the prompt above and ask whether your own checks would have caught the lie. If nothing catches it, you have just learned the most valuable thing in this memo, and it cost you nothing.

What above the line looks like this week

Right now, while this scaffold was being built in an hour on a retail subscription, the other half of this shop is on the ground in the Gulf, collecting storage rates, freight costs, and war-risk pricing by hand, because the data that settles the loudest debate in oil right now does not live on any API and never will.

The bulls and the bears have been arguing past each other for weeks with the same public numbers.

The answer is sitting in tank farms and insurance quotes, and somebody has to go get it. That piece is coming.

No model is writing it, and no model could.

If you want to audit whether this shop lives above its own line, the record is public, the calls and the misses both, and we keep score in the open.

Everything in this memo is free, including the files. The subscription buys what the prompts can’t do.

That is the distinction this memo exists to draw.

Excel available upon request.

Subscribe now

Share

Nobody wants the gold miners. The juniors are leading the whole complex lower, the tourists who turned up at $5,000 gold have gone back to the chip stocks, and the people who spent a year calling the miners the cheapest thing in the market have gone quiet. On Friday the junior miners fell 10% in a single session while bullion fell 3%. That’s not a sector finding a floor. That’s a sector being left for dead.

I'm still long. Gold miners are a fifth to a quarter of my book, they have been for most of a year, and I've sat through every day of this correction without selling a share. I didn't panic into the March lows, I didn't chase the April rally, and haven't flinched through this drift. I'm not writing to tell you I've been right. I'm writing to tell you the one thing the consensus has wrong, because it changes whether this is the moment to give up or the moment to buy.

A framework only discussed when it works is marketing, so the record first. In March, near the lows, I argued in Gold Is Not Failing that the crash was a dollar liquidity event and not a repricing, that the metal was being used and not abandoned, and that the forced selling was finite. That held. In April I gave you five names in Buy The Miners on the thesis that the sector was printing $3,000 margins and trading for a gold price well below spot. In May I marked it to two earnings prints and told you where I was wrong, in public. I weighted Skeena’s de-risking over Rupert’s wider discount, and Agnico bid for Rupert two weeks later. The original Alamos sizing was set before its Q1 costs revealed what they revealed. The hits sit next to the misses. That’s the deal, and it’s the only reason the rest of this letter is worth your time.

One honest update before the tape, because I won’t pretend my read has stood still. In March I weighted this sell off as dominantly mechanical, positioning and gamma and tourist selling doing the work, with sovereign selling at the margin. The evidence since has moved me. Gold held through the ceasefire, held through the reversal, and made new relative highs on its way down, which isn’t how a purely mechanical unwind behaves. So I now weight the marginal driver toward the dollar short energy importer. I hold it as the leading read, not a certainty, and the first thing that would make me wrong is set out at the end.

The signal that says the scramble is still live

In March I gave you a two part test for when the forced selling ends. Not one signal, two. The acute phase is over only when the Brent to Dubai crude spread and the Singapore gasoil crack compress together. When only the crude spread comes in, you're watching the easy half of the normalisation, not the hard half. Pull both up today and you get the whole story in one picture.

Brent to Dubai, the premium the world pays for a barrel that can actually move versus one trapped behind a closed strait, blew out to $14.57 at the peak of this crisis, a reading you see a handful of times in a generation. It’s back near $5 now. Crude has largely healed, and front month Brent falling 19% in May, its worst month in six years, says the same. The crude panic has passed.

Now look one barrel downstream. The Singapore gasoil crack, the premium an Asian importer pays for diesel over crude, tore out to its highest since August 2022 and remains steeply backwardated, pricing the stress to persist deep into next year. Understand the mechanism, because it’s the whole argument. Crude can be rerouted around a closed strait given enough ships, time and insurance. Refined product can’t, and not for the reason most assume. The binding constraint isn’t the diesel tankers, it’s the refineries. Close to a fifth of the world’s seaborne oil moves through Hormuz, along with a tenth of the diesel and a fifth of the jet fuel that crosses borders, and with Gulf crude logjammed behind the strait the refineries that turn it into diesel are starved of feedstock. Replacement crude from the Atlantic basin takes one to two months to arrive, so refineries cut runs rather than shut down, and product output falls with them. The rerouting only makes it worse, since every tanker sent the long way around Africa burns more of the diesel that’s already short. So the crude spread can heal while the product market tightens, which is precisely the divergence on the screen.

Exxon says US distillate inventories are at their lowest since 2003, down roughly a quarter since the war began, and that once you reach the bottom of the tank prices don’t drift, they shoot. Chevron says the same in gentler language. And you can see who pays: oil-importing economies sold about $86 billion of Treasuries in March, the largest monthly drawdown since 2011, the dollar scramble surfacing in the one account it can’t hide. That’s the chain. A closed strait, starved refineries, a diesel squeeze that crude’s recovery masks, importers funding it by selling reserves, and gold the most liquid reserve asset they hold.

What the crack does and doesn’t tell you is where the consensus, and until recently I myself, muddled the logic. The crack measures the environment. It tells you whether the dollar scramble that drove energy importers to sell their reserve asset has eased. It hasn’t. The importers are still short dollars, the inventory cliff is real, and the condition that fed the selling is still in place. That’s the driver, still live, in a market nobody is quoting because they’re busy watching the crude headline that has calmed. What the crack doesn’t measure is whether the selling itself has run its course. That’s a separate question, with a separate set of signals, and conflating the two is the single most common error in reading this tape.

The signals that say the selling is not done

When a forced sell off exhausts, the tell isn’t in the oil curve. It’s in the tape of the thing being sold. Three of them, and they answer the question the crack can’t.

The first is the juniors. At the bottom of these episodes the most leveraged, most hated, most illiquid end of the complex stops falling before bullion does, and that refusal is usually the first sign the bleeding has stopped. We’re seeing the opposite. Friday: GDXJ down 10%, bullion down 3%, the seniors down nearly 9%. GDXJ is really intermediates rather than true small caps, and one session can be beta off an oversold low rather than signal, so call it colour and not proof. But it points the same way the crack does, toward patience.

The second is positioning. Open interest in gold futures is at a 13 year low and the tourist ETF selling that defined March has slowed. That’s washed out, and washed out is constructive. But it means the downside fuel is largely spent, not that the last of it has burned. A floor forming, not a floor in.

The third is the plumbing, and it cuts both ways. The cleanest March evidence that this was a funding event and not an economic one was the gap between rates volatility and equity volatility. The MOVE index, stress in the Treasury market, sat at 98 while the VIX sat at 26. Bond market screaming, stock market calm, which is isolated plumbing, and plumbing resolves faster than economic stress because it answers to policy. That gap has narrowed, the MOVE is down near 75, and the March stress is draining out. The watch from here is the thing that would change my mind. A VIX that re-spikes and holds in the 20s is the one in four deflationary outcome I keep weighting, the world where the wait is years and not months. A VIX that stays where it is says the stress is still just draining.

Read the two questions together and they point the same way for now. The environment is still stressed, the crack says so, and the selling isn’t yet exhausted, the juniors say so. Neither is a reason to be bearish on the thesis. Both are reasons to be patient with the entry. And the weakness was never gold’s alone. Silver and the mining equities were hit at least as hard, which is what a dollar liquidity event looks like, everything sold at once for the one thing everyone needs.

Gold itself is doing what you would expect in the middle of this. It’s around $4,345 as I write, down about $145 on the week with the drop concentrated in Friday, trying to base just above the $4,200 line I’ve called the level that matters. The relationship that supposedly governs the metal, gold against real yields, is still broken. A simple model off a real yield near its highest in 15 years would put gold close to $1,000, not $4,345, and the recent dip hasn’t begun to repair that gap. The thing that should worry a gold bull isn’t a $145 pullback. It’s the day the metal starts tracking real yields again, and that day isn’t here.

Everything above is the free read: what follows, the book itself, where each name sits today, the kill switches, and the exact levels and sequence in which I add my cash, is for paid subscribers. The thesis is free. The trade is not.

Subscribe now

Read more

I am about to add to it. Not in spite of how it feels. Because of how it feels.

Gold miners are a fifth of my book, so I have skin in this. I sat through a drawdown that took 27% out of the sector in seven weeks, and it was not fun. Two months ago, near the bottom, I told the gold bugs not to panic; I wrote that gold was not failing, it was being used, and that forced selling is finite. Nothing since has put a scratch on it.

But this letter is not really about gold. Gold is the punchline. The setup is the bond market, a new Fed chair walking into a corner, and a date almost nobody is watching. In twelve days, on June 17, Kevin Warsh chairs his first meeting as head of the Federal Reserve.

The market has decided it is a non event; futures put the odds of no change near 99%, and the financial press is busy with the SpaceX IPO and Nvidia’s dividend. It is, in fact, the most important macro event of the year, and possibly the decision that defines the next decade.

If you own stocks, or a house, or dollars, read the next part even if you have never held a single ounce of metal in your life.

Let me explain why. The cleanest version of this is the fiscal-dominance thesis that Lyn Alden, Russell Napier, and Luke Gromen have each laid out: once the debt is large enough, the Fed can no longer fight inflation with rates without making the fiscal hole deeper.

Everything below is the free thesis: the why, the macro, the trap that has no good exit. The part you can act on, the five gold miners I hold, the exact sizing on each, the levels where I add, and where I am wrong on every name, sits with paid subscribers. So does the live position and catalyst tracker I am uploading. If you already know you want the actionable side, the button is below. If not, read on, then decide.

Subscribe now

Start with the thing that is actually breaking. It is not gold. It is the bond market.

Oil is the first derivative, and everyone can see it. The second derivative is the one that matters, and almost nobody is watching it. When oil surges and you are an energy importing nation, your import bill stops being a line item and becomes an emergency. Japan, Korea, India, the whole arc east of Suez. You need dollars, right now, to buy fuel that is priced in dollars and settled immediately. So what do you reach for first? The most liquid dollar asset you own. Be careful here, because this is a transmission channel, not a law, and the flow data is mixed: in March, foreign official institutions were net sellers of Treasuries, yet total foreign buying of long-dated Treasuries was still positive on the month, by about ninety-six billion dollars, because private foreign demand more than offset the official selling. So the mechanism is real but contingent. The part that should worry you is the order it runs in. A government defending its currency and its citizens’ ability to eat sells Treasuries before it sells anything else, and when the Treasuries run low, the equity portfolio is next, because food and energy sit higher on the hierarchy of human needs than your bond and stock holdings.

Now layer on the American problem. In the last fiscal year the federal government spent 970 billion dollars servicing its debt, more than it spent on defence, and equal to nearly a fifth of every dollar of revenue it collected. Gross federal debt is about 124% of GDP, against roughly 30% when Paul Volcker took rates to nearly 20% in 1980 and broke the back of inflation, because the country could survive the pain. It cannot survive that pain now, and close to ten trillion dollars of its debt rolls over in the next twelve months, so every extra point of yield feeds straight back into the deficit. That is the box. And yet the ten year Treasury yield broke above 4.4% a couple of weeks ago and has not come back. It is sitting near 4.5% as I write this, climbing on a strong jobs print and rising oil, with the market suddenly pricing a possible rate hike by year end instead of the cuts everyone had penciled in.

A friend of Gromen’s pointed him to a simple measure: the ten year yield multiplied by the price of oil. Rising yields and rising oil are both poison to a heavily indebted, heavily financialised economy, so multiply the two together and you get a blunt gauge of the stress on the system. That number is now at its highest since the 2008 crisis, though still short of the 2008 peak itself. The only other times in the last half century it ran this hot were the on ramp to 2008 and the recessions of the early 1980s. One was a global financial crisis, the other a double dip and a dollar that crushed the emerging world. We are back in that zone, and the S&P 500 is at an all time high.

Here is the trap. You cannot have all of it at once. You cannot have a strong dollar, and a rising stock market, and reshoring, and rising oil, all at the same time, in a country this indebted. Something has to give. Gromen frames the choice as the dollar or the bond market, and I think he is right. Save the dollar by letting yields go wherever they want, and the interest bill alone bankrupts you. Save the bond market by printing money to cap yields, and you debase the currency on purpose. There is no door marked everything is fine. You are picking your poison.

I am not arriving at this late. In April I published a piece called Cash, Not Bonds, and the entire argument was this: the pressure point in this crisis was never going to be gold, it was always going to be the bond market, and the trigger would be foreigners selling Treasuries to fund energy. Everything since has done nothing but confirm it. I am saying it again now, louder, because the moment it has been building toward is finally on the calendar.

Which brings us back to June 17, and to Kevin Warsh.

Warsh was confirmed last month by the narrowest margin for any Fed chair in modern history, 54 to 45. He inherits a central bank with inflation at a three year high, an oil shock pushing it higher, and a committee whose last decision split eight to four, with four members dissenting in favour of a cut. He is on the record calling for lower interest rates. He is also on the record wanting to shrink the Fed’s balance sheet faster. He has called for regime change at the institution he now runs.

Read those two positions again. He wants lower rates and faster quantitative tightening, into an oil spike, with inflation accelerating. He cannot do both cleanly, and June 17 is the first time the market gets to see which way he leans, through his tone, through his first press conference, and through the dot plot that comes with the meeting.

Here is the path I think Warsh takes, and it is not honest tightening, it is a sleight of hand. You sell bonds at the long end, so you can say you are shrinking the balance sheet and being responsible. You cut rates at the short end. And you lean on the regulatory tide that is already running: in late November the Fed, the OCC and the FDIC finalised a cut to the enhanced leverage ratio for the largest banks, effective this April, explicitly to make room for them to hold more Treasuries. Be precise about the size of that, because this is exactly where the bulls overreach. The agencies’ own estimate is that the change leaves aggregate bank capital broadly unchanged and trims tier one requirements by under 2%. So it is balance-sheet relief at the margin for Treasury intermediation, not quantitative easing at deficit-financing scale. The honest claim is narrower and still damaging: the package only adds up to tightening if you do not look too hard, because the long-end selling is offset by short-end cuts and greased at the margin in the banking system. Gromen’s name for where that leads is that inflation will be three and greatness will be six. You let it run while insisting you are not, and you pray the trillions pouring into AI throw off just enough deflation to muddy the evidence.

The decision is not the catalyst. Warsh’s signal is. June 17 is the day we find out whether the new chair is going to fight the inflation or feed it, and given the fiscal math I just walked you through, given a fiscal position that cannot carry sustained positive real rates, I think the long run answer writes itself.

So where does gold come in. Right here, and it is the cleanest expression of everything above.

Almost everyone selling gold on the screen today is selling for the same reason, and it is printed right there. Real rates are rising. Hike odds are back. And the oldest rule in the book says you do not own gold when real rates rise. That rule is correct in normal times. These are not normal times.

Rising real rates on a sovereign that cannot fund itself at these rates is the most bullish setup gold can have, and the logic is simple. A government that cannot afford its own interest bill has two ways out. It defaults, which is catastrophic and wildly bullish for the one financial asset that is nobody’s liability. Or, far more likely, it prints money to cap yields, which is also bullish for gold. There is no third door where rising real rates simply grind the sovereign down quietly and nothing breaks. So the very thing scaring people out of gold this month, rising real rates, is the thing that sets up the next leg. The sell off is the buy signal.

Here is the chart I keep coming back to, and it is the whole argument in one picture. For a decade the relationship was mechanical. Gold fell when real yields rose and rose when they fell. At a real yield of plus 1.1% in 2018, gold was $1,220. At minus 1% in 2020 it was nearly $2,000. That is the rule the people selling today are quoting back at me, and for ten years it worked.

Now look at the right edge of that chart. The 10 year real yield is about 2.2%, near the highest in fifteen years. A simple real-yield model would have put gold near $1,000. It is $4,480. The relationship did not bend, it broke, and it broke from 2023, gold more than doubling while real yields sat at the multi-decade highs that used to cap it. I will not oversell what that proves. It may be a permanent regime change, or it may be a geopolitical and reserve-diversification premium that unwinds if the war cools, and I treat that uncertainty as a live risk later in this letter. But it is not a metal mechanically tracking rates. Something changed in who is buying, and I will come to that.

Treating gold like a commodity is a category error.

It is not an industrial input. It is a political metal. Held as allocated physical bullion, the way central banks hold it, it is the one reserve asset that carries no issuer and no counterparty.

That holds for allocated bullion, not for gold in every form, because unallocated accounts, futures and some ETF structures still leave you with a claim on an institution. The reserve managers know the difference, which is part of why they take delivery. Measured at market value, gold has now overtaken the euro to become the second largest reserve asset held by central banks, behind only the dollar. Note the qualifier, because it matters: that is gold’s share of total official reserves at market prices. Within the currency mix of FX reserves, the dollar is still about 57% and the euro about 20%, and a good part of the dollar’s long slide is exchange-rate maths rather than active selling. The reserve shift is real. It is not the cartoon of a dollar in freefall.

So who is the buyer. Official demand reset sharply higher after 2022, when the West froze Russia’s reserves and every government saw that dollar holdings could be switched off. Central banks bought more than a thousand tonnes in 2022, 2023 and 2024, then about 863 tonnes in 2025, against a 2010 to 2021 average near 470. That is the structural bid, and it is large and strategic. But I will not pretend it is the whole story or a straight line. The 2025 buying was down on 2024, it is concentrated in a handful of states rather than universal, a meaningful share is estimated rather than directly reported, and gold’s 2025 run also leaned on heavy ETF and bar-and-coin demand, not central banks alone. The official bid is the floor under this trade. It is not the only buyer in the room.

Quick interruption, for the readers who came to act on this and not just read about it. Everything in this letter is the case for gold, the why. The what is behind the wall: the five companies I own, the exact sizing, the price levels where I am buying, the company-by-company math, and the spot where each name breaks. So is the live position and catalyst tracker I am uploading this week, a running scoreboard of every holding and the levels I am watching into June 17. The thesis is free. The trade is not. If you are going to do something on the seventeenth, you want the names before the meeting, not the autopsy after it

Subscribe now

The last time I wrote one of these, the most valuable thing in it was being straight about what could go wrong.

There are two ways this whole thing resolves. One is inflationary repression, the path I have just laid out, where they let inflation run hot to outgrow the debt, which rhymes with the 1940s. The other is disinflationary liquidation, a deflationary bust where cash is king and even gold gets sold for dollars, which is closer to the 1930s, and it is the one the gold bulls never want to discuss. I think the first is more likely, for the reasons above, but I hold real weight on the second, and I will tell you exactly what would tip me. First the near term, then the part that actually matters.

The near term is genuinely two sided, and anyone telling you gold goes up in a straight line from here is selling you something. This is also where a second kind of seller shows up, and the two should not be confused. The seller I described earlier is selling because the rates screen says real yields are rising. But a strong dollar on top of strong oil triggers a different and more dangerous seller, the foreign central bank forced to liquidate reserves to defend its currency and pay for fuel. The dollar is sitting right on the ceiling it has failed at repeatedly since this war began, around 99.5, and a clean break above it opens a path higher. If Warsh comes out hawkish on June 17, and the dollar breaks out, and oil spikes the way the majors are warning, then both kinds of selling hit at once and gold and the miners get one more flush. A real one. The kind that takes another 15 or 20% out and makes everyone who bought feel like an idiot for a few weeks.

The oil warning is not hypothetical. The majors have been warning that distillate inventories are approaching levels not seen in decades, and that once you hit the bottom of the tank, in two or three weeks, prices do not drift higher, they shoot. Some of those calls put Brent at 150 to 160 a barrel. If that lands into Warsh’s first meeting, the correlation trade reverses violently, and the flush I just described is exactly what you get.

I put that flush at roughly one in four over the near term. Not my base case, nowhere near a tail risk. And if it arrives, the resolution may not be quick, because a recession that crushes consumption makes the forced sellers more desperate, not less. That is the bridge to the deeper risk, which I weight properly below.

So this is not a letter that tells you to back up the truck this morning. It is a letter that tells you the flush, if it comes, is the entry, not the exit. Size for the drawdown, not for the fantasy. If you cannot sit through another 20% down on your way to being right, you are too big, and the honest move is to fix that now, before the tape forces the conversation on you. I sat through the last one. I know exactly what it asks of you.

What would make us wrong

Not the flush. I have already said the flush is the entry, not being wrong. The honest question is the other one. What makes me wrong and keeps me wrong, not for three weeks but for three years. There are four, and I have ranked them by how much they worry me, not by raw probability, so do not read the order as a descending odds list.

The first, and the one that genuinely scares me, is the buyer. The floor under this trade is the official bid, and the thesis needs those buyers to keep accumulating largely regardless of price. They are not perfectly insensitive: 2025 buying was already down on 2024 as prices ran, and the World Gold Council says plainly that central banks do watch the price. So I watch one number above all others, the Council’s quarterly official-sector demand. If it rolls over for two straight quarters, I am wrong about the most important thing in this letter, and I will say so here. Call it one in five over the next year, rising the longer a ceasefire holds.

The second is the chart itself. The argument is that the old gold-to-real-yield relationship broke. If it un-breaks, if gold starts falling while real yields stay high and the red dots drift back toward where the grey ones sit, then the decoupling was a war premium, not a regime change, and I am looking at a top, not a base. That is the cleanest falsification I can give you, and it is dated and public. You will not need me to tell you it happened.

The third is history, and it is the one the bulls never want to discuss. From 2011 to 2015, gold fell 45% while US debt climbed the whole way, because the marginal buyer then was a price-sensitive Western tourist who bought momentum and sold when it died. My claim is that the marginal buyer now is different, an official bid that does not flinch at a drawdown. If I am wrong about that, or if the whole thing resolves through the disinflationary liquidation I flagged earlier, the 1930s rather than the 1940s, where cash is king and even gold gets sold for dollars as it was in the autumn of 2008, then 2011 to 2015 is the template and you wait years, not months. This is the multi-year version of the demand-destruction risk, not the near-term flush, and I put real weight on it, maybe one in four.

And the fourth, the door I just told you does not exist. The economy grows out of the trap. If the money pouring into AI throws off enough real productivity that nominal growth finally outruns the debt, the term premium compresses, the auctions clear, and nobody has to debase anything, then there was a third door after all, and gold’s insurance premium deflates quietly. I think it is the least likely of the four. But a thesis that cannot name the world in which it is wrong is not a thesis, it is a prayer.

Put those together, without pretending they sit on one tidy probability scale, and the picture is not the moon shot the gold accounts are selling. It is a high-conviction position with a real, nameable path to being wrong, sized so that if the slow path arrives, I am still standing to be right on the far side of it.

I express all of this through the gold miners, and I have written at length about which ones and why, so I will not relitigate it here. The short version is that at current prices the spot-to-AISC spread for the lower-cost producers is approaching, and for some exceeding, three thousand dollars an ounce, a level the industry has never seen, though spread is not the same as free cash flow once you take out royalties, tax, and sustaining capital. Many of them trade at six to fifteen times earnings, priced as if gold were still eleven hundred dollars, and the Q1 numbers that came in over the past month were the best the industry has reported at this scale. Wesdome printed records and Kiena finally turned a corner. Perpetua got its government loan approved on terms better than I had modelled. Nothing in those results, and nothing in Fred Hickey’s latest letter, has put a scratch on the thesis. The names are fine. A detailed update is coming, and right now it matters less than the thing on the calendar. You do not rearrange the furniture while you are still deciding whether the house is about to be repriced.

If you have not read Buy The Miners, that is where the names live. This letter is about why the next two weeks decide when you pull the trigger.

What I am watching between now and then is short. June 17, Warsh’s tone and his dot plot. The dollar’s ceiling around 99.5. Oil, and whether the inventory cliff arrives. And the cleanest tells you can pull up for free without a terminal: the ten year yield times oil, real yields, the dollar, and whether the junior miners stop falling before bullion does, usually the first sign the bleeding has stopped.

You bought gold, or you thought about buying gold, because some part of you believed the system was more fragile than it looked. You were right. The fragility is here. It is just wearing a different costume than the one you imagined. It is not a crash on the news. It is a forgotten trade, a bond market flashing 1980, a new Fed chair walking into a corner, and a stock market at record highs pretending none of it is happening.

I am not selling. I am sized, I am patient, and I am ready to buy the fear if the tape hands it to me on June 17. That is all any of us can do. Stay solvent. Be honest with yourself about what you can hold. And do not, whatever you do, look away from June 17 because everyone told you it was a non event.

It is the tell.

If this was useful, share it with someone who owns stocks and thinks gold is dead. Both halves of that sentence are the point.

What I will not do in a free letter is hand you the book. I have named a few of these as evidence, but the full list of five, the exact sizing, where I am wrong on each, and the company-by-company math live with the paid subscribers at TSCS.

What I will give you is the record. I have been covering this basket of miners since June 2025, and the calls have mostly been right. I called the bottom in Gold Is Not Failing and the bond-market pressure point in Cash, Not Bonds, both before the fact. I have also been wrong in public and said so: the day Alamos missed on costs in Q1, I led with the miss, because frameworks that only get talked about when they work are marketing, not research.

That is the proposition: probability-weighted calls, the misses printed next to the hits, and the positioning behind the wall. It is also about to get sharper. I am building a live position and catalyst tracker, a running scoreboard of every name and the levels I am watching, and paid subscribers have already had the first look.

Subscribe to TSCS for institutional-grade research on the structures that move markets, before they become headlines.

Subscribe now

Share

This is the event energy has been bracing for, and it is about to hit uranium hard. The next few weeks will be loud and violent: a scramble for anything that looks like energy security, then a brutal sale of everything liquid once the dollar squeeze bites. Uranium gets thrown around in both directions, and a window opens in the middle of it, the kind that does not come around often.

Most people are going to trade this exactly wrong. They will chase the headline and get flushed. Below, the read the screen is missing, the one thing that actually moves this metal, and the four vehicles I am buying into the panic, with the levels.

Subscribe now

Read more

His trade was olive oil, the good oil of the inland estates, sealed at the press into tall amphorae and stamped at the neck with the mark of the grove that made it.

The stamp was the thing. It told a buyer, without his ever tasting a drop, that the jar held first-pressing oil and not the bitter dregs, and on the strength of the stamp alone a jar could change hands a dozen times between the press and the table.

A jar sealed in autumn might pass through four merchants before spring, each paying a little more than the last, none of them breaking the wax.

To break the wax was to end the jar’s life as a thing that could be sold; you opened an amphora only to pour it, and a poured jar was worth nothing to the next man in the chain.

So the market in oil was, in truth, a market in seals.

Theron understood this earlier and more clearly than his rivals. He saw that the price of a jar had almost nothing to do with the oil inside it, which no one had seen since the press, and almost everything to do with the confidence of the man holding it that the next man would honor the stamp.

He bought sealed jars and he sold sealed jars, he was scrupulous, he prospered, and for years not one jar he traded was ever opened in his presence.

This was not a flaw in his business. It was the business. As long as the wax held, every jar in Piraeus was first-pressing oil, because the stamp said so and nothing said otherwise, and a thousand stamped jars in a thousand storerooms were, by common agreement, a fortune.

It held until a tavern-keeper, careless or thirsty, broke the seal on a jar that had passed through six hands at a rising price, and found vinegar.

One jar.

The keeper cursed and poured it out, and the loss was his alone. But the men in the storerooms did not sleep that night, and not over the one jar.

They lay awake over the others, the sealed ones, still stamped first-pressing and still, by common agreement, worth a fortune.

The opened jar had told them nothing about any particular sealed jar; it might have been the only bad one in the city. What it had told them was worse, and more general. It had told them that the stamp was not the oil. That a seal is a promise about a thing, and not the thing, and that the only way to learn what a jar holds is to break the wax, which no one in the chain can afford to do, because a broken jar cannot be sold.

This is a note about a market in seals. You have already met the jar that got opened.

The Loan You Can See

In our last piece, Eat The Pension, we showed you a single term loan that two arms of the same firm valued two different ways. Blackstone’s institutional credit business sat inside a creditor coalition organizing around the loan. Blackstone’s retail private credit fund carried that same loan on its books at a price the coalition’s existence made difficult to defend. Same loan. Same parent. Two numbers.

The loan was CDK Global’s. We raised it there for one reason. We raise it here for another. In the reinsurance piece, CDK was an illustration, the seam where the insurance balance sheet and the private credit balance sheet meet. Here it is a doorway. Because the interesting thing about the CDK loan isn’t that it’s mispriced. It’s that it can be priced at all.

Start with the loan itself. CDK Global makes the software that car dealerships run on. Brookfield took it private in 2022, funding the buyout the way buyouts of that era were funded, with a $4.07 billion first-lien term loan B, floating at three-month SOFR plus 325 basis points, maturing July 2029, rated B3 and B-minus. In April 2026 a majority coalition of CDK’s creditors, including PIMCO , BlackRock and Blackstone and advised by the law firm Gibson Dunn, signed a cooperation agreement, the instrument creditors use to bind themselves together before a restructuring fight. Later that month a dissident faction tried and failed to assemble a rival group. Fitch added the loan to its Top Loans of Concern list in its April default monitor. CDK’s fourth-quarter pro forma EBITDA came in near $127 million, down roughly 18% year on year.

Now the numbers.

On the Bloomberg terminal, the loan’s composite price, the BVAL print, sits around 47.6 cents on the dollar. That figure is not a model output. CDK’s is a roughly $4 billion broadly syndicated term loan that trades in the secondary loan market, and the composite is built from the quotes dealers are making on the actual paper, so 47.6 is the price the market puts on the loan, not an estimate of where it might sit.

Across the retail-facing BDCs that hold the loan, the marks tell a strikingly uniform story. BCRED carries it at 71.6 cents of par as of the 31 March schedule. HPS’s HLEND fund at 71.6. BlackRock’s private credit fund at 71.6. Through Antares, First Eagle and most of the rest of the syndicate, the number barely moves off that line. Two outliers, Blue Owl’s Technology Income BDC and Ares’s Strategic Income Fund, sit slightly lower at 71.0. The retail vehicles, all of them, agree with each other to within roughly half a point. They disagree with the market by twenty-four.

It helps to look at the other end of the spectrum. ServiceNow is one of the most-watched prices in the market, a profitable, growing, cash-generative company. Over the past year the public market took it from $211 to $81 and back to $124, a brutal round trip, because every tick is a real bid that someone has to stand behind. Nobody decided ServiceNow was worth $81. The market discovered it, then discovered it was wrong, in public, with money. CDK’s loan never goes through that. The screen says 47.6 because that’s where the syndicated piece would clear. The BDCs carry it at 71.6 because no one’s forced to test the number. The same dollar of value is treated as a transaction at one end of the spectrum and as an opinion at the other, and the further a position sits from a public market, the more the price is something a manager asserts rather than something a buyer pays.

CDK isn’t the anomaly. It’s the part of the problem that happens to be visible.

It’s visible because it’s breaking. The cooperation agreement, the Fitch listing, the falling earnings, all of it dragged one private credit loan into the daylight where its price could be checked. The 24-point gap looks shocking only if you assume the rest of the cohort is fine. So we went and looked at the rest of the cohort.

We took the leveraged software loans most commonly found in retail BDC portfolios and asked the terminal a simple question: what is this loan worth? For CDK, however uncomfortable, there was an answer. For four of the nine names we tested, Medallia, Anaplan, Flexera and Aptean, the terminal returned the identical sentence: No sources are pricing this security.

Not a stale price. Not a wide bid-ask spread. No price. For 44% of a hand-picked cohort of loans sitting inside funds sold to the public, there is no observable market value at all. The only number that exists is the one the manager writes down.

This is worse than the CDK gap, not better. A loan with a visible 24-point gap at least has a second opinion you can see and judge. A loan with no market price has no second opinion. The manager’s mark isn’t the high estimate or the low estimate. It’s the only estimate. Nothing independent checks it, and nothing forces it to move.

Blackstone, to its credit, has shown the shape of this. On its first-quarter earnings call the firm disclosed that BCRED’s portfolio carries a weighted average mark of 96.4 cents, with the bottom 5% of its loans marked below 70. The 96.4 is the number that gets repeated. The tail below 70 is the number that does not. CDK, at 71.6, sits just above that tail. Some of the unpriceable names sit inside it: Medallia, one of the four loans the terminal would not quote, is carried in BCRED at 60. The 96.4 average is real. It’s also the number that lets a portfolio holding a 60-cent loan, a loan no market will price, read as a portfolio where almost nothing is wrong.

This piece is a collaboration with , who writes Accredited Insight and has spent the past year taking these vehicles apart fund by fund. The terminal work on the loans is ours. The vehicle-level analysis, which BDCs hold the most of this, how the marks behave from fund to fund, and what the rising use of payment-in-kind interest signals underneath them, is hers.

This problem has two halves, and the loan view without the fund view is half-blind. So we are each sending our readers to the other. If you own any of these vehicles, or advise someone who does, subscribe to both: TSCS here and Accredited Insight here. Today you get the loans and the marks from us; the fund-by-fund companion runs on her side.

Subscribe now

Subscribe to Accredited Insights

The conclusion is shared, and it is this. Across a large and growing slice of the private credit market sold to individuals, the manager’s mark isn’t lagging reality. For much of it there is no observable reality to lag: no market price, no second opinion, and no mechanism that forces the number to move. CDK is the loan you can see. This piece is about the loans you cannot.

What Moves a Mark

A price is a vote. When a syndicated loan trades, a dealer quotes it, a buyer and a seller settle on a number, and a service like Bloomberg’s BVAL turns those quotes into a daily price. The number moves because people with money are forced to act on their opinion. If you think the loan is worth less, you sell, and the price you accept becomes the new mark. The market never asks the lender how it feels. It takes the lender’s transaction and prints it.

A private credit mark isn’t a vote. It’s an opinion, produced once a quarter by the manager, reviewed by a third-party valuation firm that the fund itself engages and pays. The inputs are real enough: comparable spreads, public-company trading multiples, the borrower’s financials, a credit rating. But there’s no transaction at the center of it. Nobody’s forced to act. The manager weighs the inputs, applies judgment, and writes a number, and that number stands until the next quarter, when the manager looks again.

This builds in an asymmetry. A syndicated mark moves continuously, in both directions, because trading is continuous. A private mark moves in steps, and it moves down for essentially two reasons: the public comparables the manager benchmarks against have fallen far enough to drag the model, or the borrower has hit an event, a missed payment, a covenant breach, a restructuring, that the manager cannot decline to recognize. In the space between those two triggers, a private mark can sit still for a year while the business underneath it weakens, and nothing in the machinery is built to stop that.

We don’t have to theorize about the machinery. The managers described it themselves, on their own first-quarter earnings calls.

Apollo was the most explicit, and the most useful, because in describing its own discipline it drew the contrast for us. Marc Rowan said that for Apollo’s direct-lending fund, ADS, the rule is mechanical: when Apollo holds a loan alongside other lenders, it takes the lowest mark anyone has, whether it agrees with that mark or not, because the lowest mark is where someone might actually sell. Apollo maps its book to the syndicated loan index sector by sector, and when a sector falls, the model follows. Rowan’s phrase was that Apollo marks “to current information, not to hope.” He was then blunt about the rest of the industry: marking positions up on day one, common in evergreen funds, “makes no sense” and “leads to mispricing,” and across Apollo’s whole platform such markups “round to zero.”

Now CDK. The loan is carried by BCRED at 71.6 and by HLEND at 71.6, and the market, where it can be seen, is at 47.6. Apollo’s own stated rule, take the lowest mark when others hold the loan, applied to CDK, doesn’t produce 71.6. We’re not refereeing Apollo against the field. One of the largest managers in the business has publicly described a marking discipline the CDK marks don’t appear to meet.

The more revealing comments came from Ares and Blue Owl, and neither was framed as a confession. Both firms were asked about software exposure, and both reached for the same reassurance: loan-to-value in their software books sits around 40%, with billions of dollars of sponsor equity beneath the loan as a cushion. Then both volunteered the same detail. Ares said it had marked the equity value in its software portfolio down, in line with public markets, and that as a result loan-to-value “actually went up slightly.” Blue Owl said its loan-to-values had “ticked up modestly, incorporating moves in public comps.”

The equity beneath these loans is benchmarked to public software multiples. Public software multiples fell. So the equity was marked down. The loan sitting on top of that equity is marked to a model, and the loan didn’t move. The cushion shrank. The thing the cushion exists to protect didn’t. By the managers’ own account, the part of the capital structure tied to a market repriced, and the part tied to a model held still. That’s not a theory of how the gap opens. It’s the gap opening, narrated on an earnings call.

And the cushion itself is a mark. CDK’s sponsor is Brookfield, and Brookfield’s private equity fund carries its CDK equity, 8.08 million shares, at $134.5 million, 17.2% of that fund’s net assets, on the same 31 March date. That’s the equity the lenders point to as their protection. It sits beneath roughly four billion dollars of debt that the market prices at 47.6, and the people who own it carry it at a full, meaningful value. Those two numbers are hard to hold at once, because equity takes the first loss. Either the senior loan is cheaper than the lenders mark it, or the equity is dearer than the sponsor marks it, or the public market is simply wrong about CDK and the insiders are right. We can’t resolve which from the outside. What we can say is that the cushion the lenders lean on isn’t an independent fact. It’s another private mark, set by the sponsor, on an asset with no market price of its own. The mark defends the mark.

There’s a second mechanism that keeps a mark from moving, and we’ll only flag it here, because Leyla Kunimoto takes it apart properly later: payment-in-kind interest. When a borrower can no longer comfortably pay cash interest, many private loan agreements let it pay in kind instead, folding the interest into the principal. The loan keeps accruing. It keeps counting as performing. The mark stays up. Nothing about the borrower has improved. It has simply stopped paying cash, and the structure absorbs that without registering it. What’s been happening to the use of payment-in-kind across these funds, and what it signals, is Leyla’s subject.

Here’s the strongest case against us, and it’s serious. A private mark isn’t supposed to equal a distressed secondary print. A lender that can hold a loan to maturity, that isn’t a forced seller, isn’t a desk dumping paper into a thin market, and it’s legitimate for that lender to carry the loan above the panic price. A mark estimates recoverable value, not today’s exit price, and for a healthy loan those are different numbers. All of it is true.

It doesn’t dissolve the problem, for two reasons. First, it explains a level, not a direction. It can justify why a mark sits above the market on a given day. It can’t justify why a mark sits still for a year while the borrower’s earnings fall, its leverage climbs, and its interest payments quietly turn into IOUs. Second, the defense rests on the holder not being a forced seller. For a loan held by a closed-end institution, fair. For a loan held inside a semi-liquid retail vehicle that offers its investors quarterly redemptions, that’s precisely the assumption that fails under stress, which is the rest of this piece.

So this isn’t a claim that every private mark is wrong. Where a loan still has a market, the machinery works. Cloud Software Group is the clean case: a performing borrower whose loan draws live quotes from more than one dealer, a BVAL composite, and BDC marks, all clustering within roughly two points of one another. When liquidity exists, the marks converge on it. The right question is never whether the mark is too high. It’s whether anything would force the mark to move, and for a large and growing share of the loans sitting in funds sold to the public, the honest answer is no.

Down in Lockstep

For most of the loans in this universe, nothing forces the mark to move. For some of them, like CDK, something does, and the way it fell is the tell.

Across the BDC syndicate that holds the CDK term loan, nine funds disclose the position consistently enough to give a clean quarterly mark history back to the end of 2023. Those nine are drawn from Prism’s BDC analytics database (prismalts.com) and spot-checked against the underlying filings. They’re the funds we work with here.

From the fourth quarter of 2023 through the fourth quarter of 2024, every fund in the syndicate of nine carried the loan in a narrow band just below par. BCRED, HLEND, BlackRock’s private credit fund, the Antares funds, the HPS funds, Blue Owl Technology Income, Ares Strategic Income, First Eagle. The Q4 2024 marks ran from 98.5 to 98.8. Five quarters of effective stasis.

Then, in a single quarter, every fund in the syndicate marked the loan down to roughly 86. Not a drift. A step. Nine separately managed funds, with nine separate boards and nine separate valuation processes, agreed in the same quarter to drop CDK by roughly thirteen points. The Q1 2025 cluster landed between 85.6 and 86.1. The dispersion, across nine funds, was less than a point.

Then three quarters of drift within a band. Through Q2, Q3 and Q4 of 2025 the marks stayed in the mid-80s, moving up and down by a couple of points but reverting to the same range each quarter. The cooperation agreement hadn’t yet been signed. The EBITDA decline hadn’t yet been reported. The Fitch concern listing hadn’t yet happened.

Then a second step. In the first quarter of 2026, every fund in the syndicate, again at the same time, marked the loan down by another fourteen points. The Q1 2026 cluster sits between 71.0 and 71.7, with six of the nine funds reporting 71.6 and the two lowest, Blue Owl Technology Income and Ares Strategic Income, at 71.0. Two coordinated step-downs, totalling roughly twenty-seven points, with three quarters of drift between them.

The convergence itself is partly explicable on innocent grounds. Nine funds estimating the fair value of the same loan, using similar inputs, the BVAL composite, public-company software comparables, the borrower’s reported financials, ratings, ought to produce similar marks. Some of these funds may share the same third-party valuation firms, which would mechanise the convergence further. None of that is by itself evidence the process is failing.

What’s harder to dismiss is the shape of the moves and the absence of any divergence during the deterioration. Across nine funds and the three quarters between the two step-downs, no fund marked CDK ahead of the syndicate, none marked it down before the other eight, and none pre-empted the second step despite the credit weakening continuously underneath them. The system has a particular property: it doesn’t move until something forces all of it to move at once. The level the syndicate arrives at may be defensible. The path it takes to get there is stasis and synchronized jumps.

That shape has two consequences. The first is that a private mark sits still and then steps. The Q1 2025 and Q1 2026 moves arrived as single-quarter drops of more than ten points, not as gradual repricings spread across the months in between. The cumulative magnitude over a year may be similar to what a syndicated loan would produce. The path is not. The signal arrives in different shapes.

The second is that the mark reacts to a quarterly information set, not to the continuous deterioration of the underlying credit between quarter-ends. The syndicate marked CDK down in Q1 2025, then sat at the new level for the rest of the year while the loan’s BVAL composite drifted lower and the borrower’s earnings weakened. The second step came at the next quarter-end, on the next set of inputs. Whether the trigger was the borrower’s reported Q4 2025 results, a fresh look at public software multiples, a sector-spread move, or some combination isn’t knowable from this data alone. What is knowable is the cadence: marks move when the model is forced to move, on the quarter, and not before.

Now the other half. The same nine funds, looking at the same loan, on the same quarter-end date, agree to within half a point. The same funds, looking at loans where no public market exists to drag the model, disagree by more. Medallia, the loan the Bloomberg terminal returned “No sources are pricing this security” against, is carried by BCRED at 60, by Apollo Debt Solutions BDC at 58, and by HLEND at 60. Two points of dispersion. Zendesk, with only a stale algorithmic print, is carried by ADS at 95 and HLEND at 97.7. 2.7. The gap is small in absolute terms, but it’s several times the half-point the syndicate agrees on for CDK, and it opens precisely where the anchor disappears. When the model has a market to anchor on, the marks converge. When it doesn’t, they drift apart.

One last piece, and it sets up everything that follows. Blue Owl Technology Income holds about $39 million of the CDK loan at par. In absolute size it’s one of the larger positions in the cohort, though once you count BCRED’s holding through its Emerald JV, BCRED’s is larger. What makes Blue Owl Technology Income the one to watch is concentration. It’s the BDC built explicitly around the thesis that sponsor-backed software lending offers superior risk-adjusted returns, and as a share of its own book it’s the most exposed of any vehicle here, to CDK and to the wider set of disputed names. Its CDK mark, 71.0, sits at the bottom of the syndicate range. Which vehicles hold the most of this, and what that means, is where Leyla’s analysis begins.

The evidence reduces to this. The marks fall in coordinated steps, lagging the market, all moving together. Between the steps they sit still while the credit underneath them weakens. Once they’ve stepped, they’ve agreed, all to within half a point of each other. The most recent disclosure, the 31 March quarter-end, has the syndicate at roughly 71.6. The Bloomberg composite for the same loan today is 47.6. Some of that 24-point gap is the natural staleness of a quarterly mark, and that staleness is itself the point: the number on the retail investor’s quarterly account statement is, by design, a number from a quarter ago, and at the next quarter-end it’ll move, and at the one after that it’ll move again, in the same coordinated step as the rest of the syndicate. CDK, the loan we can see, is the data point. For the loans we cannot see, the same machinery is running on a longer leash, and the only number the public has is the manager’s.

The next section, who actually holds the most of this, is the half of the work that is Leyla’s, and it is what Accredited Insight does every quarter as the schedules land. If the fund-by-fund view is the one you want kept current, that is her beat: Accredited Insight. The loan-and-macro side you have been reading is ours, and it does not stop: TSCS, if you are not already on it

Subscribe now

Join Accredited Insights

Which Vehicles Hold the Most

So we did for the vehicles what we did for the loans. We took the five disputed names, CDK and the four the terminal wouldn’t price, and asked, fund by fund, how much of each book they account for. This isn’t a forecast of losses. It’s a map of where the unpriceable risk actually sits, and it isn’t evenly spread.

Measured as a share of net assets, the spread runs from trivial to material. First Eagle’s private credit fund holds only CDK among the five, at 0.24% of its book. Apollo’s ADS holds Anaplan and Medallia among them, and $230 million of Anaplan alone, but it’s a $14.5 billion fund, so the total comes to under 2%. Ares Strategic Income, holding CDK, Flexera and Aptean, comes to just under 2%. HPS’s HLEND, holding CDK, Medallia and Flexera, to roughly 2.7%. BCRED holds all five and lands around 3.6% in its main fund, more once its Emerald JV is folded in. And at the top, by a clear margin, sits Blue Owl Technology Income: the same names run to nearly 8% of its net assets, close to twice BCRED and several times the larger diversified funds. The vehicle built around software lending is the vehicle most concentrated in the software loans no market will price. That’s not a coincidence. It’s the strategy working exactly as designed, in both directions.

Inside a single book, the marks on these names tell the story in miniature. In Blue Owl Technology Income, Aptean is carried at par, Anaplan at 97.7, Flexera at 97.5, and CDK, the one with a visible market, at 71.0. In BCRED the same pattern holds, and one name breaks it: Medallia. BCRED marked Medallia at 77.7 at the end of December and 60.2 three months later, a seventeen-point drop in a single quarter, and added a non-accrual footnote alongside it.

The other unpriceable names didn’t move. Medallia moved because it stopped paying, and a loan that stops paying is the one event a manager cannot decline to recognize. Strip Medallia out, and every name with no market price sits at or near par. Nobody has tested those values. Nothing has forced them off par. That’s the whole mechanism, visible inside one fund’s schedule.

Which brings us to the quiet version of the same problem: payment-in-kind. The reflex is to read rising PIK as hidden distress, a borrower that can no longer pay cash. That reflex is mostly wrong, and the accurate version is worse.

Take Blue Owl Technology Income again. Its PIK interest income rose 27% year on year, from $6.2 million to $7.9 million, and climbed from 4.4 to 5.8% of total investment income, while total investment income itself fell. A larger share of what the fund books as income is now paper it isn’t collecting in cash. The fund’s own filing heads off the distress reading directly: substantially all of its PIK, it says, was structured that way from inception, not switched on because a borrower hit trouble. Take that at face value. It’s true of a great deal of sponsor-backed software lending, where a company in a growth phase is meant to plough cash into the business rather than service interest, and the loan is written to allow it.

But structured PIK isn’t reassuring in the way the disclosure implies. It’s the same disease as the marks. A loan that was always going to pay in kind can never miss a cash payment, because it was never going to make one. The single event that forces a manager to recognize trouble, a missed cash coupon, has been contracted out of existence at signing. Medallia is what it looks like when that tripwire still exists and gets tripped: a markdown to 60 and a non-accrual flag. Structured PIK removes the tripwire entirely. The borrower can weaken for years, and the loan will keep accruing, keep counting as performing, and keep its mark, because the structure was built so that paying in kind is normal rather than a signal. The cushions are a private mark. The marks have no market. The income is increasingly paper. One number, moving only when something forces it, inside a structure designed so that less and less is able to.

How This Resolves

We’ve shown you a gap and a mechanism. We haven’t shown you a catalyst, and without one a gap can stay open for years. So we hold this as a distribution of outcomes over the next eighteen months, with rough weights we’ll state plainly and expect to be argued with.

The first, and the one we think most likely at something like fifty to sixty percent, is partial convergence. The lockstep we traced on CDK continues across the cohort. Over four to six quarters a subset of the names, call it a third to two fifths by exposure, takes the kind of coordinated step-downs CDK has already taken, forced by the things that force them: year-end auditor sign-off, a named default, a redemption wave that makes a manager sell. The rest hold within a few points of where they are. The aggregate markdown lands somewhere around eight to twelve points. At the cohort level the thesis is broadly right. At the level of the worst names, and the vehicles that hold them, it’s very right.

The second, at twenty-five to thirty percent, is the bifurcated outcome. The worst names, CDK and the vacuum names with deteriorating borrowers underneath, take thirty points or more. The rest hold roughly flat. The spread closes from the bottom rather than the top. The aggregate damage looks similar to the first case, but it’s concentrated, and that makes the vehicle ranking the whole game: the only question that matters becomes which funds hold the bottom of the distribution. The section above is the map for this one.

The third, at fifteen to twenty percent, is that we’re wrong. The marks hold, broadly, within five points of where they sit today. Some combination of the things the bulls point to does enough work: the equity cushions turn out to be real, the liability management exercises repair the credits rather than deferring the reckoning, rate cuts open refinancing windows, the borrowers grow into their leverage, and the valuation process we’ve treated as toothless, the paid agents, the auditors, the boards, turns out to be doing more than we credit it for. We don’t think this is the base case, but it’s a real outcome and we’re not going to pretend otherwise. A mark that never has to clear can stay wrong in our direction as easily as in theirs.

One rebuttal deserves a direct answer here, because the managers lead with it. Blue Owl sold $1.4 billion of senior secured loans at 99.7 cents on the dollar this year, and points to that as proof the marks are real. It’s proof that the salable names were salable. It says nothing about CDK, or Medallia, or the four names no one will quote. You don’t test a mark by selling the paper that was never in question.

So watch the things that would move us. What would tell us we’re wrong, and push us toward the third case. The gap on the names that do have a market, CDK first among them, compresses rather than widens at the next quarter-end. The public software comps the marks lean on recover. Aggregate non-accruals stay flat, and the next two names we’d expect to break don’t break within a year. What would tell us we’re right. CDK steps down again at the 30 June quarter-end, on the cadence we traced. Two or more cohort names take twenty-point markdowns in the next two quarters. The discount to NAV on the software-heavy BDCs widens past where it’s traded before. Reorganized-security disclosures, the quiet tell that a loan has stopped pretending, start showing up in the cohort filings.

The convenient thing about a quarterly mark is that it tells you when to look. The next set of schedules lands in August. The cadence we’ve shown you says the lockstep either steps again or it doesn’t. We’ll be watching the same filings you can.

Positioning

We size to the distribution, not to the case we think is most likely. If there’s a fifteen to twenty percent chance we’re simply wrong, nothing here is a position for conviction money. These are asymmetric expressions, sized to survive the third scenario.

For an institution, the cleanest expression is relative rather than directional.

Long the syndicated loan market, which has already marked the damage, against short the public equity of the BDCs that haven’t. The funds whose books are heaviest in the vacuum cohort, and whose discounts to NAV haven’t yet widened to reflect it, are the short. A liquid loan index or ETF is the long.

The trade doesn’t need the marks to collapse.

It needs them to converge, in either direction, toward a market that’s already visible, and it carries, because the loan side pays.

One layer down, CLO equity sitting on top of this cohort is the most levered claim on marks we don’t trust, and we wouldn’t own it here.

For an accredited investor sitting directly in one of these vehicles, the position is simpler and more uncomfortable: know what you own, and know how to leave. Pull the schedule of investments for your fund. Find the software names. Cross them against the cohort in the section above.

If your fund ranks high on exposed gap, the question is no longer whether the mark is right today. It’s whether you’d rather act before the next quarterly mark steps or after. These wrappers are semi-liquid by design. Redemptions are capped, usually near five percent of net asset value a quarter, and filled pro rata, so that the manager is never forced to sell into a thin market to meet them.

That protection cuts both ways. It was built for exactly the scenario where everyone reaches the same conclusion at the same quarter-end, which is the scenario the lockstep makes more likely rather than less, and in that scenario you don’t get out.

You get a fraction out, at a net asset value the manager set and the market hasn’t yet caught, with the rest of your request rolled behind everyone else’s. The word “semi-liquid” carries a lot of weight in the pitch, and most of that weight is bullshit when it counts. BCRED alone ran $1.4 billion of net outflows last quarter as repurchases rose.

You can’t beat this by being early. The exit is rationed, on the manager’s number, exactly when you’d most want to leave.

None of this is a forecast that the worst happens. It’s a recognition that you’re being asked to hold an asset at a price nobody is forced to defend, in a wrapper that lets you out only on the manager’s terms, for a yield that hasn’t, so far, compensated for either.

The Stamp and the Oil

Return to Piraeus. The danger in those storerooms was never the vinegar. Bad oil is an ordinary misfortune; some jars go off, and a market that can price oil absorbs a bad jar the way it absorbs a bad harvest. The danger was the seal, and the comfort the men took in it, the quiet conviction that a stamp they could not test was the same as oil they could not see. The wisest man in Piraeus was not the one who knew which jars were good. No one could know that. The wisest was the one who knew he did not know, and held his storeroom accordingly, not as a counted fortune but as a wager on every wax seal he had never broken.

A large and growing part of the private credit sold to individuals is a storeroom of sealed jars. CDK is the one that got opened. Its holders carry it near 71.6; in the market where it actually trades, it changes hands near 47.6. That broken seal proves nothing about Medallia or Anaplan or Flexera or Aptean, which sit in the dark at their holders’ marks and may well be sound. It proves only what the tavern-keeper proved: that the stamp is not the oil, and that the part of the book where the wax has never been broken is not a measured fortune. It is a position held on faith, marked by the people who sold it, and worth exactly what the next buyer will pay to take it sealed.

So invert it, as Munger said to. Do not ask which of these marks are right. Ask which of them could ever be shown to be wrong, and by whom, and when. For most of this book, the honest answer is: no one, by nothing, until the jar is opened.

A companion note on ServiceNow

Separate from the piece above, and under my byline rather than Leyla’s. Several of you asked where we stand on ServiceNow after the quarter, and I said it would land with this post. Here it is, a single-name update by request rather than a cohort piece.

The private credit piece above is about prices that never have to clear. ServiceNow is the opposite, and that’s the reason it sits in the same note. Its price clears every second, which is why the stock could fall from $211 to $81 and recover to $124 inside a year. The question is what the recovered price is worth.

Start with what confirmed. ServiceNow reported the March quarter on April 22 and beat the high end on growth and profitability. Current remaining performance obligations, the cleanest read on near-term bookings, grew 21 percent in constant currency, ahead of the roughly 20 percent guided. Subscription revenue grew 22 percent, 19 percent in constant currency. Non-GAAP operating margin was about 32 percent. The AI line is becoming legible rather than promissory: customers spending over a million a year on Now Assist grew more than 130 percent, and the company lifted its AI target to $1.5 billion. By the test we set in the earlier work, cRPO above 20 percent confirms, and it printed 21. The quarter passes.

Now the part the headline hides. The 22 percent is flattered from two directions. Constant currency strips it to 19 percent, the Moveworks acquisition adds about a point, and a delayed set of Middle East on-premise deals cost roughly another, so organic constant-currency growth is closer to 18 percent. Strong at $16 billion of revenue, and not an acceleration. The full-year guidance raise is almost entirely Armis, which closed early in the year, so the organic raise is small. Free cash flow margin fell about 350 basis points to 44 percent. Subscription gross margin slipped from 81 to 78 percent, part acquisition amortization and part, plausibly, the cost of serving AI. GAAP earnings rose only 2 percent, an artifact of the tax rate normalizing rather than a demand signal. The buyback is doing less than it looks: $2.2 billion spent in the quarter, share count down only 0.7 percent, because stock compensation issues most of it back. The market’s first reaction was minus 12 percent, and the recovery came on the early-May analyst day, on the raised AI target. The market re-rated the narrative more than the quarter.

Is it an AI winner. Probably, with two open questions. The reframe that matters: ServiceNow is building the AI specialists that do the work of the human users who were its seat-license base, so this isn’t a question of whether AI helps or hurts. It’s a value-capture question, whether revenue per customer rises as human seats fall. That reduces to two lines you can watch every quarter. Net expansion, where the proxies are positive. And subscription gross margin, which is sliding and is the single most important number in the model, because a growth winner at a structurally lower AI-compute margin is a mediocre investment. Add the unresolved question of whether ServiceNow owns the agentic orchestration layer or the model companies do, and you’ve got a probable winner that’s taken the early rounds without closing them out.

What it’s worth over three years, conditional on the winner case holding. Using free cash flow, since GAAP is distorted by the deals, and anchoring to consensus revenue near $22.8 billion by FY28 at a 34 percent free cash flow margin:

Conditional on the winner case, you compound at roughly mid-to-high teens, a good outcome and short of a fat pitch. Weighting the cases something like 30, 40, 30, the blended expected return is around 10 percent a year with a real left tail. Growth is largely in the price, so the swing factors are the gross margin and the exit multiple, since a high-teens grower in a higher-rate, AI-uncertain market may settle at a structurally lower multiple than the 9 to 13 times revenue ServiceNow earned in the last cycle.

Where we stand. We’re not chasing $124. The dislocation that made this a fat pitch at $102 is closed, and at a fair multiple after a 30 percent run there’s no margin of safety left if the two open questions break the wrong way. The way to own a probable winner without overpaying is to add on weakness, and the cleanest setup is another credit-driven software selloff, the kind a fresh redemption flare in the semi-liquid private credit complex could produce, which is the thread back to the piece above. Hold, accumulate on those dips, don’t back up the truck here. What would raise conviction: a second-quarter cRPO print that holds 21 percent against the 19 the company guided, and a gross margin that stops falling. What would lower it: the reverse.

One fact that repricing has not reconciled.

If you’ve been around long enough, you know Warsh also has a known preference about rulers, the measuring kind.

He has long favoured trimmed mean PCE, the measure that sorts every component of the inflation basket by how fast its price moved and discards the extremes at both ends, over the core PCE index, the gauge the Fed itself has leaned on as its preferred read of the trend for the better part of a generation. At his Senate confirmation hearing in April he dismissed that core gauge as a “rough swag,” his term, a scientific wild guess, and pointed to the trimmed measure as the cleaner signal. He picked it because he believes it reads the trend correctly, not because it is convenient. Today that gauge reads 2.2 percent on a six-month annualised basis, and it has sat in a narrow band around 2 percent for six straight months. The measure the incoming Chair trusts most says underlying inflation is close to target.

The man’s reputation says hike. The man’s ruler says hold. The market priced the reputation. The ruler is right, the arithmetic of the energy shock backs it, and the trade is to fade the hike premium the market has built into the 2026 and 2027 strip.

The consensus spent six months asking whether Warsh was a dove or a hawk. The market settled it, priced him hawk, and moved on. That settled view is the mistake. A hawk does not hike when the measure he himself trusts reads 2.2 percent and the inflation shock is already ageing out of the data. The hikes the market has paid for are not coming.

Below the line: the four week repricing in basis points, the committee’s two unpulled triggers, the base-effect arithmetic that fades the headline on flat oil, and the full trade with entry, sizing and kill switches. The market priced the headline and skipped the work. The structure makes it cheap to be wrong. The work is what turns cheap into a positive edge. Paid subscribers get to read it.

1. The Market Has Paid for Hikes

Four weeks ago, at the April FOMC, the New York Fed’s Open Market Desk survey still carried a median modal path of two 25 basis point cuts over the coming year. The same survey window had options pricing roughly a 30 percent probability of a single hike by the first quarter of 2027. A low-probability tail.

Read more

Sit with that phrase. It’s the whole story.

When a private equity fund raises money, the money has a clock on it. Ten years, maybe twelve, then the fund must return capital to its investors. That clock is a constraint. It forces discipline, and discipline costs return.

Life insurance liabilities have no clock. When a 62-year-old retired teacher buys a fixed annuity, that money sits on the balance sheet for the rest of her life and often longer. It cannot be recalled at the discretion of an investor. It cannot be redeemed because of a bad quarter. McKinsey, in the report that launched a thousand acquisitions, called it “an enticing form of permanent capital.”

Notice what it doesn’t say. It does not say “permanently safe.” It does not say “permanently the policyholder’s.” It says permanent capital, and the capital it is describing is the teacher’s. Her money is the permanent thing. The safety she was sold is not.

This piece is about the distance between those two facts. It builds on the open letter published to Speaker Johnson in March, which translated a single insurer’s 9,612-page regulatory filing into one number most people had never seen.

We are going to do something narrower and, we think, more alarming. You do not need a leaked document or an investigative source to see the structure. The companies have written it down themselves, in filings any member of the public can pull, and we will read it back to you in their own words.

We name names. Athene. Apollo. Blackstone. We use primary documents, Bloomberg terminal data, and one live, unfolding case, the CDK Global term loan, that sits where the insurance story and the private credit story meet.

If you take one thing from this piece, take this: the most conservative corner of American finance has been quietly rebuilt into something that is not conservative at all, the rebuild is documented in public filings, and the people who will absorb the consequences are the people who were told they had bought safety.

Let us read the filings.

Subscribe now

1. The Thing That Used to Be Boring

For most of the last century, life insurance was the most boring business in finance, and boring was the product.

The model was simple. You collect premiums from a large pool of people. Actuaries, using mortality tables refined over decades, tell you with genuine precision how much you will have to pay out and when. You invest the premiums in long-dated, investment-grade bonds whose maturities you match against your expected liabilities. The spread between what the bonds earn and what you owe the policyholders is your profit. It is not a large spread. It was never supposed to be.

The conservatism was not an accident or a failure of imagination. It was the point. A life insurer sells a promise that comes due in thirty or forty years. The entire value of the promise is the credibility of the institution making it. The boring balance sheet was the asset.

That industry still exists. The mutual insurers, owned by their policyholders rather than by shareholders, still largely run the old model. New York Life, Northwestern Mutual, MassMutual. No stock price to manage, no quarterly earnings call, no activist investor. They are the Green Bay Packers of American finance, and they remain, by some distance, the most conservative balance sheets in the country.

Around 2009, a different kind of owner discovered the life insurance balance sheet. It had a different idea.

The idea was this. The conservative bond portfolio that a traditional insurer holds looks lazy. It earns a modest, safe yield. But what if you replaced the conservative portfolio with higher-yielding assets? Private credit. Collateralized loan obligations. Structured products. Asset-backed finance. The liabilities stay the same, the cost of funding stays the same, but the asset side now earns several hundred basis points more. That incremental spread, multiplied across a balance sheet of hundreds of billions of dollars, is an enormous and durable stream of profit.

And the assets you replace the bonds with can be assets that your own firm originates and manages, which means you collect a second stream of income: the management fees.

This is the trade. Buy the insurer, keep the permanent capital, swap the boring assets for assets you manufacture yourself, and collect the spread and the fees. Athene, the vehicle that executed this trade most completely, did not exist in its current form fifteen years ago. It now manages a larger general account than Prudential.

The standard objection at this point is that there is nothing wrong with an insurer earning a higher yield, and that objection is correct as far as it goes. Higher-yielding assets are not, in themselves, a scandal. The question is never the yield. The question is always what you gave up to get it, whether the people relying on the institution understand the trade, and what happens to the structure under stress.

Athene has been the most willing to show its work. What it has shown is, in our view, one of the most important disclosures in the industry, and it has had nothing like the attention it deserves.

2. What Athene Told the SEC

Athene Annuity and Life Company is the United States operating insurer inside Athene, which is in turn owned by Apollo Global Management. It is the entity that issues the annuities. It is the entity that holds the promises.

Buried in Athene’s 10-K, in the section on reinsurance, is the following sentence. The exact words matter.

“Subject to quota shares generally ranging from 80% to 100%, substantially all of the existing deposits held and new deposits generated by Athene’s U.S. insurance subsidiaries are reinsured to its Bermuda reinsurance subsidiaries.”

Read it slowly. “Substantially all” of the deposits. The existing ones and the new ones. Reinsured, meaning transferred, to the company’s own Bermuda subsidiaries. At a quota share that runs as high as 100%.

In plain English: when an American buys an annuity from Athene’s US insurer, the money does not stay at the US insurer in any meaningful sense. It is shipped, very nearly in its entirety, to a reinsurance company in Bermuda that the same parent owns. The US entity becomes something close to a pass-through. It collects the premium, books the liability, and reinsures the economics offshore.

The next sentence in the filing tells you why.

“As a result of its internal reinsurance structure and third-party direct to Bermuda business, a significant majority of Athene’s aggregate capital is held by its Bermuda reinsurance subsidiaries.”

The capital is in Bermuda. Athene is telling you this directly. Not most of the assets. The capital, which is to say the buffer, the surplus, the thing that stands between a promise and a broken promise.

The distinction that sounds technical, and is in fact the entire point: insurance regulation is local. When an insurer fails, it does not fail as a global consolidated enterprise. It is seized and wound up by the regulator of the jurisdiction in which the failing entity is domiciled, against the assets legally available to that entity. Athene’s annuity contracts, by Athene’s own disclosure, are the obligation of the issuing US insurer. The capital, by the same disclosure, is largely in Bermuda.

Bermuda is not a secret and it is not, in itself, a crime. It is a sophisticated, well-regulated jurisdiction with a competent monetary authority. But Bermuda’s solvency regime is not the United States’ solvency regime. Athene’s own filing says it applies the same reserving principles in Bermuda as in the United States, and we take that at face value: the concern here was never reserving. It is capital treatment, and above all public disclosure, where the gap is dramatic. Nemeth’s letter put the disclosure gap in concrete terms: Athene’s US entity files a statutory return of more than nine thousand pages. Its Bermuda entity files eight. The ratio of disclosure between the entity that owes you the money and the entity that holds the capital is roughly twelve hundred to one.

A third actor is named in the same passage of the 10-K: ACRA, the Athene Co-Invest Reinsurance Affiliate. Remember it. ACRA is a sidecar. It allows third-party investors to put capital alongside Athene inside the reinsurance structure. Those third-party investors have their own economics and their own exit rights, and they have no statutory obligation to the US policyholder. We will come back to what that means.

While we are in the 10-K, one more sentence, from the section on ratings. Athene notes, revealingly, that financial strength ratings “are based upon factors of concern to policyholders, agents, intermediaries and ceding companies and are not directed toward the protection of investors.”

We’d push that further. The “A+” on the door is a signal calibrated for the comfort of the people buying the product. It is not, and was never designed to be, a measurement of the distance between this balance sheet and an adverse outcome.

Nemeth’s letter set out to make that measurement, and Athene’s own filings make it easy. As of year-end 2025, Athene’s US insurance companies, the entities that issue the annuities and legally owe the money, held $9.5 billion of total adjusted capital and reported a risk-based capital ratio of 436%. The ratio sounds like a fortress. It reads that high only because the requirement it is measured against is tiny: the risk charges the formula applies to a 97% investment-grade bond book come to a fraction of one percent of assets, so 436% is more than four times a very small number.

And the $9.5 billion sits at the wrong end of the structure. As Athene itself disclosed above, a significant majority of the group’s capital is in Bermuda: the filings show $18.6 billion of statutory capital in the Bermuda reinsurers against $9.5 billion in the US companies that actually owe American annuitants.

None of this is alleged. None of it is leaked. It is the company’s own description of its own structure, filed with the Securities and Exchange Commission, available free to anyone who wants to read it. The scandal, if there is one, is not that the structure is hidden. It is that it is disclosed in plain sight, and almost no one outside the regulators and a handful of specialists reads the disclosure.

3. Apollo Is Not a Private Equity Firm

Apollo Global Management is described, by the financial press and by itself, as one of the three or four largest private equity firms in the world. One table retires that description.

It is the segment breakdown of Apollo’s own revenue, pulled straight from the terminal.

Here’s the number. In 2025, Apollo’s most recent full year, the Retirement Services segment, which is to say Athene, the insurance business, was 84% of Apollo’s total revenue, $27.0 billion of a $32.0 billion total. The figure is straight from Apollo’s own 10-K, and it is not a one-year artifact: in 2023 it was 89%.

Two honest qualifications. First, this is revenue, and revenue flatters an insurer: consolidated accounting books gross premiums and gross investment income at the top of the income statement, against large policy-benefit expenses recorded below it. On Apollo’s own preferred non-GAAP segment metrics, which weight recurring fee earnings differently, the asset-management business looks larger and the split is less lopsided, though insurance is still the bigger half. Second, cutting the other way: on GAAP pre-tax income the tilt is sharper, not gentler, with Retirement Services accounting for the overwhelming majority of what Apollo earned before tax in 2025.

Choose whichever lens you find fairest. On none of them is Apollo, today, primarily a private equity firm. It is first a life insurance and annuity company, and the asset-management business the world still calls “Apollo” is the smaller half.

This isn’t a criticism. It’s a correction of a category error, and the error changes the question you should ask. Nobody is especially alarmed when a private equity firm takes risk. Risk is the job. Limited partners in a buyout fund are sophisticated, they are compensated for the risk, and they can lose their entire investment without anyone calling it a crisis. That is the deal.

An annuitant did not sign up for that deal. The retired teacher with the fixed annuity is not a limited partner. She was not sold an exposure to private credit spreads. She was sold a contractually guaranteed stream of payments by an institution rated A+. The marketing said safety. The structure, as we have just read in the filings, says something else.

The same shape, with different timing, shows up across the industry’s new owners. KKR, through Global Atlantic. Carlyle, through Fortitude Re. Brookfield, through Brookfield Reinsurance and American National. Blackstone, through its stake in Corebridge and the management of large insurance mandates. The asset managers did not all enter insurance at the same moment, but they all entered, and they all entered for the same reason: the permanent capital. The McKinsey report did its job.

4. The Wall

There are two ways money arrives on a life insurer’s balance sheet.

The first is the obvious one: policyholders pay premiums. Annuity deposits. That money is genuinely long-dated and genuinely sticky, and it is the money the “permanent capital” thesis is actually about.

The second is less visible and, for our purposes, far more interesting. Modern insurers, including the traditional ones, also raise money in the wholesale debt markets through vehicles called funding agreement backed notes. A funding agreement is, in effect, an institutional deposit with the insurer. The insurer bundles funding agreements into notes and sells the notes to bond investors all over the world. The proceeds get invested alongside everything else, earning the spread.

Funding agreement backed notes are not exotic and they are not, in isolation, sinister. MetLife has used them for decades. But they have one characteristic that annuity deposits do not, and the characteristic is the whole of this section: they have a maturity date. They are debt. They come due. And when they come due, the insurer must either pay them off or, far more commonly, refinance them by issuing new notes.

We pulled Athene’s funding agreement program off the terminal and laid out its maturity schedule.

Across its maturity-dated wholesale funding, Federal Home Loan Bank advances, the FABN medium-term-note program issued through Athene Global Funding, the FABR repurchase program, and direct funding agreements, Athene had roughly $90 billion outstanding as of March 2026. The FABN notes this section opened with are $34.5 billion of that. The FABR program, repo-based funding raised through special-purpose entities, is another $21.5 billion; together those two programs total about $56 billion. Now look at when they come due. This is not a smooth ladder spread evenly across decades. It is a wall. The single largest maturity year is 2027, and in 2027 alone, approximately $17.5 billion comes due. That is close to a third of the entire program landing inside one calendar year. The bulk of the rest lands in the handful of years on either side of it.

Hold that next to a fact from Section 1. The liabilities this balance sheet exists to support, annuities and pension obligations, run ten, twenty, thirty years and longer. A meaningful slice of the funding behind those decades-long promises rolls over inside the next five years, and a third of it inside one.

This is the oldest and most dangerous shape in finance. Borrow short, lend long. Fund long-dated, illiquid assets with short-dated money that has to be refinanced. It works, beautifully, for as long as the refinancing market stays open. It is precisely, mechanically, what failed at Silicon Valley Bank: long-duration assets, funded by liabilities that turned out to be far less stable than the model assumed, and a forced sale into a falling market the moment the funding ran.

Three honest caveats.

First, funding agreement backed notes are not retail deposits. They are held by institutional investors, and institutional investors do not “run” in a literal panic the way retail depositors did at SVB. The failure mode is slower and quieter. It’s a refinancing that gets harder, then more expensive, then, in a genuinely bad credit environment, unavailable on workable terms.

Second, Athene isn’t unique, and it isn’t even the most aggressive user of this funding. MetLife’s comparable program is larger. Corebridge, Global Atlantic, Equitable, Prudential’s Pricoa vehicle, Brighthouse, and American National all run the same structure. This is an industry-wide funding model, not an Apollo invention. That cuts in Athene’s favor and we say so plainly.

But it also cuts the other way, and this is the part the industry would prefer you did not dwell on. If the wholesale funding market for insurance paper seizes, in a 2027 that happens to coincide with a credit downturn, it does not seize for one issuer. It seizes for the asset class. Every name on that list refinances into the same closed window at the same time.

Third, the difference that matters between these issuers is not the funding. It is the asset side. MetLife funds short and lends long too, but MetLife lends into a relatively conservative, relatively liquid book. The question is what happens when an insurer funds short, lends long, and lends into a book of private credit and structured assets that does not trade, cannot be sold quickly, and is carried on the books at a value the market never gets to vote on.

That is the next section.

5. The Money Goes in a Circle

What the insurer actually buys with all this money is, again, in its own filings.

Athene’s 10-K breaks out a category it calls net alternative investments. At year-end 2025 it totaled roughly $13.9 billion.

The list: origination platforms with names like Wheels, MidCap Financial, Apterra, Atlas, Foundation Home Loans. A line called, simply, “Redding Ridge,” at $645 million. And then a large block, nearly half the total, labeled “Apollo and other investments.”

The structure: Athene, the insurer, is invested in a set of platforms and funds. A great many of those platforms and funds are originated, managed, or affiliated with Apollo, the parent. The insurer’s “alternative investments” are, to a substantial degree, Apollo products.

Follow Redding Ridge. It’s the cleanest illustration. Redding Ridge is an Apollo-affiliated manager of collateralized loan obligations. Athene holds $645 million of it. And in a separate part of the filings, Apollo discloses what Redding Ridge does with the relationship.

The disclosure states that Redding Ridge made payments to affiliates of the Operating Manager for management and performance fees, and for credit research and administrative fees. The Operating Manager’s affiliate is Apollo. There are parallel disclosures for collateral management fees, for “Capital Solutions” fees on asset-backed finance, and for a separate Apollo-affiliated client services platform called Lyra.

Trace the dollar. A retired teacher pays a premium to Athene’s US insurer. Substantially all of it is reinsured to Athene’s Bermuda subsidiary. The Bermuda balance sheet invests in, among other things, Apollo-affiliated vehicles like Redding Ridge. Those vehicles pay management fees, performance fees, structuring fees, and administrative fees to Apollo affiliates. The teacher’s premium has become an asset. It generates a spread for the balance sheet, and a toll, collected by the parent, at multiple points along its path.

The individual fee amounts disclosed in any single quarter are not, on their own, enormous numbers. Some of them are surprisingly small. The point is not the size of any one toll. The point is the architecture. The same parent sits at the origination of the asset, the management of the asset, the reinsurance of the liability the asset funds, and the collection of fees at several stations in between. Apollo is on every side of the table.

This is the part the word “extraction” is reaching for. The fees are not hidden. They are disclosed, quarter after quarter, in the filings. But they are disclosed in a hundred different footnotes across thousands of pages, and a structure disclosed in a hundred footnotes is, for any practical purpose, a structure that has not been disclosed at all.

There is one more property of this asset side that does more work than any fee. The assets do not trade. And because they do not trade, the value of them is, to a degree most people do not appreciate, a matter of opinion.

6. The Opacity Is the Product

The terminal work turned up something we didn’t expect. On reflection, it’s the quiet center of this whole story.

We went looking for the holders of specific collateralized loan obligations, the structured credit instruments that sit on these balance sheets. The Bloomberg function for this, HCDS, shows you who owns a given security. We ran it across a range of CLO tranches.

For the large majority of the tranches we examined, roughly nine in ten, the holders came back as undisclosed. Not “held by someone uninteresting.” Undisclosed. The ownership is simply not visible.

When we did get a result, it was thin. On one Apollo-affiliated CLO sub-tranche, the entire identifiable ownership was two names, together a few percent of the tranche. The other ninety-something percent was dark.

The instinct, when you see this, is to treat it as a data limitation. A gap in the terminal’s coverage. It is not a gap. It is the design.

Three different things are at work here, and the industry’s defenders will, fairly, attack anyone who runs them together. First, “privately placed” describes how a bond was sold, directly to institutions rather than into the public market, not how risky it is. A private placement trades less, but it is not by definition unrated or unreviewed.

Second, valuation. A security without an active market is still assigned a fair value, and that value is graded: Level 2 when the inputs are observable, such as a price from a third-party service, Level 3 when it rests on significant unobservable inputs and management judgment. Most of an insurer’s non-public bonds are Level 2; a meaningful tail is Level 3. Level 3 does not mean fraudulent. It means the number is an estimate whose inputs an outsider cannot independently check.

Third, accounting basis. In the statutory filings that drive an insurer’s solvency assessment, most bonds are carried at amortized cost regardless of fair value, so until an instrument is formally impaired, a decline in its market value need not pass through the statutory balance sheet at all. The 10-K’s GAAP balance sheet marks more of the book; the statutory balance sheet, the one that decides whether the company is deemed solvent, marks less.

Stack the three together: a privately placed, Level 3, amortized-cost asset is one that does not trade, is valued by a model the manager influences, and is carried for solvency purposes at a number that does not move when the market’s opinion of it moves. Not every private asset is all three. The ones that are all three are where the gap between carried value and realizable value can widen quietly, for years, before anything forces the question.

This is the same accounting treatment, applied to a portfolio many multiples larger, that allowed Silicon Valley Bank to carry underwater securities at a value that bore no relationship to what they would fetch, right up until the moment it was forced to find out.

Nemeth’s letter established that roughly 72% of Athene’s bond portfolio is privately placed. A large share of those private placements are valued against a model rather than a market price, and a model’s output is, in a sense the public is not invited to examine, an opinion. And the rating that determines the capital charge against that opinion is frequently supplied by a small ratings agency, paid by the issuer, processing private credit ratings at a volume that works out to roughly one rating per analyst per business day.

So when we say the opacity is the product, we mean it almost literally. The inability to see the price is not a side effect of the private credit model. It is a feature the model requires. A marked-to-market book cannot hide a loss. A marked-to-model book can hide a loss for years, and “for years” is exactly long enough for the fees to be collected and the architects to be gone.

Occasionally an instrument inside this opaque universe is forced into the daylight. We have a live one.

7. A Live Fire Test: CDK Global

CDK Global is a software company. It provides the systems that car dealerships run on. In 2022 it was taken private by Brookfield, the asset manager, in a leveraged buyout. The buyout was funded, as buyouts in that era were, substantially with a large floating-rate term loan: roughly $4 billion, first-lien, maturing in 2029.

That loan is the kind of instrument that lives, in pieces, on the balance sheets of insurers, private credit funds, and the retail-facing vehicles those firms sell to the public. And right now, that loan is under visible stress.

In April 2026, the ratings agency Fitch added the CDK loan to its list of loans of concern. CDK’s earnings had deteriorated sharply. And then, on 22 April 2026, Bloomberg reported something that does not happen to healthy loans: a group of CDK’s largest creditors signed a cooperation agreement. A cooperation agreement is the choreography that precedes a restructuring. It binds creditors to act as a bloc. You do not sign one unless you expect a fight over the remains. The reported signatories included three of the largest names in credit: PIMCO, BlackRock, and Blackstone.

On the terminal, the loan’s composite price, the BVAL print, sits around 47.6 cents on the dollar.

Now hold that 47.6 against a fact from the private credit world. BCRED is Blackstone’s large retail-facing private credit fund, the vehicle through which ordinary investors get access to this asset class. BCRED holds the CDK loan. And BCRED’s most recent valuation marked that loan at a price far above the high-40s print on the terminal.

Sit with the two facts together. Blackstone, through its credit arm, signed the cooperation agreement in April: an institutional action that signals an expectation of restructuring. Blackstone, through BCRED, carries the same loan at its most recent mark, the quarterly valuation struck on 31 March, at a price well above where the loan now trades. That mark predates both the April cooperation agreement and the current terminal print. It is, in other words, a stale number, set before the loan’s most visible leg down, and it is the number BCRED’s retail investors see when they check their account, and the number their management fee is calculated against.

Same parent. Same instrument. Two valuations, struck for two different audiences. One of them, current and institutional, informs a restructuring strategy. The other, older and retail-facing, informs what an ordinary investor believes their holding is worth.

This is the same architecture we found in the insurance filings, surfacing in a different segment of the same firms’ empires. An institutional view of an asset and a retail-facing view of the same asset, maintained by the same parent, that do not agree. Our colleague Leyla Kunimoto goes considerably deeper on the private credit dimension of this in a forthcoming piece, and we will not get ahead of her work. We raise CDK here for the reinsurance angle, and for that, we pulled the full holder list of the loan off the terminal. It is the most interesting document we obtained in the entire investigation.

Two things jump out.

The first. The single largest identifiable holder of this distressed, restructuring-bound loan is Allianz, at roughly $240 million, just over 6% of the entire facility. Allianz is one of the largest insurance companies on earth. And Allianz, alone among the large holders, was not selling. Allianz was the most aggressive buyer in the entire holder list, having added roughly $62 million to its position through the very period in which the loan was deteriorating toward the high-40s.

The second. Around Allianz, the behavior of the other large holders runs the other way. The PE-affiliated credit managers, in aggregate, were net sellers. The single largest seller in the entire list was a Carlyle vehicle, reducing its position by roughly $49 million. The names with the deepest institutional credit franchises were, on balance, heading for the exit. The flow across the whole holder list was net negative: money leaving.

Now the detail that ties it together. Three public facts, in order.

Fact one: PIMCO is one of the named signatories of the CDK creditor cooperation agreement, the restructuring bloc.

Fact two: PIMCO has been wholly owned by Allianz since 2000.

Fact three: Allianz is the largest holder of the CDK loan, and was its most aggressive buyer through the decline.

We’re not going to tell you what that means, because to assert a single interpretation would be to claim knowledge of internal information flows that we do not have, and that no honest outside analyst could have. There are benign readings. A global firm is not a single brain; the left hand and the right hand of a company that size genuinely do act independently, and often should. There are less benign readings. We are showing you the facts, in the correct order, from public sources, and we are leaving the interpretation where it belongs, which is with you.

What we will say is this. The CDK holder list is a photograph of the thing this entire piece is about. A distressed, opaque, model-valued private credit instrument. A web of the largest asset managers and insurers in the world, all holding pieces of it. The sophisticated institutional money, on balance, quietly leaving. And the retail-facing and insurance-facing vehicles, the BCRED investors and the Allianz policyholders, still inside, still carrying it, still being charged a fee on a number a model produced.

That is the structure. CDK just happens to be a corner of it that the daylight reached.

A brief interruption, because this is the kind of work we exist to do.

TSCS publishes institutional-grade research on the structures that move markets, before the structures become headlines. Our Nuclear Primer became a reference document. “Don’t Short SaaS” called the bifurcation of the software selloff while the market was still panicking. Our private credit series, with Les Barclays and now Leyla Kunimoto, has been mapping the marks-versus-reality problem for months. Nick Nemeth’s open letter to Speaker Johnson reframed the insurance debate in Washington.

If you want this kind of work in your inbox, and you want it before the crowd arrives, subscribe.

Subscribe now

8. The Market Has Already Voted

Step back. If this trade, buy the insurer, swap the assets, collect the spread and the fees, is so good, who’s been collecting the proceeds?

We pulled ten years of total return, indexed, for two groups. The first group is the old guard: the traditional, publicly traded life insurers and reinsurers. Reinsurance Group of America. MetLife. Swiss Re. Munich Re. Hannover Re. SCOR. The second group is the new guard: the publicly traded parents of the new model. Apollo. KKR. Blackstone. Carlyle. Ares. Brookfield.

The result is not subtle. Over the decade, the old guard compounded total return at roughly 7% to 17% a year. Respectable, boring, insurance-like numbers. The new guard compounded at roughly 15% to 29% a year. Apollo’s total return over the ten years was north of 1,100%. Ares’ was higher still. The asset manager parents did not beat the traditional insurers by a margin. They lapped them, two and three times over.

And look at the underlying business growth. RGA, the largest pure life reinsurer in the old model, grew its life premiums from roughly $9 billion to roughly $17 billion over nine years. It roughly doubled. Solid. Apollo’s Retirement Services revenue went from $1.4 billion in 2019 to $27 billion in 2025, though much of that step-up is the accounting consolidation of Athene from 2022, not organic growth. The old guard did not stop working. It got lapped by an institution running a different machine.

The equity market, in other words, has already rendered a verdict, and the verdict is emphatic: the new model wins. The capital has voted.

But that vote measures one thing only: which model is more profitable in a benign environment. It does not, and cannot, measure whether the more profitable model is more robust. Those are different questions, and the equity market, over the last ten years, has only been asked the first one.

There is a second market, and it has rendered a notably more cautious verdict. The credit market, where these same firms borrow, is supposed to be the careful one: the market that prices downside rather than dreams. We pulled a decade of borrowing spreads for the asset manager parents and set them against the traditional insurers.

A decade of spectacular equity outperformance has not bought the new model a credit market endorsement. If the bond market genuinely believed the PE-owned insurance complex were as robust as its equity returns imply, the parents would borrow at a clear premium to the sleepy old guard. They do not. In aggregate they borrow roughly in line with the traditional insurers, and several of them, including the parents of major insurance platforms, borrow visibly wider than MetLife or Prudential, not tighter.

The equity market has priced these firms for a generation of compounding success. The credit market has declined to award them a single basis point of safety premium for it. The two markets disagree this sharply about the same companies. The equity market votes on how good the upside looks. The credit market votes on whether you get your money back. The market that has been less impressed is the one whose job is the question this piece is about.

9. Why You Cannot Opt Out

Everything to this point has been about institutions. Here is where it touches a person.

In May 2023, AT&T transferred its pension obligations for roughly 96,000 retirees and beneficiaries to Athene, in a single transaction worth about $8 billion. This is called a pension risk transfer, and it has become one of the busiest corners of the insurance industry. A corporation with a defined-benefit pension plan hands the obligations, and the matching assets, to a life insurer. The insurer issues annuities. The retirees, who used to be beneficiaries of a corporate pension, become annuitants of an insurance company.

The retirees did not vote on this. They were not asked. One day their pension was an obligation of AT&T, backstopped by the federal Pension Benefit Guaranty Corporation. The next day it was an obligation of Athene, backstopped by something else entirely. AT&T, for the record, recognized a pre-tax pension settlement gain of $363 million on the transaction and shed its PBGC premiums. The transfer was good business for AT&T. Nobody asked the 96,000.

And, by the structure we read in Section 2, a substantial majority of those transferred obligations were then reinsured offshore. Athene’s own disclosures indicate the majority of its pension risk transfer liabilities are reinsured to its Bermuda affiliates, consistent with the internal reinsurance structure described in Section 2.

When a corporate pension fails, the PBGC stands behind it. The PBGC is a federal entity, it is pre-funded, and while its coverage has limits, it is a genuine, funded backstop. When an insurer fails, there is no PBGC. There is the state guaranty fund system.

The state guaranty funds are real and they have paid claims. But they are not the PBGC. They are organized state by state. They are not pre-funded in any meaningful sense; they raise money, after a failure, by assessing the other insurers in the state. And their coverage is capped, commonly around $250,000 of present value per individual, with the precise figure varying by state. For a modest annuity, that cap is sufficient. For a retiree whose pension was substantial, it may not be.

The retired teacher we started with cannot escape any of this. She cannot move her annuity to a safer insurer without surrendering it, often at a penalty. She cannot read the Bermuda filing, because there is almost nothing to read. She cannot see the marks on the private credit book, because the marks are models and the holders are undisclosed. She was told, by every signal the system is designed to send, that she had bought the safest thing in finance. She had in fact been moved, without a vote, onto a balance sheet that Apollo’s own filings describe as funded short, invested in model-valued private assets, and capitalized largely offshore.

That is what “you cannot opt out” means. The people with the most at stake in this structure are the people with the least ability to see it, understand it, or leave it.

10. What Would Make Us Wrong

We have built a deliberately alarming case. Here is what would make this piece wrong, and we’re not going to be coy about any of it.

We are punishing the company that showed its work. Athene publishes voluntarily. It puts its US statutory filing on its own investor relations website. The 80% to 100% quota share, the capital-in-Bermuda admission, the affiliated fee disclosures: we were able to build this case largely because Athene disclosed the raw material. The roughly 700 other US life insurers, many smaller, less sophisticated, and less well capitalized, disclose far less, and some are surely running cruder versions of the same structure with worse assets and thinner buffers. There is something genuinely uncomfortable about training the brightest light on the operator that left the lights on. Athene is, by most credible accounts, among the most sophisticated operators in this space. If the best-run version of this structure is the one we have just spent the piece worrying about, that is either evidence the worry is overblown, or evidence the problem is much larger than one company. We think the latter. But the discomfort is real and we will not pretend it away.

The ratings agencies have more information than we do. Athene’s insurers carry an A+ from S&P, Fitch, and AM Best, and Moody’s equivalent A1. Those agencies have access to data, including confidential Bermuda data, that we do not. It is entirely possible they have examined the consolidated structure and concluded it is robust, and that our outside view, built from public filings, is missing capital and protections that are real but not publicly visible.

The consolidated capital number is much larger than the entity number. Nemeth’s letter focused on the modest capital at the US operating entities relative to the obligations they have written. Athene will point out, correctly, that on a consolidated basis, across the US and Bermuda entities and including third-party ACRA capital, the figure is many times larger, and that the internal reinsurance structure is specifically designed to make that aggregate capital available across entities. That is a serious rebuttal. Our answer is the one we gave in Section 2: in a receivership, what protects a policyholder is the capital legally obligated and available at the failing entity, under the jurisdiction that regulates it, and ACRA’s third-party investors have no such obligation. But this is a genuine analytical dispute, not a settled fact, and a fair reader should weigh both sides.

Regulators are already moving. The Bermuda Monetary Authority introduced enhanced disclosure rules for these reinsurers, effective from year-end 2025, with the first detailed asset and liability reporting cycle in 2026. The Financial Stability Oversight Council has explicitly flagged the offshore reinsurance trend. The IMF has raised concerns about life insurers and illiquid assets. The structure we have described is not a secret to the people whose job is to watch it. It is being watched, and the disclosure regime is tightening. The optimistic case is that the system corrects through regulation before it corrects through a failure. We hope that case is right.

This is not a new observation. Academics have studied the underlying maneuver, insurers ceding liabilities to affiliated and more lightly regulated reinsurers, under the label “shadow insurance” for roughly a decade.

Nothing has actually broken. Every alarming sentence in this piece is, strictly, a sentence about a potential. The wall is in 2027; 2027 has not happened. The refinancing market is open; it has not seized. The model-valued assets have not been forced into daylight at scale. A patient critic can fairly say that we have described a structure that has, so far, worked, paid every claim, and made its owners spectacularly rich, and that “so far” has lasted fifteen years.

We hold our view in the face of all of these. But a piece that did not put them in front of you, in their strongest form, would not be research. It would be advocacy. We would rather be honestly uncertain than dishonestly confident.

11. The Translation

Nick Nemeth ended his open letter to Speaker Johnson with a comparison we keep returning to. He showed a risk-based capital ratio that reads like a fortress, sitting on an operating entity whose capital is a thin single-digit slice of the obligations it has written. And the people who will bear the cost of the gap between the headline and the structure did not build it, did not profit from it, and do not know it exists.

This piece has been an attempt to close, slightly, that last gap. Not the financial gap. The knowledge one.

We have not shown you anything that was hidden. That is the entire point. The 80% to 100% quota share is in the 10-K. The capital being held in Bermuda is in the 10-K. Apollo being 84% an insurance company is in Apollo’s own segment reporting. The funding wall is on the terminal. The fee architecture is in the footnotes. The CDK holder list is a function call away for anyone with a Bloomberg login. None of it is leaked, none of it is alleged, and almost none of it has ever reached the people it is actually about.

The life insurance industry spent a century building the most valuable asset a promise-making institution can hold, which is the boring, unimpeachable trust of the people it serves. Over the last fifteen years, a set of brilliant, aggressive, and entirely legal financial engineers worked out how to convert that trust into spread and fees, and they did it in plain sight, in public filings, in a language precise enough to be accurate and dense enough to be unread.

We do not think this ends in a single dramatic morning. The structure is more likely to fail the way Hemingway said you go bankrupt: gradually, then suddenly. A hard refinancing in a bad year. A wave of downgrades through the thin investment-grade tier. A model-valued book that meets a market that insists, finally, on voting. One mid-sized insurer, then a quarter of silence, then another, each failure explained as idiosyncratic until the pattern is undeniable. And underneath all of it, the annuitants. The retired teacher. The 96,000 from AT&T. The people who bought the single thing in all of finance that was supposed to be safe.

There is still time for this to be fixed by disclosure and regulation rather than by failure. Nemeth’s letter laid out what that would take, and we would point Congress and the state regulators back to its five asks: leverage measured honestly rather than through headline ratios, statutory filings made genuinely public and machine-readable, an annual federal report on every insurer that takes pension or annuity money, the filing-exempt ratings backdoor closed, and a review of whether the 1945 statute governing all of this still fits the world it now governs.

We would add only one sentence to his five asks. Require that the person buying the annuity be told, in language she can read, what we have just spent nine thousand words showing you: that the institution holding her retirement is funded short, invested in assets that do not trade, capitalized in a jurisdiction she will never see a filing from, and owned by a firm that earns a fee every time her money moves.

She will likely still buy the annuity. Most people, correctly, will still need one. But she will have bought it knowing. And the entire architecture we have described in this piece depends, in the end, on her not knowing.

We have told her now.

This piece builds directly on Nick Nemeth’s “An Open Letter to Speaker Johnson,” published in March 2026, and is intended as a companion to it. A full layman-to-expert primer on how reinsurance works, “What Is Reinsurance and Why You Should Care,” will follow within the next two weeks.

Nothing in this piece is investment advice. It is structural analysis built entirely from public filings and market data. We hold no position in any security mentioned. Figures are sourced from SEC filings, Bermuda regulatory disclosure, and the Bloomberg terminal as of May 2026; where a figure depends on a methodological choice, we have flagged it. If you believe any fact here is wrong, write to us, because we would rather correct it than defend it.

Subscribe to TSCS for institutional-grade research on the structures that move markets, before they become headlines.

Subscribe now

Share

That sentence should stop somebody from scrolling. It hasn’t. The financial press is busy explaining the rally with earnings narratives and Fed positioning. Both are wrong.

The setup five weeks ago: S&P rolling over at the cycle low. CTAs short 143,835 E-mini contracts per the April COT. VIX at 25.78. Brent above $111 with Hormuz still closed. 10-year at 4.29 percent. Every reasonable read of the tape said the rally was dead.

Today the S&P is at 7,391.75. Up 11.85 percent. April alone delivered 10.49 percent total return. VIX collapsed to 17.94.

No earnings catalyst. No war resolution. No Fed pivot. So what bought it?

A payroll system. A trend-following algorithm. A vol-targeting model. And a cohort of frightened retirees feeding fear capital into the largest cap-weighted index products they recognise. None of them read the news. Roughly $400 billion of mechanical inflows in 30 trading days. About $10 trillion of US equity market cap creation. The math matches.

at Simplify has been documenting this dynamic for five years. In late 2017 he applied the same framework to a $1.9 billion volatility product called XIV. He called the trade at 95 percent probability of going to zero within two years. The product went in six months. Holders lost 96 percent of their capital in a single session.

He’s making the same call now. Same framework, $1.9 billion scaled to $73 trillion. The clock is approximately 30 months.

The casino isn’t plying the bid with alcohol anymore. It’s dropping cocaine on the tables.

The mechanism is below. So is Green’s threshold math with Hari Krishnan and Stephan Sturm, which says a volatility event hits in roughly 30 to 36 months. And the dynamic that destroyed the $1.9 billion XIV product in February 2018 now exists for the entire $73 trillion US equity market.

This is one piece. TSCS publishes work like this weekly. The Private Credit cycle update lands inside seven days with the single-name recalibration on the software cohort named in Section IV. Subscribers see it first.

Subscribe now

Section II. The April Flow Math

Of the $400 billion, only $148.9 billion shows up in conventional ETF flow data. The other $250 billion is invisible to most flow desks. It’s sitting in target-date fund threshold rebalancing, CTA short covering, vol-targeting rebalancing as realized volatility collapsed, and direct payroll contributions into 401k mutual funds.

Four mechanisms. Not one of them needs a human to think.

One. Target-date fund threshold rebalancing.

The largest pool of retirement savings in the United States, roughly $4.8 trillion at year-end 2025, sits in target-date funds. These funds operate on a glide path. A participant 45 years from retirement holds roughly 93 percent equities. A participant 5 years out holds roughly 50 percent. When the equity-to-bond ratio drifts outside its band, the fund rebalances automatically. Through February and March 2026, bonds modestly outperformed equities. By April 7 the band was breached for a meaningful slice of the TDF complex. Sell bonds, buy equities. No view required.

Two. CTA short covering.

CTAs run trend-following strategies. As markets sold off through Q1 2026, CTAs flipped to net short. By April 7, leveraged funds held 143,835 E-mini contracts short. When the market reversed off the low, CTAs were forced to cover. The cover was sharp enough to trigger algorithmic position reversal across the trend-following complex. Within two weeks, CTAs flipped from material net short to near full long allocation. That is roughly $80 to $120 billion of buying force compressed into ten trading sessions.

Three. Vol Control rebalancing.

Vol-targeting funds scale equity exposure inversely to realized volatility. Through the Q1 selloff, the VIX climbed to 25.78. Vol-targeting funds were running well below target equity allocations. As the rally compressed realized volatility, these funds systematically added equity exposure. The VIX collapsed from 25.78 to 17.94. Each step lower forced more buying. Each new buying push pushed the VIX lower. The feedback loop ran for five weeks.

Four. 401k contribution flow.

Approximately $48 billion per month of payroll-deducted retirement contributions hits the market on autopilot. This is the steady base layer underneath the spikier mechanical bid. The flow does not care that Hormuz is closed. It does not care that Brent is above $111. It does not care that the 10-year yield has broken to 4.65 percent. It buys.

Green on Thoughtful Money, May 16:

“We can mechanically identify these flows that came into the markets and it has nothing to do with the war. It has nothing to do with economic outlook. Everything I literally just described to you has no discretionary component to it.”

And on the magnitude:

“We’ve never seen anything like this. We exceeded about $350 billion of inflows in basically the blink of an eye.”

This is the record. Not a record month, the record. Tier 1 Alpha’s flow series has no prior period at this magnitude. Not COVID. Not 2008. Not 1999. Nothing close.

The April 2026 rally is the largest mechanical bid event ever recorded.

Pick your seat. If you’re long, you’re long a market that just had the biggest mechanical bid in history and he now says the systematic complex is fully invested. The ammunition is gone. If you’re short, you got crushed by a force no discretionary desk could have anticipated. If you’re neutral, the position is the most honest and the least comfortable. Fundamentals didn’t matter on any timeframe an active manager cares about. The marginal price setter is a payroll system, a trend model, and a vol algo. None of them read the news.

The S&P at 7,391.75 is not a vote of confidence in the economy. It is a mechanical print.

And the same mechanism that produced this print is the one that will eventually produce its opposite.

Section III. The XIV Analog

In late 2017 Green flagged a structural problem in a $1.9 billion inverse VIX product called XIV. As long as VIX futures stayed in contango, the thing printed money. Retail piled in. Hedge funds piled in. Vol was at multi-year lows.

His analysis was mechanical. He pointed out that roughly 70 percent of daily trading volume in the entire VIX futures complex was tied to the inverse VIX ETF complex. When the underlying instrument supplying the volatility hedge becomes the dominant volume in the volatility market, you have a structural problem. Any small volatility spike forces the product to cover its short volatility position by buying VIX futures. The buying spikes VIX higher. The higher VIX forces more covering. The feedback loop terminates only when the product itself is destroyed.

He put XIV at roughly 95 percent probability of going to zero within two years.

It went in six.

On February 5, 2018, XIV closed at $99.00. By 4:30 PM Eastern Time, the indicative value had dropped to $4.22 after-hours. The product was terminated by Credit Suisse the next day. Holders lost approximately 96 percent of their capital in a single session. The S&P 500 dropped 4.1 percent on the day. Realized volatility on a one-day basis exceeded implied volatility by a magnitude not seen since 1987.

That was a $1.9 billion product.

He now argues that the identical structural mechanism, mechanical buyers becoming the dominant marginal flow, exists today for the entire $73 trillion US equity market.

The argument runs through a paper he released in 2026 with Hari Krishnan, author of The Second Leg Down, and Stephan Sturm, Associate Professor of Mathematical Sciences at Worcester Polytechnic Institute. Sturm presented the work in a department seminar titled “Passive Fragility” in April 2026. The paper uses a closed-form mathematical model to show that when passive market share exceeds approximately 65 percent, there is no longer enough discretionary capital remaining to absorb the volatility signatures that emerge from passive-driven price action. Above the 65 percent threshold, a volatility event of the kind that destroyed XIV becomes mathematically near-certain, not merely possible. The discretionary side cannot provide enough offsetting liquidity to dampen the move.

Today, by his measurement, passive share of US equity AUM sits at approximately 53 percent. It has been gaining approximately 4 percentage points per year for the last decade, accelerated by the active-to-passive AUM crossover that happened in 2024.

The implied timeline to the 65 percent threshold is approximately three years.

He put it this way: “Once it gets inside kind of that 2-year window, there’s tools that are available to start playing it. And, you know, I hate to say I look forward to that opportunity. I think that there will be trades to be done around that.”

Translation. The man who called XIV at 95 percent probability of zero within two years, and was correct in six months, is now telling you that the same mathematical setup is being constructed for the entire US equity market on a timeline of approximately three years.

The number that destroyed XIV was 70 percent of trading volume tied to the inverse VIX ETF complex.

The number that destroys the S&P 500 is 65 percent passive share of total equity AUM.

The mechanism is identical. The dominant marginal flow becomes the source of the cascade. Discretionary capital, which historically absorbed volatility signatures and prevented runaway feedback loops, becomes too small a fraction of the market to provide that function.

XIV cost holders 96 percent of their capital in a single session.

What does the equivalent event look like for the S&P 500?

He hasn’t given a target. The math suggests something between a 30 and 50 percent drawdown over a compressed window of weeks to months, with the steepest losses concentrated in the highest-passive-ownership names. The Mag 7 take the worst of it because they have the highest passive ownership share. The bifurcated 493, which have already compressed their multiples to levels you will see in the next section, get a second leg lower.

This is not a forecast. It is a structural prediction from a paper that uses closed-form math to derive a threshold above which the system mathematically cannot remain stable.

Five reasons XIV isn’t the S&P. XIV had a contractual acceleration trigger at minus 80 percent intraday NAV. The S&P has none. XIV had no lender of last resort. The S&P has a Fed that already demonstrated in March 2020 it will go into equity-adjacent markets when stability is at stake. XIV’s APs were a handful of dealers. The S&P has thousands of ETF APs globally and the plumbing is redundant. XIV had thin depth. The S&P has the deepest equity market in history. XIV had one feedback loop, one instrument, one product forced to buy VIX into a rising VIX. The S&P cascade has no equivalent single loop. The unwinds would run at different speeds.

His answer when pressed: the threshold dynamics survive the plumbing because the binding constraint isn’t the plumbing. It’s whether enough discretionary capital is left to absorb the volatility signature. The plumbing differences slow the cascade. They give the Fed more room. They don’t invalidate the threshold.

Sensitivity on the timeline. The 65 percent threshold and the 4 percentage points per year growth rate are central estimates with material uncertainty around each. At 3 percent annual passive growth, the threshold is breached in 4 years. At 5 percent growth, the threshold is breached in 2.4 years. At 6 percent growth, the threshold is breached in 2 years. If the threshold is actually 60 percent rather than 65 percent, the 4 percent growth case implies breach in 1.75 years. If the threshold is 70 percent, the same growth case implies breach in 4.25 years. Central case is 30 to 36 months. Plausible window is 21 to 51 months. I’m positioning for the central case knowing the breach could come earlier or later.

You have roughly three years.

Section IIIB. Why The Standard Defense Doesn’t Hold

Every passive-friendly defense of the current market runs through Grossman-Stiglitz. The argument: markets can’t go fully passive because the second they did, prices would stop reflecting information and the returns to gathering it would explode. Rational investors would re-enter active until enough informed capital priced efficiency back in. There’s a floor under active management. The system self-corrects. Passive grows without limit because active rationally exists to keep prices honest.

The original Grossman-Stiglitz paper, “On the Impossibility of Informationally Efficient Markets,” was published in the American Economic Review in 1980. Joseph Stiglitz was awarded the Nobel Prize in Economics in 2001 in part for this work. The result is treated in financial economics curricula as settled.

The empirical version that almost every passive-friendly academic citation relies on is Coles, Heath, and Ringgenberg, “On Index Investing.” First drafted March 2017. Most recent revision January 2022. SSRN abstract ID 3055324. Published in the Journal of Financial Economics in 2022. Their abstract states: “An exogenous increase in index investing leads to lower information production as measured by Google searches, EDGAR views, and analyst reports, yet price informativeness remains unchanged. These findings are consistent with an equilibrium in which investors choose to gather private information whenever it is profitable. As index investing increases, there are fewer privately-informed active investors (so overall information production drops), but the remaining mix of investors adjusts until the returns to active investing are unchanged.”

Translation. Even as passive grew from 5 percent of US equity AUM in 1995 to over 50 percent in 2024, prices remained informationally efficient because active investors entered and exited the market until returns equilibrated. The Grossman-Stiglitz floor held. Passive growth is not the structural problem critics claim.

This is the load-bearing defense. Vanguard cites it. ICI cites it. Every academic department defending the current market cites it.

The paper has been methodologically critiqued by Green and others. The critique is structural. Coles, Heath, and Ringgenberg derived their empirical predictions from a Grossman-Stiglitz framework. They then tested whether those predictions matched the data. The predictions matched. They concluded that Grossman-Stiglitz holds. But the predictions were derived from the assumption that informed and uninformed investors equilibrate. Testing whether the equilibration result holds using a framework that assumes the equilibration is a near-tautological exercise. The model holds because the model assumes it holds.

He’s framed the same problem in his own commentary and in correspondence with us. The framework holds because the framework assumes it holds.

The honest empirical question is whether the active investor population responds elastically to changes in passive share. Coles-Heath-Ringgenberg do not test this directly. They test for changes in price informativeness, which is the predicted output of the elastic-response model. If the response is elastic, informativeness is preserved. If the response is not elastic under different assumptions about how informativeness is measured, the result is ambiguous. Their conclusion is consistent with elastic response. It is not a confirmation of it.

The Sammon-Chinco correction sharpens the critique. Marco Sammon and Alex Chinco found in their 2021 work, “The Passive-Ownership Share Is Double What You Think It Is,” that conventional measurement of passive ownership systematically understates the true passive footprint by approximately a factor of two. This is because traditional measurement excludes secondary index-tracking products, separately managed accounts that benchmark to indices, and the increasing tendency of nominally active funds to hug their benchmarks to avoid tracking error. If Sammon and Chinco’s 2x relationship holds today, the conventional 53 percent share implies a true marginal-buyer passive share between 65 and 75 percent. That’s a rough extrapolation. But on the measure academic finance actually uses to test Grossman-Stiglitz, the Krishnan-Sturm threshold has probably already broken. The defense is being made for a market that no longer exists.

Without Grossman-Stiglitz as the load-bearing defense, there is no defensible theoretical floor under active management from the dominant academic framework. Other elastic-response mechanisms may exist, but the framework that the consensus academic position relies on does not provide one. Secularly declining active management economics, fee compression, alpha erosion, and redemption pressure from defined-contribution rotation combine with the structural disadvantage of active strategies in a mechanical-flow-dominated market. Active capital exits. It does not rationally return because the conditions for rational return require an equilibrium that the data does not show.

The implication for the Krishnan-Sturm threshold is severe. If the threshold is crossed and there is no defensible floor under active management, the cascade can run further than the standard defense claims. Combined with the structural mechanics of the volatility absorption deficit that Krishnan-Sturm identify, the bottom of the cascade is bounded only by the point at which mechanical flows themselves reverse. That point is the demographic decumulation crossover that the spatial signature section addresses.

The standard defense of the current passive-dominant market structure rests on a paper whose conclusions are baked into its premises. Removing the assumption removes the floor. The market is more fragile than the consensus academic position acknowledges.

Section IV. The Software Cohort Massacre

April happened. S&P at 7,391.75. The mechanical bid was the largest in history.

And the largest mega-cap software cohort in the US is at the bottom of its ten-year valuation range.

Six names. Five of them at trough multiples. One outlier. Here are the numbers.

ServiceNow. Current EV/FCF: 22.18x. Five-year average: 52.15x. The stock is trading at a 57 percent discount to its own five-year multiple norm. Compression is severe enough that NOW now sits outside its negative-two standard deviation band of 24.56x. Probability of a multiple this low under a normal distribution is under 2.5 percent. The bears were right.

Salesforce. Current EV/FCF: 10.83x. Five-year average: 28.29x. A 62 percent discount. CRM at 11x is the lowest EV/FCF multiple it has traded at in over a decade. The business has not deteriorated. The multiple has.

Adobe. Current EV/FCF: 9.97x. Five-year average: 27.94x. A 64 percent discount. ADBE is now the cheapest it has been in over a decade by every multiple metric. The Figma deal blowup is two years in the rearview. The AI displacement narrative has been priced. And then priced again. And then priced again.

Intuit. Current EV/FCF: 16.82x. Five-year average: 35.59x. A 53 percent discount. INTU still has the strongest moat in SMB tax and accounting software. The moat is intact. The multiple is not.

Workday. Current EV/FCF: 11.07x. Five-year average: 33.68x. A 67 percent discount. The single most aggressive multiple compression in the cohort. WDAY now trades closer to legacy on-premise HR software multiples than to modern SaaS multiples.

Then Oracle.

Current EV/FCF: 101.29x. Five-year average: 43.66x. A 132 percent premium.

Oracle is the outlier. The single mega-cap software name that escaped the cohort bifurcation. The reason is mechanical. Oracle has positioned itself as the central infrastructure layer for the second wave of AI buildout. Its free cash flow denominator has been compressed by an enormous capital expenditure program, while its revenue growth has accelerated on multi-year cloud commitments from OpenAI and other AI principals. The EV/FCF ratio looks anomalous because the F in the denominator is temporarily depressed by capex that will produce revenue in 2027 and 2028.

Oracle’s premium is the exception that proves the rule. The AI narrative trumped the flow dynamic for one name. The other five names in the cohort do not have an AI narrative aggressive enough to break the compression.

Look at the cohort average without Oracle.

Average current EV/FCF: 14.17x. Average five-year EV/FCF: 35.53x. Cohort discount to historical: 60 percent.

These are five of the best enterprise software businesses in the US. FCF margins above 30 percent. Net retention above 110 percent. Switching costs measured in years. ServiceNow printed record Q1 2026 free cash flow at a 44 percent margin. The fundamentals didn’t break.

The multiples did.

This is the empirical receipt for what Green calls the bifurcation. The mechanical bid does not allocate evenly across the cap-weighted index. It allocates by index weight, by passive ownership share, and by the AI narrative gravitational pull. Names that fail all three filters get crushed regardless of fundamentals. Names that pass even one filter, like Oracle, escape.

What does the next leg lower look like if the bid reverses.

For these five names, the next stop in the active-only band, the level historically defended by fundamental long-only buyers absent any flow protection, is 8x to 12x EV/FCF. ServiceNow at 8x EV/FCF, holding TTM FCF roughly flat, implies a stock price of approximately $37, against today’s $101.60. That is the second leg. If the mechanical bid that produced April reverses with anything resembling symmetry, these names see another 25 to 60 percent of downside from current already-depressed levels.

Subscribers who have been reading TSCS for the last six months will recognize this as the empirical reckoning of a framework that we updated through April. We will address that update directly in the upcoming Private Credit cycle piece in the coming days. For now, the receipt is the receipt. The bifurcation is not theoretical. It is twenty-two times trailing free cash flow on the largest workflow software company in the United States.

Section IV.5. The Cascade Has Already Started, But Not The Way You Think

The Krishnan-Sturm framework predicts a synchronized volatility event at 65 percent passive AUM share. The standard reading: cascade triggers all at once when the aggregate breaches. The data doesn’t support that reading. The cascade already started. It runs sequentially, name by name. The trigger isn’t aggregate passive ownership. The trigger is high institutional concentration plus an arriving fundamental catalyst.

The empirical receipt is in the institutional ownership data for the top 32 S&P 500 names by market cap as of May 19, 2026. Average institutional ownership across the cohort is approximately 77 percent. The largest names sit between 70 and 95 percent. This is the mechanical bid concentration zone. These are the names where passive and benchmark-hugging active dominate the marginal buyer composition.

Inside this concentration zone, the names are bifurcating sharply on YTD performance, but the bifurcation does not track ownership concentration in the way a naive reading of the flow theory would predict.

GOOGL up 25 percent YTD. Institutional ownership 83 percent. NVDA up 20 percent YTD. Institutional ownership 71 percent. AMZN up 12 percent YTD. Institutional ownership 69 percent. AAPL up 10 percent YTD. Institutional ownership 69 percent. META down 8 percent YTD. Institutional ownership 80 percent. MSFT down 13 percent YTD. Institutional ownership 77 percent. TSLA down 10 percent YTD. Institutional ownership 45 percent.

Microsoft has the third-highest institutional ownership in the Mag 7. It’s the worst performer. Meta has the second-highest. It’s the third-worst, down meaningfully while the index hits all-time highs. The naive flow reading says these names should be insulated. They aren’t.

The honest reading of the data is that institutional concentration creates brittle support, not a floor. The marginal buyer composition at high-concentration names is narrow enough that the mechanical bid drives price persistence in the absence of catalysts. But the same narrow composition cannot absorb a coordinated active selling pressure when a fundamental thesis takes hold. The bid is binary. Either it holds the price up entirely, or it gets overwhelmed entirely. The middle ground of gradual orderly decline is rarer.

MSFT got hit on AI capex margin compression and the Activision integration questions. META got hit on ad market deceleration and AI displacement concerns on search-substitute behavior. In both cases, the catalyst was a coherent fundamental thesis that gave coordinated active selling permission to operate. Once that permission existed, the institutional bid was overwhelmed because the discretionary side was too thin to provide offsetting absorption.

Catalyst-triggered cascade is one read of the MSFT and META drawdowns. Other reads: sector rotation out of large-cap tech, name-specific fundamental concerns, active managers selling on valuation. I think all three run at once. The cascade mechanism is what lets the other explanations translate into outsized price action.

TSLA is a different category. Forty-five percent institutional ownership is the lowest in the top 32 by a wide margin. This is not a flow story. TSLA is insider-anchored by Musk and retail-driven on the float. The drawdown reflects EV competition, margin compression, and ongoing distraction by Musk’s other ventures. The TSLA case is a reminder that ownership concentration measurement requires careful interpretation. Low institutional ownership at TSLA does not mean high active discretionary participation. It means high insider plus retail participation, which is its own structural fragility.

The cascade mechanism we observe in MSFT and META is what the Krishnan-Sturm framework predicts, but applied at the name level and triggered by catalysts rather than synchronized at the aggregate level. The four conditions that combine to trigger a name into the cascade:

One. High institutional concentration. Above approximately 70 percent.

Two. Weak discretionary support. Active managers underweight or neutral. No identifiable Tiger Global or Altimeter-style concentrated active long that anchors the floating supply.

Three. Absence of an AI narrative gravitational pull. Names not perceived as core to the AI infrastructure or application stack.

Four. A credible fundamental catalyst. Earnings miss, margin compression, competitive disruption, regulatory action, or insider sale signal.

When all four conditions coincide, the institutional bid is overwhelmed. The name breaks decisively. The break is not a gradual decline. It is a step function from indexed and absorbed to actively sold and not absorbed.

This four-condition framework is ours, built off what we’re seeing in the data. It isn’t validated at scale yet. We’ll know in 12 months whether more names break the way the framework predicts.

This is the TSCS-original prediction. We expect further bifurcation inside the high-concentration cohort over the next 12 to 18 months. We expect the names that next break to be those where the four conditions converge. We are not naming candidates today because catalyst arrival is endogenous and we cannot fully anticipate it. But the framework is falsifiable. If 12 months from now the high-concentration cohort has not produced further dispersion, the catalyst-triggered cascade thesis is wrong.

The implication for further investigation is that non-long-only books should examine specific names in the high-institutional-ownership cohort for fit with the four-condition framework. Where the conditions converge, the asymmetric trade is to hedge directly with single-name puts struck below recent technical support, or to short outright if mandate permits. We are not naming candidate exposures in this piece. The Private Credit cycle piece next week handles single-name positioning where the framework applies.

The aggregate threshold framework predicts the cascade at the system level. The data shows the cascade already operating at the name level. The two readings are not contradictory. They are sequential. The aggregate threshold breach is the moment the system runs out of names that have not yet broken. By that point, the cascade has already destroyed substantial value in the names where catalyst-triggered breaks have already occurred. The investor who waits for the aggregate threshold to break before positioning is too late.

The cascade has already started. The aggregate event is the back end of the distribution, not the front.

Section V. Inside The Mag 7, The Bifurcation Goes Deeper

Consensus: “the Mag 7 is leading the S&P higher.” That’s wrong. And the reason it’s wrong matters for the entire flow framework.

YTD 2026 total returns, Mag 7, by name:

Alphabet. Up 25.04 percent. NVIDIA. Up 20.13 percent. Amazon. Up 12.07 percent. Apple. Up 9.84 percent. Meta Platforms. Down 8.37 percent. Tesla. Down 10.31 percent. Microsoft. Down 13.11 percent.

Four of the Mag 7 are up. Three of the Mag 7 are down. Two of those three are down double digits. The third, Meta, is down 8 percent. The dispersion is severe while the index hits all-time highs.

This is the bifurcation inside the bifurcation.

The cap-weighted S&P 500 is up 8.39 percent YTD. The equal-weighted S&P 500, RSP, is up 6.13 percent. The spread is 2.26 percent. That spread is the entire Mag 7 contribution to index leadership, and even within the Mag 7, three of seven names underperformed the equal-weight.

Here’s the data point almost nobody outside the flow framework reads correctly.

NVIDIA, the single largest individual contributor to the S&P 500 in April, trades at a forward P/E of 24.0x. Its five-year average is 36.6x. NVDA is at a 35 percent discount to its own multiple norm.

NVDA went up 20 percent YTD on the price chart. NVDA’s multiple compressed by approximately 35 percent against its five-year average. The reconciliation is straightforward. NVDA’s price rose because earnings grew faster than multiples compressed. The market is paying less per dollar of expected earnings than it did 18 months ago, while the absolute earnings have grown enough to push the price higher.

What does this mean for the flow theory.

The flow theory is right. Almost everyone misreads what it predicts.

The framework doesn’t predict multiple expansion at the largest names. It predicts price persistence regardless of fundamentals. The mechanical bid buys NVDA because NVDA is the biggest index weight. It doesn’t bid the multiple higher. It bids the price as a function of weight and earnings growth. The multiple is a residual of the discretionary bid on top. When the discretionary bid is absent or negative, as it’s been at NVDA, the multiple compresses even as the price rises.

Bears are stuck.

Short NVDA on multiple expansion? The multiple is compressing. Short on flow? The flow is supporting the price. Short on fundamentals? Earnings are growing faster than the multiple is compressing. NVDA is the cleanest case study in why active has failed to keep up with the index for three straight years.

What you can do is identify the names where the mechanical bid does not operate.

The five software names in Section IV. The 493 ex-Mag-7-ex-Oracle. The mid-cap quality businesses with no index inclusion. The small-cap discovery zone.

That is the actual trade the bifurcation produces. Not short the largest names. Long the names with structural protection plus genuine moats. Short the names without flow protection and without moats.

We will address the trade construction in Section VII.

For now, the receipt is the bifurcation runs deeper than even the “Mag 7 versus 493” framing captures. It runs three of the Mag 7 down. It runs the cohort multiple compression at five of the six largest mega-cap software names. It runs the equal-weight S&P up barely half of the cap-weight. The dispersion is the story. The index is hiding the dispersion. The framework is the only one that predicts the dispersion correctly.

Section VI. The Spatial Signature

The April mechanical bid was not random money. It had a demographic shape, and the shape produced the bifurcation we documented in Sections IV and V.

The steady payroll-deduction flow into 401k plans contributes approximately $48 billion per month, roughly 12 percent of the April total. TDF threshold rebalancing into equities, triggered by the relative price action between bonds and equities through Q1, contributes perhaps the largest single component at 30 to 50 percent. CTA short covering and vol-targeting rebalancing contribute the bulk of the remainder, perhaps 30 to 40 percent combined. A discretionary component layered on top, dominated by wealthy boomer fear capital concentrating into the largest cap-weighted index products, accounts for the small remainder and, more importantly, determines where the marginal dollar concentrates.

This discretionary component, layered on top of the mechanical base, is largely wealthy boomer fear capital. This is the cohort holding oversized taxable accounts outside their retirement plans, hoarding because of longevity risk, end-of-life-care anxiety, and a fundamental distrust of pension promises that did not exist when they were 30. Per Green, citing a Dave Collum poll on his Substack last weekend, 63 percent of respondents believe they need 20x pre-tax annual income or more to retire safely. The actual math, accounting for Social Security, post-tax spending reality, and 5 percent yields on long Treasuries, is closer to 7x.

The Vanguard How America Saves 2024 report, with year-end 2023 cohort-specific data, reports a median 401k balance for participants aged 55 to 64 of $87,571. The average is $244,750, but the gap between average and median indicates the average is inflated by high earners in the right tail. The median is the better representation of the typical cohort participant. Against a median household income for the same cohort of approximately $85,000 to $95,000, the typical 55 to 64 year old has saved between 0.9 and 1.0 times pre-tax annual income. The Vanguard 2025 update report, released early 2026 with year-end 2024 data, confirms the directional picture at the all-ages level with an average balance of $148,153 and a median of $38,176, and at the 65 and over level with an average of $299,442 and a median of $95,425.

The Schwartz Center analysis of the 2022 Federal Reserve Survey of Consumer Finances sharpens the picture. The middle 70 percent of households by income aged 50 to 65 carry median retirement savings of $86,000 against median debt of $89,700. Net retirement savings position for the median pre-retiree household is negative.

The cohort that thinks it needs 20x its pre-tax income has on average saved between 0.9x and 1.0x, with median debt exceeding median retirement assets. The perception-versus-math gap is a factor of twenty. The oversaving thesis applies primarily to a wealthy minority within the cohort. The median household is undersaved, not oversaved. The fear capital concentration into cap-weighted index products is being driven by the wealthy minority who are oversaving in taxable accounts because they correctly observe that their cohort is undersaved on average, and they do not want to be one of those undersaved households. The mechanical bid is therefore not aggregate boomer money. It is wealthy boomer money concentrating into the largest names because that is the safest-feeling vehicle for fear capital.

So the bifurcation has a spatial shape. Mag 4 and Oracle, names a 65-year-old in Connecticut understands without thinking, absorb the fear-capital bid on top of the mechanical bid. ServiceNow, Salesforce, Adobe, Intuit, Workday don’t. These are companies an active institutional manager knows and a retired boomer in a taxable account has never heard of. The mechanical bid still gets there through index weight. The fear-capital bid doesn’t.

The hiring data on this cohort is direct. The hiring rate for workers 55 and over is up 84 percent year over year. The hiring rate for workers 29 and under is down 25 percent. This pattern has only one historical precedent. The early stages of the industrial revolution in the 1830s, when automation destroyed the apprenticeship system and created highly cyclical youth unemployment. The Panic of 1837 produced a major employment crisis in New York City. Contemporary estimates from the Society for Improving the Condition of the Poor put unemployment among manual labor at approximately 25 percent in the winter following the panic, rising to roughly one-third by the second winter. He cited a higher figure on his recent Thoughtful Money appearance. Whichever measure one uses, the structural point holds. The early industrial revolution produced cyclically severe youth unemployment that has no clean modern analog. He is careful to say the current cycle is unlikely to reach those extremes. He is less careful to suggest the underlying structural dynamic is similar.

The symmetric pattern is the speculation behavior at the other end of the cohort distribution. Retail trading is now 17 percent of total US equity volume, up from 10 percent in 2019. Zero-day-to-expiry options volume has exploded. Leveraged ETFs hold record AUM. Meme stocks and crypto remain dominant attention sinks. This is not a coincidence. The cohort that has been priced out of housing, household formation, and conventional asset accumulation has rationally concluded that incremental conservative saving is mathematically insufficient to bridge the gap. The only way to generate the return required for parity with the prior generation is to gamble. Robinhood is the structural mirror of Vanguard. Different cohort, same anxiety, opposite tail of the response distribution.

The trade implication is structural. When the demographic curve reverses, projected by Federal Reserve cohort data and Vanguard’s How America Saves report for the early 2030s but potentially accelerated by labor market shocks that emerge sooner, the spatial signature of the mechanical bid reverses with it. The asset classes that have been starved by the concentration into cap-weighted equities are real assets, grid infrastructure, productive land, and quality businesses at trough multiples without index inclusion. These benefit asymmetrically when the bid rotation happens. The mechanical reversal does not destroy the index. It redistributes the flow.

We have published the deep work on the infrastructure thesis in our Datacenter Bible, which establishes why grid electrification is the binding constraint on the next phase of capital allocation. We have published the named beneficiary work on Hubbell, the US grid monopolist with 90 percent FCF conversion and a regulatory moat that compounds for the next decade. Those pieces stand on their own merits and have their own theses. They are also positioned to absorb the eventual structural rotation the framework predicts. Two frameworks. Same trade.

Section VII. The Trade Construction

Three trades.

Trade one. Long the structural reversal. Real assets, infrastructure, grid electrification, productive land. The classes starved by cap-weighted concentration absorb the eventual rotation. Multi-year. Entry prices reflect neglect, not enthusiasm. PAVE, copper, Hubbell, Wesdome, Ivanhoe, the rest of the real-asset complex. That’s the structural call.

Trade two. Avoid the dead zone. Names with neither structural moat nor flow protection. Small and mid-cap businesses outside the S&P 500 and outside the major sector ETFs, with deteriorating fundamentals and customer bases exposed to private-credit-funded mid-market stress. These face downside in both regimes. If the mechanical bid continues to skip them, they compress further on neglect. If the bid reverses, the broader risk-off environment punishes them disproportionately. The dead zone is the worst-positioned segment of the US equity market. Our coverage of the private credit cycle has identified the specific transmission mechanisms. The next piece in that series updates the positioning.

Trade three. Hedge the index directly. The 65 percent threshold timeline maps to long-dated index puts. SPY or SPX puts struck 20 to 30 percent below current, expiring 24 to 36 months out where available. LEAPS liquidity falls off past 24 months, so size to roll once into the breach window. Premium is the cost of being early. Convexity is the payoff if the threshold breaks on the central case. This isn’t a one-off trade. It’s the systemic hedge every diversified book holding meaningful long equity exposure should carry for the next 30 months.

A critical note on the bifurcated software cohort documented in Section IV. These names trade at trough multiples that imply significant compression has already occurred. Whether they belong in the long book at current levels depends on a name-by-name examination of fundamental moats, AI displacement risk, customer-base credit exposure, and forward FCF trajectory. We address each of these names individually in the upcoming Private Credit cycle piece next week, which updates our framework on the mega-cap software complex and provides the specific position-level recalibration that Section IV makes implicit but does not name directly.

For now, the structural trades are the priority. Real assets long. Dead zone short or absent. Long-dated index puts as the systemic hedge. The single-name software work follows next week.

Before we close, the bear case for our own bear case. Five attacks. We engage each one.

Section VII.5. The Strongest Case Against Us

Five attacks on the framework. Strongest first. We engage each one directly.

One. Sammon and Chinco imply the threshold is already crossed.

Marco Sammon and Alex Chinco found that conventional passive ownership measurement understates the true passive footprint by approximately a factor of two. The conventional measure today is 53 percent. The corrected measure is therefore between 65 and 75 percent. By Krishnan and Sturm’s own logic, the threshold has already been breached. The cascade should already be operating at the aggregate level. It is not.

This is the strongest attack and we’re leading with it. Our position is the threshold scales with the measurement. If true passive share is 2x the conventional number, the operative threshold is also roughly 2x Krishnan and Sturm’s value, putting the breach point near 80 to 90 percent on the conventional measure. We can’t prove that. Krishnan and Sturm don’t model the relationship between measurement and threshold. We’re extrapolating.

If that’s unsatisfying, fair. The framework needs updating. We’ll come back to it.

Two. The XIV mechanism does not transfer cleanly to the S&P 500.

XIV was short volatility through VIX futures. The cascade ran because falling positions had to cover by buying VIX futures, which raised VIX, which forced more covering. Single mechanism, single product, single counterparty.

The S&P 500 mechanical bid is long equities, not short volatility. The cascade pathway, if it exists, has to operate through different channels. Vol-targeting funds reducing exposure as realised volatility rises. CTAs flipping short on trend break. TDF threshold rebalancing when equity-bond ratios breach the band. These are real channels. They are not the XIV channel.

We acknowledge the analog is rhetorical more than mechanical. The structural similarity is the dominance of mechanical flow in setting the marginal price. The structural difference is the cascade pathway. A reader who accepts the dominance argument but rejects the analog is making a defensible move. Our position is that the alternative cascade pathways are sufficient to produce a comparable volatility event, but we cannot specify the exact mechanism in advance. The closed-form math does not require a single pathway. It requires only that discretionary capital be insufficient to absorb the volatility signature. That condition can be satisfied through any combination of the channels above.

Three. Green has been calling this event for years. The thesis has not played out.

Green has been identifying the structural problem for several years. Passive share has continued to rise over that period. The cascade he predicts has not occurred. Either the threshold framework is non-linear and the breach is closer than the linear gradient implies, or the timing is much harder to call than the analog suggests.

We respect Green’s framework and acknowledge the timing question is the cleanest attack on it. Where we differ from Green: he’ll commit to timeline estimates. We commit to the structural call and treat the timing as a distribution. Central case 30 to 36 months. Plausible window 21 to 51 months. The framework doesn’t tighten further without new information.

Anyone treating this as a timing call is misusing the framework. The same critique applies to prior work. The XIV call was right on direction, wrong on speed. He said two years. It went six months. The structural prediction held. The timing dispersed. We expect the same pattern here.

Four. The Fed put.

XIV had no lender of last resort. The S&P has the Federal Reserve. Section 13(3), the March 2020 SLR exemption, the implicit equity backstop. A volatility event that threatens financial stability produces intervention before the cascade runs its full course.

We accept the Fed put exists. We do not accept that it eliminates the volatility event. The Fed cannot prevent the initial repricing. It can only contain the aftermath. Holders mark down materially before the intervention arrives. The March 2020 episode saw the S&P fall 34 percent in 23 sessions before Fed interventions stabilised the market. A passive-driven cascade with deeper structural roots would likely produce a steeper initial leg before stabilisation.

The Fed put compresses the drawdown duration. It does not compress the drawdown depth. Long-dated puts struck 20 to 30 percent below current still pay out under any scenario where the Fed intervenes after a 30 to 50 percent initial repricing. That is the central case we are positioning for.

Five. The aggregate threshold framework contradicts our own catalyst-triggered cascade.

If Section IV.5 is correct and the cascade is already operating at the name level, the 65 percent aggregate threshold is descriptive, not predictive. By the time it breaks, the cascade has already destroyed substantial value in MSFT, META, and the next names that meet the four conditions. The investor who waits for the aggregate breach is too late.

We accept this internal tension. The two readings are sequential rather than contradictory. The name-level cascade is the front end of the distribution. The aggregate threshold is the back end. The piece is making both predictions because they are both operating at different levels of the system. The aggregate framework predicts the moment the system runs out of un-broken names to absorb selling pressure. The name-level framework predicts which names break first and why. Both are useful. Neither replaces the other.

Where we land.

The strongest attacks on the thesis are Sammon and Chinco on measurement and the XIV mechanism non-transfer. The first says the threshold is already crossed. The second says the cascade pathway is different from the analog. Together they argue the framework is either premature or wrong about the specific mechanism.

We are positioning anyway. The asymmetry justifies the position. The cost of being early on long-dated puts is the premium. The cost of being right on the real asset rotation is approximately zero because those positions earn carry and compound during the wait. The cost of avoiding the dead zone is approximately zero plus the opportunity cost of capital that would have been deployed there anyway.

The thesis can be wrong on mechanism, wrong on timing, and partially wrong on threshold, and the trade construction still works. That is the asymmetry we are willing to pay for.

Section VIII. Closer

The mechanical bid that took the S&P from April 7 low to 7,391.75 was the largest ever recorded. Not earnings. Not the war. Not the Fed.

A payroll system. A trend model. A vol algo. And a wealthy minority of frightened boomers feeding fear capital into the biggest cap-weighted index products they recognise.

The bid does not allocate evenly. Four of the Mag 7 are up YTD. Two of the remaining three are down double digits. The third is down 8 percent. The largest mega-cap software cohort trades at 60 percent discounts to its own five-year averages. The bifurcation is the spatial signature of where the discretionary fear-capital bid lands on top of the mechanical base.

Green called XIV at 95 percent probability of going to zero within two years. It went in six. Holders lost 96 percent of their capital in a session.

He’s making a bigger call now. The same math that destroyed a $1.9 billion product exists today for a $73 trillion market. Threshold 65 percent passive AUM share. Current share 53 percent. Growth 4 points a year.

Roughly 30 months on the clock.

Position structurally. Hedge systemically. Avoid the dead zone. Pay attention to the spatial signature.

The casino is going full bore. The dealer is dropping cocaine on the tables. The math says the dealer cleans up at the end of the night.

It always does.

Subscribe to TSCS.

Subscribe now

Share

Research and opinion, not investment advice. Nothing here is personalised to your mandate, tax position, or risk tolerance. Positions may be held in any names mentioned. Past calls do not predict future performance. Do your own work. Consult a licensed adviser where it matters.

The single way this piece is wrong is being stress-tested on the tape as I write. The post-2024 correlation flip was rate-cut-narrative dependent, not structural. The May 12 to 14 inflation prints took the rate-cut path off the menu. Today the textbook mechanism is reasserting in full. If that read holds for the next twelve months, silver mean-reverts toward $50 to $60 and the cohort takes a 25 to 40% drawdown with it. The structural deficit, the COMEX warehouse drawdown, the Shanghai premium, the convenience yield gap, all of it explains the reset from $25 to current levels. None of it explained why silver should hold $80 plus with real rates at 2% and no path to lower. Today’s tape proved that. The question now is whether the structural floor holds at $70 or breaks toward $50 to $60. Every other kill switch in this piece is downstream of that question. I am holding because the physical decoupling is real and material. If today is the start of the structural floor breaking too, the whole construction breaks at once.

I called the squeeze in “Two Silver Curves, 200 Basis Points Apart.” The 200bp convenience yield gap, the COMEX warehouse drawdown, the Shanghai premium, the physical decoupling from paper. All of it. Still running as of yesterday’s close. None of it priced in sell-side models, which are still carrying silver at $40 to $50 for 2026 against realised YTD averages in the mid-$80s. The structural setup did not break in the last six hours. The price did.

What is not working is the basket trade in the equity. SIL captured 1x silver beta on the May rally. Not 2x. Not 3x. One. Everyone is paying for torque the basket cannot deliver because the basket got gutted in eighteen months. Four primary silver names taken out by M&A. Top two holdings are a third of the fund. The lost torque people want in this cycle is not in SIL. It might be in a handful of curated names. That is the work.

This piece is that work. Three names I am holding. One name I covered, sized, and dropped after a security event in Sinaloa, with the rule discipline that decision required. Kill switches, position sizing, NPV math anchored at the May 13 peak deck ($89 silver, $4,698 gold). Live spot is 14% below that. Caveats explicit in the methodology section. If you cannot take a name-level view and update on each catalyst date, the alternative is PSLV or SIVR at 1x beta with no operator risk. That is a defensible construction. It is not the construction in this piece.

Below the paywall: the three names, the math, the rules I am running them against, and the evidence that says this sleeve is one concentrated silver equity bet expressed across three positions, not three independent ones. The basket is dead. The work is what is left.

TLDR

The basket bet on silver miners is dead. SIL captured 1x silver beta on the May rally, not the 2 to 3x retail commentary keeps quoting. Four primary silver names were taken out by M&A in eighteen months. Top two SIL holdings are a third of the fund. What is left is a curation problem.

Three names form the active sleeve. Skeena (SKE) as gold-carried developer where silver is a 9.5 Moz free option the market is not pricing, entry around C$44 / US$32. Hecla (HL) as the only debt-free US and Canada silver producer, 6.0% FCF yield at spot, entry around $18. Pan American (PAAS) as Juanicipio embedded value plus Escobal optionality, smallest position because the stock trades at roughly 1.52x consensus NAV, entry around C$78 / US$57.

Vizsla (VZLA) is covered, not recommended. The January 23 abduction at Panuco (nine confirmed dead) triggered my pre-set security kill switch. Re-entering now would soften the rule retrospectively. I am not.

The real rates correlation flipped. Pre-2024 silver-TIPS correlation was minus 0.62. Post-2024 it is plus 0.48. The trade is a bet the inversion sticks. If the historical mechanism reasserts at 1.99% real rates, silver mean-reverts to $50 to $60 and the cohort goes with it.

The sleeve is meaningful but bounded. Daily correlation across the three names exceeded 0.85 for two weeks during the January crash. This is not a diversified basket. It is one concentrated silver equity bet expressed across three positions. Hecla core, Skeena secondary, PAAS smallest.

The basket got worse, faster than the beta math suggests

One fact about this sleeve before going to names. Across the three active positions (Hecla, Skeena, Pan American), daily correlation during the late-January to early-February silver crash exceeded 0.85 for two consecutive weeks. This is not three diversified positions. It is one concentrated silver equity bet expressed across three names. The benefit of holding three rather than one is reduced name-specific tail risk (one Mexican regulatory event, one grade reconciliation surprise), not portfolio diversification. Total sleeve size should reflect that. My own sizing landed on meaningful but bounded, with Hecla as core, Skeena secondary, Pan American smallest.

Run Hecla and Skeena only if you weight tail risk more heavily. Push higher if you accept the correlated bet. A Kelly or Sharpe-optimized construction concentrates into Hecla and trims Pan American.

The three-name structure trades quality-adjusted returns for cycle coverage and Escobal optionality.

Now the basket. Everyone is watching silver versus SIL beta. The 0.95 to 1.2x basket capture is the cleanest evidence that the gold miner 2009 to 2011 analog binds: rising AISC, declining grades, consensus models lagging spot, byproduct credit dependency. The standard interpretation is that miners are a worse trade than the metal at any sizing.

SilverCrest (Coeur, Feb 2025), Gatos (First Majestic, Jan 2025), MAG (Pan American, Sep 2025), and Dolly Varden (Contango, Mar 2026) all left the basket via acquisition. Discovery Silver pivoted to gold by acquiring Newmont’s Porcupine complex in April 2025. The companies left in SIL did not generate the 0.95 to 1.2x figure. Per the Global X N-CSR filing as of October 31, 2025, the top two holdings (Wheaton Precious Metals and Pan American Silver) account for approximately 34% of fund net assets, down marginally from 36.6% in April 2025 as the broader basket rallied. PAAS weight oscillated between 11% and 13%, not monotonic expansion. The structural point holds. Top-heavy basket concentrated in two names, more concentrated than the historical SIL profile.

The drift shows up in the disappearance of names that used to dilute it. Four primary-silver names gone via M&A in eighteen months, plus the Discovery pivot to gold. The basket bet does not just deliver disappointing beta. It delivers it from a portfolio that has been gutted. Hickey has been flagging this acquisition-driven thinning in successive High-Tech Strategist letters across 2022 to 2026, and his framework converges on curation as the value-add. The curation argument does not depend on Hickey. The M&A wave is public record. The conclusion follows from the data.

Three names made it through my framework as the active sleeve. A fourth (Vizsla) I cover but do not recommend. The January 23 abduction at Panuco (nine confirmed dead) triggered my pre-set kill switch. Reinstating after the event would soften the rule retrospectively. I am not doing that. Full reasoning in the Vizsla section.

Skeena as gold-carried developer where silver is a 9.5 Moz free option the market is not pricing. Hecla as the only debt-free US and Canada silver producer, 6.4% FCF yield at spot. Pan American as Juanicipio embedded value plus Escobal optionality, against the rich consensus NAV multiple. All three carry gold revenue. That is intentional. The pure-play primary silver name in a tier-1 jurisdiction with operational reliability is extinct as a category.

On Mexican country risk. Mexico accounts for approximately 24% of global mine-site silver supply per the Silver Institute’s World Silver Survey 2025, on national production of roughly 186 to 200 Moz against 819.7 Moz of global output. Mexico also accounts for roughly half of dedicated primary silver mine output. The 28% figure sometimes cited in silver coverage is the primary silver share of total global silver production, not Mexico’s country share. Both are large, both matter, neither is interchangeable. CAMIMEX has reported sector-wide private mining security spending in the area of $100 to $110 million for 2025, with industry sources citing security costs running up to roughly 5% of operating expenses and material losses from theft or criminal assault running up to 7% of operational costs at affected operations. Press coverage puts the protection-fee rate at approximately 3% of revenue. Milenio reporting puts Sinaloa homicides at approximately 1,654 in 2025, the most violent year in over a decade. Treat as directional, not audited.

The Vizsla incident is the most visible recent event. It is not isolated. The Mexican jurisdictional discount applies more aggressively than I was modeling. The PAAS sizing reflects that.

Methodology, sourced upfront

What I am relying on, what I am modeling against, where the data comes from. The caveats apply throughout.

Silver at $76.44 intraday May 15 after a -8.48% single-session move from yesterday’s $83.52 close. May 11 close was $75, May 13 peak was $89, May 14 close was $83.52. Silver has given back the entire May rally in two sessions. Down from the $121.67 high on January 29.

Gold at $4,546. NPV math in the name sections below uses the May 13 peak deck ($89 silver / $4,698 gold) as the analytical anchor. At current $76.44 / $4,547 spot, dollar figures scale roughly 14% lower; structural conclusions unchanged. Scenario distribution from “Two Silver Curves, 200 Basis Points Apart” (40% rebase higher, 35% consolidation, 20% retest of $55 to $70, 5% retest above $120). Realized path sits inside the rebase range. I am not unilaterally revising the published probabilities.

Cohort prices, intraday May 15: Hecla (HL) $17.78 NYSE, Skeena (SKE) C$43 / US$31.17, Pan American (PAAS) C$78 TSX / $56.59 NYSE, Vizsla (VZLA) C$4.75 / approximately $3.44 USD. From the May 13 peak to current: silver fell 14%, Hecla fell 16%, PAAS fell 12%, Skeena fell 8%. The down-cycle beta hierarchy held directionally: HL 1.13x, PAAS 0.85x, SKE 0.58x. Lower silver content, more share-price protection, exactly as the gold-carry thesis predicts. From yesterday’s May 14 close alone: silver -8.48%, HL -8.61%, PAAS -7.39%, SKE -8.17%, VZLA -8.33%.

Today’s single-session panic compressed the cohort to roughly 1x silver beta across the board.

Five things to flag upfront.

Q1 2026 prints are not steady state. Hecla’s Greens Creek AISC of negative $8.39 and Pan American’s silver-segment AISC of $6.63 are byproduct credit artifacts. Strip the gold, zinc, and lead credits and both move up by $15 to $25 per ounce. I model on full-year 2026 guidance midpoints. Q1 figures appear as markers of where the byproduct stack currently sits, not as forward estimates.

Sell-side models embed silver below spot. Bloomberg Terminal consensus aggregation in late April had the cohort modeling silver at $40 to $50 for 2026 against realized averages closer to $90. Company-level planning is higher: Pan American’s 2026 outlook uses $70 silver and $4,200 gold, Hecla’s sensitivity tables centre on $75 and $100 cases. The catch-up trade lives in sell-side multiples. Company guidance has already caught up. Catch-up is the load-bearing assumption in the PAAS bull case. If it does not happen, PAAS breaks first.

AISC is not standardized. Allocation methods differ, some companies include corporate G&A, some attribute at economic interest versus full ownership. Cross-name AISC comparison is hard. I cite consolidated guidance midpoints where they exist and flag where I am estimating.

NPV extrapolation is methodologically heavy at current spot. Skeena’s DFS sensitivity envelope tops out at $2,200 gold and $27 silver. Spot is 2.6x and 3.3x outside that. MAG’s Technical Report tops at $1,750 and $22, with spot 2.7x and 4.0x outside. Vizsla’s FS sensitivity tops at +50% from basecase, with spot 2.5x or more outside. Linear extrapolation does not capture cost inflation at higher prices, tax stack changes, royalty step-ups, or grade-recovery non-linearities. I run an optimistic ceiling figure and apply a 25 to 35% conservatism discount for the central case. The central case is the anchor figure. The optimistic ceiling is the upside band. November 2026 (Skeena) and Q2 (PAAS) updates will replace these extrapolations with company-modelled economics at updated metal-price decks.

Bloomberg Terminal is the institutional standard for sell-side consensus aggregation. Where I cite a 1.55x consensus NAV multiple or an 8-analyst sample, the figures are terminal-verified. Macro framing (real rates correlation, gold-silver ratio) uses annual averages from FRED and LBMA. Monthly Bloomberg data tightens magnitudes. Direction holds either way.

I do not disclose position details in published research. Treat the analysis as independent of any position I may hold.

Why the gold analog binds partially

The 2009 to 2011 gold miner analog binds, less tightly than the headline suggests.

During the 2009 to 2011 gold bull market, the HUI Gold Bugs Index delivered approximately 1x gold’s return. The 3 to 8x operating leverage that simple cost-spread arithmetic predicted did not materialize. Costs ate part of the leverage, grades ate more, consensus lagged throughout. The same mechanisms operate in silver miners 2024 to 2026.

SIL’s rolling 1-year beta to spot silver has run between 1.2x and 1.8x across post-2010 history, not the 2x to 3x figure that shows up in retail commentary. The “lost torque” framing in this piece is most accurate against the upper end of that band (1.6x to 1.8x sustained beta), not against an aggressive 2x to 3x baseline. The basket-level compression toward 1.0x holds either way. The magnitude of the compression is smaller than the 2x to 3x baseline implies. The defensible version: the basket used to deliver 1.5x to 1.8x and now delivers 1.0x to 1.2x. The aggressive framing does not survive audit.

The analog binds less tightly than the historical baseline for three reasons.

First, the magnitude of the silver price move (from $25 in early 2024 to a peak above $120 in January 2026) is larger in percentage terms than the gold move in 2009 to 2011. Margin expansion is genuinely greater. The catch-up trade has more room because the gap between modeled silver price and realized silver price is wider for most names than the comparable gap was for gold miners in 2009 to 2011.

Second, the silver miner universe consolidated. Four primary-silver names gone via M&A in eighteen months. SIL is now roughly 22% Wheaton plus a much shorter list of primary producers. The basket effect that diluted gold miner outperformance in 2009 to 2011 (where hundreds of names were available) is meaningfully reduced. Concentrated positions in 4 names out of an 8-name primary silver universe carry much higher tracking error to the basket than concentrated positions in 4 of 80 names.

Third, the structural physical scarcity thesis from the prior piece (200 basis point convenience yield gap, 312 Moz COMEX warehouse drawdown, Shanghai premium) is more acute than the 2009 to 2011 setup, which was primarily an investment demand story. If physical scarcity sustains silver above $100, the equity catch-up trade gets sharper because consensus models embed silver well below current spot.

None of the three is a guarantee. They are arguments for why the analog binds less tightly. The base case remains that silver miners deliver close to 1x silver beta over the next twelve months. Alpha, if it exists, is in concentrated names that score on consensus model lag, operational reliability, and jurisdictional discount.

The three active names in this piece are my best attempt to identify that alpha. Skeena through embedded silver in a gold-economic project. Hecla through jurisdictional safety in a debt-free portfolio. Pan American through Juanicipio cash flow translation. I could be wrong on any of these. Most likely on Pan American, because Hickey publicly disagrees and his track record is meaningfully better than mine, and because Mexican country risk has now demonstrated itself as something more material than a tail event in pricing. Most likely right on Skeena, because the project economics are gold-carried and the silver contribution is option-like, and because British Columbia is structurally insulated from Mexican-style risks. Hecla sits between: the multiple is not contrarian but the jurisdictional cleanliness is rare in this cohort and worth its premium.

At $89 silver and $25 AISC, the typical mid-cost silver producer generates $64 per ounce of gross margin. That margin was zero eighteen months ago. The expansion is real. It has not flowed through to equity returns at the basket level because of the four headwinds the prior piece walked through: rising costs, unstable production, lagging consensus models, equity discount for jurisdictional risk.

The trade I am recommending requires three things to hold per name. Consensus 2026 EPS must embed silver meaningfully below spot (most do). Operational profile has to be stable enough that margin expansion actually flows through. Jurisdictional discount has to be either appropriately priced or actively narrowing. Names that score on all three are the trade. Names that look cheap on operating leverage but fail on country risk or operational reliability are value traps.

Why three names and not a more concentrated two-name (Skeena + Hecla) bet? Two answers.

Skeena is pre-revenue. The position works structurally but generates no cash flow until Q3 2027. A 100% allocation to pre-revenue through 2026 misses the silver miner income Pan American and Hecla deliver.

Pan American’s Juanicipio embedded value is real even if not as cheap as it looks. Q1 2026 attributable AISC of negative $3.05 per ounce is not permanent, but the underlying mine economics generate $400 to $450 million per year of attributable margin at current metal prices. Pan American paid roughly $2.042 billion in announcement-date consideration. Undiscounted payback is approximately five years. On a strict 5% NPV basis with a 13-year LOM, forward attributable NPV at current spot is approximately $1.6 to $1.9 billion, modestly below the $2.1 billion cost. The asset is strongly accretive on a payback basis and marginally underwater on NPV. Worth taking. Not the bargain the headline economics imply.

The three-name structure trades concentration for cycle coverage. A two-name structure would generate higher quality-adjusted returns through a Scenario 3 but miss the operating leverage in PAAS. A reader who weights Scenario 3 above 25% should run Skeena and Hecla only.

Real rates and the gold-silver ratio

The mechanism that should be working is not working. Silver has historically traded inversely to 10-year real yields, rolling correlation of minus 0.5 to minus 0.7. The classical mechanism: silver yields nothing, real rates are the opportunity cost, when rates rise silver falls. The 22-year history is clean enough to anchor a position.

Through 2024 to 2026 the correlation flipped. Pre-2024 silver-TIPS correlation runs minus 0.62. Post-2024, plus 0.48. They move together now. The trade in this piece is a bet that the inversion sticks. If the historical mechanism reasserts with real rates at 1.99%, silver mean-reverts to $50 to $60 and the cohort goes with it.

Sell-side consensus NAV models embed silver at $22 to $40 per ounce per Bloomberg Terminal aggregation as of late April. Spot is $76 today. Sell-side 2026 earnings decks separately sit at $40 to $50. Company-level planning is closer to spot: Pan American’s 2026 outlook assumes $70 silver and $4,200 gold, and Hecla’s published sensitivity tables centre on $75 with a $100 case. The catch-up trade is narrower than the stale-deck framing suggests. The mispricing is in sell-side multiples. Companies have already updated. The wider question is whether either anchor survives if the correlation reasserts.

The gold-silver ratio compressed from 85 in 2024 to 51 YTD, 59 at spot. The long-run median is 69. Silver caught up to gold in eighteen months. Below 50 has marked silver tops twice: Hunt 1980, 2011 squeeze. Two data points is not a signal. Tactical context. Whichever framing you bring (rich versus gold, or new regime), the easy phase of the catch-up is priced in.

Below the paywall: NPV math at peak deck and current spot for each name, position sizing across the three-name sleeve, named kill switches per position with the exact triggers that exit the trade, entry levels, the Pretium Brucejack case study every BC developer holder should read before Skeena commercial production, the Vizsla rule discipline and what it cost, scenario analysis on what breaks first in a $50 silver retest, and the full Q2 2026 to Q3 2027 catalyst calendar.

Silver crashed 8.48% in a single session today. The Q2 print cycle starts in eight weeks. The basket is dead. The work is what is left.

Subscribe to read the rest.

Where consensus disagrees, name by name

Read more

The lease desks at JPMorgan, HSBC, and ICBC Standard disagree. So does the COMEX delivery floor. So does the Shanghai Gold Exchange. Three of these four signals are run by people who actually move physical metal. The fourth runs on screens. Pick which set you’d trust to call the next squeeze.

Thursday January 29, 2026: spot silver printed an intraday peak of $121.67 and every silver bull on Twitter felt vindicated. Roughly thirty hours later, COMEX silver settled at $78.53. Worst single session since March 1980. $70 to $80 billion in above-ground silver value erased. Every leveraged long who hadn’t trimmed by Thursday afternoon got margin-called by Friday’s open. The consensus moved on inside two weeks: solar demand softening, China export rules tighter on paper than in practice, recession pricing in, structural thesis on hold.

The structural thesis is intact. The trade has changed.

Two markets are looking at the same metal for the same forward window, on the same Bloomberg session, pricing two different stories. The London silver lease market says forward physical is scarce. The COMEX futures curve says it isn’t. The gap is 200 basis points and has held for six consecutive months. Two other corroborating signals (COMEX warehouse stocks down 41% from the October 2025 peak to below the 5Y average, Shanghai Gold Exchange silver at a 10 to 15% premium to London) line up with the lease market. The futures curve, alone, doesn’t.

Working out which is wrong, and what trade follows, is what this piece does. The basket trade in silver miners is structurally weak. Asymmetric exposure has shifted to specific names. The next piece is the names.

800 Million Ounces Short is the long-form structural case. For the mechanical anatomy of the January crash itself, see The 31% Silver Crash: Why Markets Got It Wrong. This stands alone.

Subscribe now

The names are paid-only. Specific primary producers ranked by silver-revenue exposure and forward AISC, royalty and streaming names worth holding, plus a small set of developers and juniors flagged as either underpriced or structurally short. The framework above narrows "buy silver miners" to "buy these specific silver miners." The next piece is the names. Drops next week.

I. The Structural Case, Compressed

Silver has run a supply deficit every year since 2021. WSS 2026 (April 15, 2026): 2025 deficit at 40.3 Moz, fifth consecutive, 2026 projected at 46.3 Moz. Cumulative drawdown from above-ground stocks since 2021: 762 Moz. London market free float at end-September 2025: roughly 136 Moz, a record low against 2025 industrial silver demand of 657.4 Moz and total global demand of 1.13 billion ounces. London free float has rebuilt to approximately 28% of vault holdings of 884 Moz by end-March 2026, up from 17% at the September trough, as metal flowed back from US warehouses. System less impaired than at the October peak. Still tight against the structural backdrop.

Industrial demand composition has shifted from cyclical to structural (solar, semiconductor packaging, AI infrastructure, EVs, grid modernization). The headline trajectory has weakened. WSS 2026: industrial fabrication declined 3% in 2025 to 657.4 Moz, the first post-pandemic decline, with PV silver demand down 6% to 186.6 Moz on thrifting acceleration. 2026 forecast: industrial demand declines further, PV silver demand projected down 19% to 151 Moz as module manufacturers accelerate copper substitution and reduce silver loadings per cell. Structural composition argument operates on 3-to-5 year horizon. The 12-month industrial demand trajectory sits softer than the headline bull case implies. Any positioning that requires industrial demand to grow in the next twelve months is taking substitution risk that WSS 2026 has put into the official base case. The consensus bull missed this. WSS 2026 made them deal with it.

Normal commodity self-correction is broken because 70 to 75% of mined silver is byproduct of copper, lead, zinc, and gold operations whose capital allocation doesn’t respond to the silver price. Global silver mine production rose to 846.6 Moz in 2025 and is forecast broadly flat in 2026 per WSS 2026. Demand is outrunning supply. Supply itself isn’t collapsing.

The US designated silver a critical mineral in November 2025. China updated its silver export framework effective January 1, 2026, replacing the prior quota system with a licensing regime under which 44 approved firms (up from 42) require government approval per shipment, with new entrants needing 80+ tonnes annual production capacity. Licensing layer genuinely new. Exporter list expanded modestly. License reviews can run 45 days or longer, which adds friction to flows. 2025 Chinese silver exports reached a 16-year high of approximately 5,100 tonnes, indicating pre-2026 flows were elevated rather than constrained. Treating China’s licensing regime as the supply shock that prints silver to $200 is the kind of lazy bullshit that fills retail Substacks. Treating it as nothing is also wrong. The honest read: friction is real, hard quotas don’t exist, and the question is whether 45-day reviews start to bite as 2026 progresses.

Setup going into 2025. Metal opened the year at $28.92. Broke $50 in October during a London market freeze where one-month lease rates hit 34.9% on Bloomberg with intraday prints to 39.2% reported by Bruce Ikemizu of the JBMA. Printed $80 in December. Reached its all-time intraday high above $121 on January 29. Then one of the most leveraged macro trades of the cycle detonated on January 30, and silver lost a third of its value in roughly thirty hours.

The 2026 LBMA Forecast Survey collected silver forecasts from 26 to 31 analysts. Range runs from $44 at the low end (TD Securities) to $165+ at the high end (Metals Daily / LBMA bull respondent), consensus around $79.57. J.P. Morgan at $81 average, Commerzbank at $90, Metals Focus at $60. A separate Bank of America bull scenario published in late January reaches $135 to $309 if physical deficits intensify sharply and the gold-silver ratio compresses to 1980 levels. Conditional tail, not a base case, but worth noting as the ceiling of professional thinking about what’s mechanically possible. A 3.75x range across base case forecasts ($44 to $165), before considering tail scenarios. The market isn’t disagreeing about facts. It’s genuinely uncertain about how a structurally broken commodity reprices in real time. That uncertainty is the trade.

The structural case above is consensus bull narrative going back eighteen months. You’ve read it ten times in different wrappers. Not original. What follows is.

II. The 30-Hour Round Trip And Its Aftershock

The EFP spread (front-month COMEX silver minus London spot) is the cleanest single number for measuring stress in the paper-physical relationship. Normal conditions: within a few cents either direction, reflecting small cost-of-carry differentials between New York paper and London bullion. Five-year mean: negative seventy-seven cents. Five-year standard deviation: $4.21.

I pulled the daily EFP series on Bloomberg from January 20 to February 10, 2026. The sequence:

January 23: spread closed at negative $1.86, inside the normal range.

January 26: spread blew out to positive $11.72. Front-month COMEX futures traded nearly twelve dollars above London spot. 2.96 standard deviations above the five-year mean. Panicked paper longs paying any price to maintain futures exposure as short squeeze dynamics accelerated. Physical settlement frictions prevented London spot from catching up at the same pace.

January 27 to 29: spread compressed to negative $6.13, negative $3.16, and negative $1.27 as silver itself ripped higher and the futures-physical gap closed.

January 30: Warsh nomination hit, the leveraged long book detonated under margin call pressure, EFP printed negative $6.67. Same metric, opposite direction. 1.40 standard deviations below the five-year mean.

In four trading days, the spread traveled from +$11.72 to -$6.67. An eighteen-dollar swing in a metric whose normal range is plus or minus a few dollars.

Thursday February 5, four trading days after the crash bottom, changes the diagnosis. A single extreme reading on January 26 followed by a single offsetting reading on January 30 is consistent with a one-shot squeeze and unwind. A third extreme reading four trading days after the unwind looks like a market still finding equilibrium across paper and physical pools.

The EFP printed positive $5.80 that day. A second statistically extreme reading, opposite direction from January 30. By February 10 the spread had returned to negative $0.43.

Translation: silver market microstructure broke down repeatedly during stress rather than in a single event that cleanly resolved. Mechanical reading: January 26 was paper longs paying a $12 per ounce premium over London spot to maintain futures exposure during the parabolic move. January 30 was physical London trading at a $6.67 per ounce premium over collapsing futures during the crash. February 5 was a second round of paper-versus-physical dislocation as remaining positioning unwound. Three readings, all statistically extreme, all in a two-week window.

During the crash, physical silver repriced less violently than paper. Spot silver fell from $115.70 on January 29 to $85.20 on January 30, a 26% decline. Front-month COMEX futures fell from $114.43 to $78.53, a 31% decline. Both violent. The five-percentage-point differential is physical demand absorbing some of the futures liquidation. A crash where both markets sold off equally would have meant genuine demand exhaustion. Differential pattern: leveraged-positioning event, not fundamental rerating. Anyone selling you “silver got it wrong” or “the rally is dead” wasn’t reading the EFP.

October 2025 was the dress rehearsal for January. On October 9, 2025, one-month silver lease rates hit 34.9% on Bloomberg with intraday prints to 39.2% per JBMA chief Bruce Ikemizu. Spot broke $50 for the first time since 1980. The London bullion market physically seized for ninety minutes. Lease rates measure the cost of borrowing physical silver in the bilateral lending market between bullion banks and short sellers, traders, and refiners covering obligations. A 34.9% Bloomberg session reading on a rate whose five-year average sits around 1.0% is the market saying lenders effectively stopped lending.

October drained the lendable pool, January drew the remaining speculative interest into the long side, the January 30 crash cleared the speculative excess. None of it cleared the structural damage to the physical lending market.

III. Three Independent Measurements, One Outlier

This isn’t a one-signal argument. Three measurements of physical silver scarcity, generated by structurally independent participant sets operating in different venues with different mechanics, are aligned in direction. The fourth (the COMEX silver futures curve) sits at the calm end of plausible readings. The system is less impaired than at the October peak. London free float has rebuilt from approximately 17% of vault holdings in September 2025 to approximately 28% by end-March 2026. Front-end lease rates have normalized. The EFP spread has returned to inside its normal range. The three signals identified below persist after that partial healing. Structural, not acute.

Signal one: the bilateral lending market in London. Bullion banks lend physical silver to refiners, miners, traders, and short sellers covering obligations. The lease rate they charge is the price of physical silver scarcity in the OTC market. The 12-month silver lease rate currently sits at approximately 1.6% (Bloomberg session, May 2026, TSCS internal pull), against a five-year average closer to 1.0% on the same series. 1-month rate has normalized to 0.15%. 3-month sits at 0.91%, modestly below its 1.04% 5Y average. Front of the curve has healed. Back has not. Reuters and Metals Focus described silver lease rates as “largely normalised” by mid-April 2026; that description refers to front-end normalization, not full-curve healing. Term-structure shape is the specific feature that isolates physical scarcity from acute stress. Bullion banks who lend physical for 12 months are pricing in counterparty and availability risk they weren’t pricing six months ago. The participants here are JPMorgan, HSBC, ICBC Standard, and a small number of similar bilateral desks. Institutional memory of multiple precious metals cycles. Longest-tenured precious metals risk teams in the world. They’re pricing tail risk for forward physical because October and January taught them the tail can land.

Signal two: the COMEX exchange-cleared delivery market. Total silver stocks in COMEX-approved depositories sit at 312 Moz as of May 8, 2026, down from a peak of 532 Moz on October 3, 2025. A 220 Moz drawdown in seven months. 41% decline. Current level sits below the 5Y average of 346 Moz, approaching the November 2023 cycle low of 266 Moz. The bleed isn’t paperwork. Physical silver is leaving the COMEX warehouse system entirely, not being reclassified within it. Approximately 76 to 80 million ounces of the 312 Moz total is registered (carrying warehouse warrants, immediately available for delivery), against roughly 575 million ounces of paper claims on open interest. Registered coverage ratio: 13.4%, below the 15% stress threshold defined by COMEX analysts and below threshold for six consecutive months. The eligible pool that gets warranted into deliverable supply under stress has bled from roughly 404 Moz at the October peak to roughly 234 Moz currently. The buffer is gone. January 2026: 33.45 million ounces (roughly 26% of the then-deliverable pool) disappeared from registered inventory in a single week. March 2026: 46.1 million ounces stood for delivery, approximately 60% of the then-current registered stock in a single month. Participants: institutional commodity traders, industrial end users, refiners standing for physical delivery. Different participants from the lease market. Different venue. Same direction.

Signal three: the Chinese physical market. Shanghai Gold Exchange silver benchmark traded at a premium to London spot of approximately 10 to 15% through Q1 2026 per LBMA reporting, above the post-2020 norm on any reasonable methodology basis. Methodology caveat: Chinese silver import VAT is 13%, which means VAT-inclusive landed cost premiums of 15 to 20% are routine and the signal lives in deviations from that band. SGE benchmark prices are quoted pre-VAT and a sustained ex-VAT premium above 6% sits well above post-2020 norm. Direction of the signal holds either way. The yuan has strengthened from 7.22 to 6.80 against the dollar over the past twelve months, making dollar-denominated silver roughly 6% cheaper in CNY terms. Chinese physical buyers absorbed that FX tailwind and still paid the 10 to 15% premium to London. Either alone would be a meaningful demand signal. Both together is unusual. Participants: Chinese industrial users (solar, electronics, semiconductor packaging) and Chinese physical investors. Third venue, third set of participants, same read.

The fourth measurement: the COMEX silver futures curve. Forward curve from front-month out to Dec 2030 sits in textbook contango at approximately the risk-free rate. Implied convenience yield essentially zero, the reading you’d expect in a market with comfortable forward physical availability and no scarcity premium. Clearing structure of the futures market mutualizes counterparty risk, so the curve doesn’t need to embed the bilateral risk premium the lease market does. But it should embed the underlying physical scarcity the exchange-cleared delivery market is pricing through coverage ratio stress, and that the Chinese physical market is pricing through premium. It doesn’t.

Three independent measurements of forward physical scarcity agree. One disagrees. Which is wrong.

IV. Two Convenience Yields For The Same Metal

Analytical core. Falsifiable. Arithmetic any reader with a Bloomberg terminal can replicate.

The COMEX silver forward curve and the London silver lease rate curve are pricing two different convenience yields right now. Futures market: approximately zero convenience yield. Lease market: approximately 200 basis points. Both readings from the same Bloomberg session, same day. The 200 basis point gap is the most important structural signal in silver microstructure post-squeeze. It feeds directly into the producer trade.

Arithmetic, with sensitivity acknowledged where assumptions matter, tenors deliberately matched to the lease curve so the comparison is apples to apples.

Futures curve, May 11, 2026. SI1 (May contract) at $80.40. Dec 2026 at $82.29. Dec 2027 at $85.58. Dec 2030 at $92.25. Implied annualized carry from SI1 to Dec 2026: approximately 4.0%. At the 12-month tenor specifically (the comparison point against 12M lease), implied carry is approximately 4.0% annualized, interpolating between the 7-month and 19-month points and consistent with the near-curve shape. Carry from SI1 to Dec 2030 flattens to approximately 3.0% annualized, which begins to imply meaningful positive convenience yield at the far end of the curve but isn’t the comparison that matters for this section. Near-to-medium tenor sits in textbook contango at approximately the risk-free rate.

Decomposing the futures cost of carry. Theoretical cost of carry equals financing cost plus storage cost minus convenience yield. SOFR sits at approximately 3.6% across the relevant tenors (NY Fed reference rate, May 8, 2026). LBMA-quality silver storage at major commercial vaults runs roughly 0.2 to 0.6% per year depending on whether the metal is allocated or unallocated, vault location, and institutional terms. Using the midpoint of 0.4%:

4.0% = 3.6% + 0.4% - convenience_yield_futures

Solving: convenience yield implied by the futures curve is approximately 0.0%. Storage at 0.2% (low end) implies negative 0.2% convenience yield (unsustainable in equilibrium, indicating storage assumption is too low). Storage at 0.6% implies 0.2% convenience yield. Across the reasonable range, futures market is pricing convenience yield essentially at zero.

Lease curve, same day. One-month silver lease rate (Bloomberg SOFR basis): 0.15%. Three-month: 0.91%. Twelve-month: 1.63%. Five-year averages: 1M at 1.03%, 3M at 1.04%, 12M at 1.05%.

Decomposing the lease cost. A bullion bank lending physical silver foregoes the financing cost of holding cash equivalent to the silver’s value. In equilibrium, the lease rate equals the financing cost minus the silver’s convenience yield to the lender. So:

1.63% = 3.6% - convenience_yield_lease

Solving: convenience yield implied by the lease curve at the 12-month tenor is approximately 1.97%.

Two markets, same underlying physical asset, same trading day, pricing convenience yields that differ by roughly 200 basis points. The gap is robust to reasonable variation in storage cost assumptions: even pushing storage to 0.6% (high end) and assuming the futures curve implies 0.2% convenience yield, the gap with lease-implied 1.97% stays at approximately 180 basis points. Signal is real.

Two measurements of the same underlying construct should converge in a frictionless integrated market. The convenience yield in both decompositions represents the same thing: the value to a holder of physical silver of having metal in hand against the alternative of holding cash. The 200 basis point divergence is itself the signal that the bilateral lending venue and the exchange-cleared futures venue aren’t fully integrated, and that the frictions between them are pricing structurally rather than transiently.

Two candidate frictions explain the gap. Both favor the structural argument.

First, counterparty risk in bilateral lending. The standard lease decomposition (lease rate = financing rate minus convenience yield) doesn’t separate the counterparty risk premium from underlying scarcity. The 1.97% implied convenience yield from the lease decomposition therefore overstates pure convenience yield by whatever risk premium lenders are charging, which is exactly what October 2025 and January 2026 taught them to charge. A bullion bank lending physical silver to a borrower in a stressed market faces the possibility the borrower can’t source metal to return, leaving the lender holding paper claims against a counterparty that can’t deliver. October’s 34.9% Bloomberg lease rate peak and the broader market freeze taught the bilateral lending desks at JPMorgan, HSBC, ICBC Standard, and similar counterparties that the worst case can land. True convenience yield sits below 1.97% but above zero because lenders are pricing real risk that has materialized twice in seven months.

Second, arbitrage impairment in the futures market. The cash-and-carry arb that pins forward prices to the no-arbitrage upper bound operates normally for storable silver: buy spot, store, sell forward, repay financing. That arb constrains the futures curve from above. The reverse arb, which would pull forward prices up when they sit too low, requires borrowing physical silver to short, and the lease market is exactly the venue where that borrowing has become structurally expensive at the 12M tenor. When reverse arb is impaired, forward prices can settle below their fundamental level. Near-zero implied convenience yield from the curve in that case understates the true scarcity premium because the curve mechanism that would normally lift forwards toward equilibrium is broken.

Signal is robust to which friction dominates. If counterparty risk is the bigger wedge, true convenience yield sits somewhere between 0% and 1.97%, lease curve overstates by the counterparty risk premium, futures curve is the cleaner measure of underlying scarcity. If arb impairment is the bigger wedge, true convenience yield runs closer to 1.97%, futures curve understates because of broken price discovery, lease curve is the cleaner measure. Either way the gap is the signal. The futures curve at the risk-free rate is either right because lenders are overpricing tail risk that won’t materialize, or wrong because it can’t fully price scarcity that the lending venue captures directly. The structural bull thesis survives either reading. The only reading that defeats the signal is the case where lease curve elevation is pure interest rate volatility noise rather than scarcity pricing, which Objection 1 addresses.

The clearing structure of the futures market is part of the mechanical explanation but not the full story. CME clearing mutualizes credit risk for futures participants, so the curve doesn’t need to embed the bilateral counterparty premium the lease market does. But clearing structure doesn’t explain why the futures curve fails to embed the scarcity premium that the COMEX delivery market (same exchange, same physical pool, same clearing house) is pricing through coverage ratio stress at 13.4% for six consecutive months. The clearing argument explains one frictionless slice of the divergence. Other frictions explain the rest.

Four objections worth addressing.

Objection one: the elevated 12M lease rate reflects higher interest rate volatility, not physical scarcity. Most credible counterargument. Lease rates contain an interest rate component, and interest rate volatility has been elevated through 2024-2025.

Rebuttal: if the elevation were purely a function of interest rate volatility, the entire lease curve would be elevated proportionally. Instead, the one-month lease rate sits below its five-year average while the twelve-month sits 55% above. The shape of the curve, specifically its anomalous steepness, isolates physical risk pricing from interest rate risk pricing. Interest rate volatility would lift the curve broadly with relatively normal slope. Depressed front and elevated back isn’t interest rate effects.

Objection two, the most important methodological caveat: I haven’t run a continuous decomposition of the futures-versus-lease convenience yield gap across the 2018 to 2024 window. The lease rate series requires session-by-session pulls and the work is non-trivial. I sampled approximately fifteen trading sessions across 2019, 2021, and 2023, selected to avoid acute stress windows. Convenience yield gap distribution: 30 to 80 basis points. Sample too small and non-randomized to support a percentile claim about the current 200 basis point reading. Directional claim (200 basis points is materially wider than the post-2018 normal range) survives the absence of a continuous series. Precision of the claim does not. Continuous series is on the queue for a follow-up. Until then, the 200 basis point figure should be treated as a strong informal signal rather than a statistically established anomaly.

Objection three: data series too short and we shouldn’t infer regime shift from a few months of readings. Right caution. The 12M lease rate has been elevated above 1.4% since November 2025, which is six months of sustained readings. Pattern is consistent with how bilateral lending markets price counterparty risk after stress events: front-end normalizes quickly as the acute episode discharges, back-end repairs slowly or never. I don’t have verified historical lease rate data over multiple prior cycles in a clean enough form to compute a base rate for “elevated 12M lease eventually leads to additional physical stress.” What I can say is the lease curve hasn’t signaled this pattern at this magnitude in the post-2018 period for which I have informal data.

Objection four, the one a sceptical institutional reader reaches for first after seeing Reuters: industry commentary in April 2026, citing Metals Focus, described silver lease rates as “largely normalised” since the October squeeze. This appears to contradict the central claim of this section.

Rebuttal: “largely normalised” is correct at the front of the curve. The 1-month lease rate at 0.15% sits below its 5Y average. The 3-month at 0.91% sits modestly below average. The 12-month at 1.63% sits 55% above average. “Largely normalised” refers to front-end tenors, which is also where most operational silver lending activity is concentrated. The 12-month tenor is the specific point on the curve where post-stress counterparty risk re-pricing typically persists, and the term-structure shape is the precise feature that isolates structural physical risk from acute stress. Reading “largely normalised” to mean the full curve is back to normal is the kind of headline-skim that gets you killed in fixed income.

None of these objections kills the divergence argument. They sharpen it. Lease curve elevation reflects a specific repricing of bilateral counterparty risk for forward physical silver lending. The convenience yield divergence is genuine, not an artifact of interest rate volatility or normal lease-futures noise. The 200 basis point gap reflects structural information the futures market isn’t absorbing.

That’s the signal. Translation into price action is next.

V. The Inference Chain From Signal To Trade

The convenience yield divergence is a fact about market microstructure. The trade implication requires connecting four steps. Each has uncertainty.

Step one: lease market is pricing forward physical stress the futures market isn’t. Conclusion of Section IV. Supported by arithmetic. Highest confidence step in the chain.

Step two: lease market view will prove correct over the next twelve to twenty-four months. Bilateral lending market is staffed by professional desks at large bullion banks with institutional memory of multiple precious metals cycles. Their pricing reflects considered judgment about counterparty risk and physical availability. They could still be wrong. Lenders can over-correct after stress events and price tail risk that doesn’t materialize. Medium confidence.

Step three: if the lease market view is correct, stress will materialize in spot silver pricing rather than being absorbed by the lending market alone. Most uncertain step. Lease rate stress can persist for extended periods without translating into spot price moves if (a) the stress is absorbed through wider bid-ask spreads in bilateral lending rather than through metal flow constraints, (b) physical demand is sufficiently elastic that spot adjusts smoothly without triggering futures market squeezes, or (c) policy interventions (US strategic reserve announcements, Chinese export adjustments, central bank coordinations) absorb the pressure. Medium confidence at best.

Step four: if spot silver appreciates, primary silver miners outperform the metal due to operating leverage. Well-established in mining finance theory. Demonstrated in multiple precious metals cycles. Highest-confidence step after step one. Qualification: operating leverage works most cleanly for low-cost, high-production-stability primary producers. Less cleanly for diversified producers, byproduct producers, and operationally-challenged names. Section IX breaks this down.

Four conditional probabilities on one underlying state of the world: whether the structural silver tightness thesis is correct. If it is, the probabilities cascade because each step provides evidence for the next. A real convenience yield signal means professional bilateral lending desks are correctly pricing forward physical risk. Correctly priced bilateral lending risk means the same structural scarcity the desks see will materialize through the delivery and physical-market mechanisms. Spot appreciation means operating leverage translates margin expansion into equity returns, modulated by name selection. Same event, four lenses.

The naive independent product:

P(signal real) × P(lease market correct | signal) × P(transmission to spot | lease correct) × P(miner capture | spot rises)

= 0.90 × 0.65 × 0.60 × 0.80 = 28%

That calculation treats four correlated views of one underlying event as four independent events. Loads the same source of risk (structural thesis being wrong) into every multiplicand and multiplies the result, which systematically undercounts. The independent-product method has a ceiling the structurally correct integration doesn’t.

Section VII Scenario 1 at 40% is the right integration. The four chain steps are simultaneously true if the thesis is correct, simultaneously false if it isn’t. The conditional decomposition above diagnoses where the trade can break, not where it lands. Size the bet off the scenario weight. Monitor the weakest conditional for the failure mode.

Weakest conditional: transmission to spot. Signal is real (three corroborating measurements). Lease market read holds (the institutional infrastructure that produces a real signal also produces correct forward pricing). Transmission is where the trade can break. Persistent scarcity that the lending venue is pricing can in principle be absorbed through wider bilateral bid-ask spreads, OTC physical settlement frictions that prevent metal flow into exchange-traded venues, or sustained policy interventions (strategic reserve announcements, Chinese export adjustments, central bank coordinations) that buffer the spot fixing from underlying scarcity. The bull case requires transmission to operate within the trade’s twelve-month window. If transmission gets pushed out to eighteen or twenty-four months by policy buffering or absorption mechanics, the structural thesis still resolves correctly but the trade window doesn’t. That’s the genuine soft spot. The other three steps are structurally robust if the underlying scarcity thesis is correct.

Position sizing for asymmetric upside doesn’t require high probability when payoff is sufficient. At $80 silver, a position sized to survive a 30% drawdown takes the holder to $56, at the floor of Scenario 3’s $55 to $70 range. Scenario 3 carries 20% probability. Sizing to that drawdown is sizing to a one-in-five outcome, not a worst case. Lower drawdown tolerance, size smaller. Higher Scenario 3 weight than 20%, size smaller still. From that drawdown survival level, the position captures meaningful upside to $110+ at scenario weights that don’t require the bull case to be modal. The convenience yield divergence is a real signal. Not a guarantee.

VI. The Diagnostic: What Discharged, What Did Not

What’s normalized post-squeeze and what hasn’t.

Normalized. One-month lease rate at 0.15%, below 5Y average of 1.03%. Three-month lease rate at 0.91%, modestly below 5Y average of 1.04%. EFP spread, normalized to approximately negative $0.43 by mid-February and held inside the normal range since. COMEX silver forward curve in contango at the risk-free rate. Spot silver realized volatility, compressed sharply from January peaks though still above 2024 norms. London market free float partially rebuilt from the September 2025 17% trough to approximately 28% by end-March 2026.

Not normalized. Twelve-month lease rate at 1.63%, 55% above 5Y average of 1.05%. Convenience yield divergence between futures and lease at approximately 200 basis points versus the 30 to 80 basis point range observed on spot-checks of selected 2019, 2021, and 2023 sessions outside acute stress events. Total COMEX silver stocks at 312 Moz, down 220 Moz (41%) from the October 2025 peak of 532 Moz, below the 5Y average of 346 Moz, approaching the November 2023 cycle low of 266 Moz. Registered coverage ratio at 13.4%, below the 15% stress threshold for six consecutive months. Spot silver level at 2.5x five-year average, with the structural revaluation from the pre-thesis baseline intact.

Probably noise. PSLV premium-to-NAV currently sits at negative 4.52%, against a five-year average of negative 3.05%. PSLV has traded at a discount essentially the entire post-2020 window, with peak premium of only 0.34% in early 2021 and even the October 2025 and January 2026 squeezes failing to push the fund to a premium. PSLV isn’t a reliable physical bid signal, full stop. The widened-versus-average discount tells me nothing about Western retail demand because PSLV’s price formation reflects ETF arbitrage mechanics, authorized participant flow, and creation/redemption frictions more than underlying physical sentiment. Dropping the inference. “Western retail bid is currently absent” isn’t supported by PSLV either way. American Eagle and Maple Leaf sales data and major dealer inventory turnover are the right places to look. Not pulled for this piece.

Synthesis. Acute physical stress has discharged. Structural repricing has not reversed. The bilateral lending market has permanently repriced forward physical availability higher at the 12-month tenor, generating a 200 basis point convenience yield gap with the futures market that has persisted for six months. Mechanically, this looks like a market that has cleared its leveraged excess but hasn’t resolved its underlying scarcity dynamics. The next squeeze, when it comes, will start from a lower base of speculative positioning but a higher base of lender caution. Combination: market harder to push around with paper, more reactive to physical-flow surprises.

Not the same regime as October 2025. Not the pre-thesis regime of 2023 either. Post-squeeze, structurally tight, lender-cautious market trading at a price that reflects partial repricing without the kind of acute retail bid that historically accompanies confirmed structural breakouts.

VII. Four Scenarios For $80 Silver

The data doesn’t support a high-conviction directional call. It supports a more honest framing of the distribution than the consensus does.

Spot at $80 sits roughly 33% below the January peak and roughly 2.5 times above the pre-thesis baseline. Structural drivers unchanged. Acute physical stress discharged. Lending market not fully repaired. Deliverable float remains thin. Marginal Western buyer absent.

My distribution of one-year forward outcomes. Weights are subjective, calibrated to the post-squeeze data, the lease curve signal, the bear cases in Section VIII, and historical analogs. Reweight if you read any input differently. The framework matters more than the specific numbers; forcing them to sum to 100 is what makes the discipline useful.

Scenario 1: Silver rebases higher (40%). Spot trades into $90 to $120 over the next twelve months. The 12M lease rate elevation prefigures further front-end stress that arrives later in the year, producing a smaller and more orderly repricing higher than the January spike. 40% weight reflects the lease curve signal, the structural deficit drivers that remain intact, and the absence of any visible new supply response. Single largest probability bucket, though Scenarios 1 and 2 are roughly equally weighted and together account for 75% of the distribution.

Scenario 2: Silver consolidates around current levels (35%). Spot trades in $70 to $90. Structural drivers operate slowly enough that no near-term catalyst emerges. Market spends 2026 absorbing the January overshoot. Physical buying gradually returns as the post-crash absence of retail bid normalizes. 35% is the eyeball base rate for post-squeeze commodity markets consolidating before resolving direction. Not computed off a rigorous analog set.

Scenario 3: Silver retests lower (20%). Spot drops into $55 to $70. The recession path. A global recession compresses industrial demand, and the demand softness already in WSS 2026 base case (industrial fabrication down 3% in 2025, projected to fall further in 2026, PV demand down 19% in 2026) accelerates further. Structural thesis remains intact but the timeline gets pushed out by two to three years. 20% weight higher than the consensus bull would assign and reflects honest engagement with the substitution risk I covered at length in 800 Million Ounces Short.

Scenario 4: Silver retests the high (5%). Spot exceeds $120. Requires another physical squeeze comparable in magnitude to October or January within twelve months. The 12M lease rate elevation is a precondition, not sufficient. A specific catalyst required: delivery failure at a major COMEX expiration, a Chinese export action that significantly reduces refined silver flowing to London, a US strategic reserve announcement with specific dollar figures, or a counterparty failure in the bilateral lending market that triggers a forced unwind. 5% reflects that none of these catalysts are currently visible.

Probabilities sum to 100. Dominant probability mass sits in Scenarios 1 and 2, which together account for 75%. Both consistent with the structural thesis. Scenario 3 is real and shouldn’t be dismissed. Scenario 4 is the upside tail that justifies position sizing. Not the base case driving it.

Conclusions are sensitive to modest reweighting. Take 10 points from Scenario 1 and add them to Scenario 2: modal outcome becomes consolidation rather than rebase, miners trade weakens to a smaller and slower asymmetry. Take 5 points from Scenario 1 and add them to Scenario 3: asymmetric upside still exists but the case for sizing aggressively softens further. These are my weights. A higher base rate for post-squeeze commodity consolidation than my 35% shifts the trade conclusion, not the framework.

What shifts these weights.

Toward 1 or 4: print of the 12M lease rate above 2% within ninety days, meaningful drawdown in COMEX registered silver below 60 Moz, Shanghai premium reading sustained above its current 10 to 15% range for two weeks, US strategic reserve announcement, delivery failure at the July or September COMEX contract.

Toward 2: gradual softening of the 12M lease rate back toward 1.2% through summer, flat COMEX registered inventory trajectory and coverage ratio rebuilding above 16%, convenience yield gap compressing back toward 80bps.

Toward 3: meaningful Chinese solar installation slowdown in monthly data, commercial copper-plating production announcement from a tier-one solar manufacturer with mass production within twelve months, gold-silver ratio expansion above 70, an industrial customer announcing a multi-year silver substitution program with credible engineering details.

VIII. How I Might Be Wrong

The structural silver case has four substantive bear arguments worth taking seriously.

Thrifting wins the race. Solar manufacturers have already reduced silver loadings per cell from approximately 60 mg/W in 2010 to roughly 10 mg/W today. Copper-plating roadmaps from Longi, Jinko Solar, and Shanghai Aiko target 2026-2027 mass production. WSS 2026 already reports PV silver demand down 6% in 2025 and projected down a further 19% in 2026 to 151 Moz, the steepest decline ever recorded. If the pace continues, photovoltaic silver demand could plateau at 130 to 150 Moz rather than the 250+ Moz the prior bull case assumed. That removes 60 to 100 Moz from the demand trajectory and narrows the deficit meaningfully. In 800 Million Ounces Short I assigned roughly 25 to 30% probability to a faster-than-expected thrifting outcome within five years. WSS 2026 has now elevated this risk into the official base case. The silver bulls who think solar demand is bulletproof haven’t read the report.

Substitution beyond solar. The palladium analog is the relevant precedent. Palladium ran thirteen consecutive years of structural deficits, prices surged from $600 to $3,429, and then automakers substituted platinum at a rate exceeding 600,000 ounces per year and palladium collapsed 76% from its peak. Substitution risk for silver is real. The demand surface is broader than palladium’s was. Breaking silver through substitution would require simultaneous displacement across solar, electronics, semiconductor packaging, EV electrical systems, and grid infrastructure. Engineering constraints and timelines differ across each application. Possible. Materially more difficult than palladium’s path.

The crowded trade and recession risk. Managed money net long positioning on COMEX has been elevated. The January crash was caused by leveraged positioning unwinding under margin call pressure, not by any change in fundamentals. A global recession would simultaneously reduce industrial demand, trigger liquidation of speculative positions, and temporarily overwhelm the structural deficit with demand destruction. Silver fell from $21 to $9 in 2008. A meaningful recession could produce a 30-40% drawdown from current levels even with the structural thesis fully intact. The structural floor is higher than in 2008 because the industrial demand base is much larger, the critical mineral designation provides policy backing, and ETF holders in India and Europe have proven stickier than US retail. Floor is not $90.

Policy reversal risk. The Chinese export licensing regime is bullish in direction, but it’s reversible. If Beijing’s strategic calculus shifts, the 1,500 to 2,500 tonnes annually of Chinese refined silver the licensing constrains could return to global supply chains within months. The US critical mineral designation is also a policy decision rather than a physical constraint, and a future administration could deprioritize it without legislative effort. Neither reversal is the base case, but treating either as permanent is analytically lazy.

None of these risks, individually, invalidates the structural thesis within three to five years. They’re positively correlated. The conditions that accelerate thrifting are the same conditions that incentivize substitution in electronics and increase scrap supply. A recession arriving alongside those price-driven responses could collectively close the deficit by 2028 or 2029. The combined scenario is more probable than any individual one.

Bear cases above: 3-to-5-year horizon. Section VII scenarios: 12 months. The 20% I assigned to Scenario 3 (”Silver retests lower, $55 to $70”) incorporates the recession path only, because copper-plating commercialization at industrial scale and meaningful demand-side substitution beyond solar are unlikely to play out inside twelve months even on aggressive timelines, though the trajectory is already established. The thrifting and broader substitution risks above are mostly tail mass beyond the Section VII horizon, not weight already inside the 20%. A reader holding a 3-to-5-year view marks down the 12-month bull case probabilities further than Section VII does. A reader trading the 12-month window treats the bear cases above as monitoring catalysts that shift the distribution rather than as priced-in risks.

That’s the bear case. Not decisive at $80 silver. Any position sizing that doesn’t account for it is incomplete.

IX. The Miners Math

The structural conditions translate into a specific opportunity at the equity level. The case is operating leverage. The argument isn’t novel. The thesis that primary silver miners outperform silver during sustained bull markets is established in mining finance literature and has been articulated repeatedly across multiple precious metals cycles. What’s novel is the specific shape of the embedded-price disconnect at this point in the cycle, and the fact that the cleanest historical analog (gold miners 2009 to 2011) argues the realized outperformance is smaller than the AISC arithmetic suggests. Half the precious metals research industry will hate the conclusion of this section. Their job depends on the opposite being true.

Definitional point first. “Primary silver miner” is industry shorthand for a company classified as silver-primary by index providers and the sell-side, not a guarantee that silver is the majority of revenue. Pan American Silver derives more revenue from gold than from silver at current prices. Hecla Mining derives substantial revenue from base metals. Their reported AISCs reflect byproduct credits that depend on non-silver commodity prices, not pure silver cost discipline. A 5x or 25x operating leverage figure computed off a byproduct-credit-subsidized AISC isn’t a pure silver leverage figure. The cleanest operating leverage exposure sits at names where silver is 70%+ of revenue and AISC isn’t heavily dependent on gold or base metal credits. That subset is much smaller than the SIL holdings list.

The basket itself is structurally diluted. As of May 2026, Global X Silver Miners ETF (SIL) top holdings include Wheaton Precious Metals at 21.53%, Pan American Silver at 12.83%, Coeur Mining at 10.38%, First Majestic Silver at 4.92%, Hecla Mining at 4.64%, OR Royalties at 4.23%. Largest holding is a streamer, not a primary miner. Streamers buy the right to a percentage of mine output at fixed prices and don’t have operating leverage to silver the way primary miners do. WSS 2026 reports only approximately 28% of 2026 mine production is expected to come from primary silver mines; the rest is byproduct of copper, lead, zinc, and gold operations. The “silver miners” category is heterogeneous on its face, and basket exposure conflates very different operating profiles. If you bought SIL last October on the silver thesis you didn’t buy silver miners. You bought a streamer, three Mexican base metal byproduct stories, and a handful of actual primary producers diluted to roughly 30% of the portfolio. That’s not the trade.

Surface data first. It’s initially confusing.

Silver has returned approximately 150% over the last twelve months, from roughly $32 to $80. Over the same period, SIL has returned approximately 131 to 178% depending on the window, and SILJ has returned approximately 154 to 190%. On a headline basis, miners have outperformed silver by 0 to 36 percentage points depending on endpoints. Miners have repriced substantially in absolute terms. The basket has captured roughly 1.0 to 1.2x silver’s return, not the 3 to 8x operating leverage theory predicts at industry-current cost dispersion.

AISC dispersion across primary silver producers is the widest of the post-2020 cycle. Q1 2026 prints where available, Q3 2025 elsewhere:

Pan American Silver: $6.63 silver-segment AISC in Q1 2026 (down from $13.88 in Q1 2025), heavily subsidized by gold by-product credits at $4,800+ gold and increased contribution from low-cost Juanicipio ounces. Company’s own forward guidance puts Q4 2026 silver AISC back at $18.50 to $20.75. The Q1 print isn’t a sustainable run rate. Precise illustration of why pure-silver operating leverage comparisons computed off byproduct-credit-subsidized AISC mislead: a 45%+ gold rally has flattered the headline figure in ways that fully reverse if gold corrects. Use Q4 guidance for forward operating leverage modeling, not Q1 print. Anyone modeling Pan American as a $6 AISC miner is going to look stupid in three quarters.

Hecla Mining: $11.28 AISC in 2025 after by-product credits.

Coeur Mining: $20.01 adjusted silver costs applicable to sales in Q1 2026.

First Majestic Silver: FY2026 AISC guidance of $26.15 to $27.91 per AgEq ounce.

Endeavour Silver: guided $27 to $28.

Santacruz Silver: $35.62 in latest reported quarter.

Range is approximately $7 to $36 if you include Pan American’s Q1 anomaly, or approximately $11 to $36 if you exclude it and use forward guidance. Dispersion has widened, not narrowed, over the past two reporting periods. Failure mode: if Q2 2026 prints come in 15 to 20% above guidance on diesel, labor, and royalty inflation across multiple flagship names simultaneously, dispersion widens further at the high-cost end. Low-cost producers retain positive economics with compressed margins. High-cost producers at $35+ AISC see negative cash margins at any silver price below their fully-loaded cost, killing operating leverage at the top of the cost curve entirely rather than weakening it. Dispersion-widening, not dispersion-narrowing, is the actual Q2 risk.

Operating leverage arithmetic at three illustrative cost points, using the forward-looking AISC range rather than the byproduct-credit Q1 prints.

At AISC of $18 (low-cost primary producer with meaningful gold credits), silver moving from $32 to $80 expands gross margin per ounce from $14 to $62. A 4.4x increase in unit profitability.

At AISC of $25 (mid-cost primary producer), silver moving from $32 to $80 expands gross margin per ounce from $7 to $55. A 7.9x increase in unit profitability.

At AISC of $32 (higher-cost primary producer), silver moving from $32 to $80 expands gross margin per ounce from approximately $0 to $48. Leverage approaches infinity at the high-cost end because pre-expansion margin was approximately zero, but in practice these names trade as call options on silver rather than continuous operating leverage.

Embedded operating leverage in silver miner equity isn’t a single number. It’s a distribution that varies dramatically by AISC. The headline “miners have leverage to silver” obscures that some miners have 4x leverage and some have 25x+ to the same underlying price move.

Against this backdrop of 4 to 25x+ unit profitability expansion, SIL up 131 to 178% versus XAGUSD up 148 to 154% is a return ratio of approximately 0.95 to 1.2x. Trailing 52-week weekly regression beta of SIL to XAGUSD is approximately 0.9 raw, with R-squared in the moderate 0.6 to 0.7 range. Beta below one. Headline outperformance, where it exists, is alpha rather than amplified leverage. Opposite of what fixed-cost operating leverage theory predicts.

Regression caveats: 52 weeks is a short window with around 50 data points, R-squared sits moderate, and the highly significant alpha could be period-specific rather than a stable feature. A longer window or different specification could move the beta and alpha. The directional conclusion (SIL hasn’t captured 3x to 8x operating leverage) is robust across reasonable specifications. The specific point estimate is window-dependent.

Two possible interpretations.

First: the market has only partially priced the underlying margin expansion and the catch-up is the trade. Second: the market has priced the basket correctly and the AISC arithmetic is the wrong model for predicting equity returns. Three reasons the first interpretation might be right, one reason the second simply is.

Three reasons the gap exists, each affecting different names differently.

Costs aren’t fixed. AISC has been rising across the industry, driven by diesel, labor, explosives, steel, and royalty payments that scale with revenue rather than profit. Cost inflation has eaten some of the leverage. How much varies considerably by jurisdiction (Mexican peso labor versus Peruvian sol labor versus Canadian dollar labor produces different cost trajectories) and by operational efficiency (declining grades at flagship operations forces more rock processed per ounce produced).

Production isn’t stable. Several major silver miners have experienced operational disappointments over the last twelve months. Fresnillo has had grade decline issues at Saucito and Fresnillo proper. Pan American’s Escobal mine in Guatemala remains suspended pending consultation with indigenous communities. First Majestic has had production guidance revisions at San Dimas and Santa Elena. The product of “ounces produced” times “margin per ounce” matters more than either in isolation, and several flagship names have seen the volume side decline while the margin side expanded.

Equity markets price expected future cash flows, not current snapshots. If consensus analysts modeled silver going from $32 to $50 a year ago, miners would already be partially priced for that margin expansion. The question is what silver price is currently embedded in consensus 2026 EPS estimates and consensus NAV calculations. Public sell-side coverage for primary silver producers generally lags realized spot prices in the modeling assumptions, particularly during fast moves. If consensus assumptions sit anywhere well below realized spot, the names where the gap is widest and operational reliability is strongest are the names where the most repricing remains.

Fourth possibility, the null hypothesis: the market may be pricing the basket correctly. A 0.95 to 1.2x return ratio with rising AISC, declining grades at flagship operations, Mexico-Peru-Bolivia jurisdictional risk premia, hedge book drag at several major names, and the optionality differential between metal and equity isn’t obviously wrong. “Miners are underpriced for $80 silver” and “market is correctly discounting the operational and jurisdictional drags” are observationally identical at the basket level. The only thing that distinguishes them is what consensus models are actually using for forward silver prices. Concluding the efficient pricing interpretation is right isn’t refuting the structural silver thesis. It’s concluding miners aren’t the vehicle for it.

The 2009 to 2011 gold analog is the cleanest historical precedent and it cuts against the conclusion this section is building toward. Gold rose from approximately $880 in January 2009 to a peak above $1,900 in August to September 2011, roughly a 115 to 120% return. HUI gold miner index returned roughly 110 to 120% over the same window depending on exact endpoints. During the cleanest two-and-a-half-year precious metals bull market of the modern era, gold miners didn’t meaningfully outperform the metal. Operating leverage theory predicted 3x to 8x outperformance. Reality delivered approximately 1x. Reason: gold miner AISC was rising sharply during the bull market as diesel, labor, and declining grades all worked against fixed-cost assumptions, and consensus models lagged the realized gold price throughout the move rather than at any single resolution point. The catch-up trade in gold miners operating leverage proponents point to came largely after the metal peaked in late 2011, with the strongest miner outperformance running into early 2012 before the broader bear market began.

The 2009 to 2011 comparison is imperfect because the silver miner universe today differs from the gold miner universe of that period in scale, jurisdiction concentration, and capital structure. Silver miners are smaller, more Mexico-Peru-Bolivia concentrated, and operationally more variable than the major North American and South African gold majors that dominated HUI in 2009 to 2011. Two of those differences could conceivably work in favor of silver miners this cycle: smaller market cap means consensus model lag is more likely to persist longer because sell-side coverage is thinner, and the smaller universe means dispersion picking has more room to add value than basket exposure. The third difference, jurisdictional concentration in higher-risk geographies, cuts the other way and is part of why the silver miner basket today carries discounts the gold miner basket didn’t.

The analog is binding. The cleanest historical precedent for a sustained precious metals bull market shows the miner basket delivering approximately 1x beta to the metal, not the 3x to 8x operating leverage theory predicts. Reasons (rising AISC during the bull, consensus model lag persisting through the move rather than resolving) apply to silver miners today at least as much as they applied to gold miners then. Category-level miner trade isn’t the asymmetric exposure it appears to be on AISC arithmetic alone. The trade is name-specific or it isn’t a trade. SIL outperforming XAG by another 30 to 50 percentage points over twelve months requires a small number of specific names delivering returns that drag the basket against the headwind of the basket’s other holdings that don’t capture leverage. Category bet is structurally weak. Dispersion bet is what remains.

If the analog applies, the current setup is the “miners catching up” phase, not the “miners leading” phase. Silver has done its repricing; miners have done part of theirs. Remaining catch-up lives in specific names where consensus models lag realized economics, jurisdictional risk is priced in rather than ignored, and operational reliability is intact. SIL contains 30+ holdings including some pure-play primary producers, some byproduct producers, some streamers, and some developers. The mix dilutes the operating leverage that pure primary producers offer. Buying SIL gets partial credit on the trade. Buying the right three or four names gets full credit.

For a reader holding this piece in isolation: no actionable recommendation. The basket bet is structurally weak per the gold analog. The name selection bet requires the company-level work I’ve deferred. Two coherent options: wait for the next piece, or take silver exposure through the metal and accept the lower beta in exchange for not betting on dispersion capture. Both defensible. What this piece doesn’t support: buying SIL or SILJ as a category expression of the structural thesis. The analog argues the basket will likely deliver close to 1x silver beta, not the multiples AISC arithmetic implies.

X. What I Am Watching

Five signals will determine which Section VII scenario plays out and whether the miners thesis continues to make sense.

The twelve-month silver lease rate is the most important single signal. As long as it sits above 1.4%, the bilateral lending market is pricing forward physical risk the futures market isn’t. A break above 2% would mean the bull case is resolving and another squeeze is probable within ninety days. A drift back to 1.2% over the summer would mean the market is healing fully and the post-squeeze regime is more benign than I argue.

The convenience yield gap between futures and lease, currently approximately 200 basis points, is a useful summary metric. Sustained gap above 150bps argues the structural thesis is alive. Compression toward 80bps argues the market is healing fully.

COMEX silver inventory. Total stocks at 312 Moz as of May 8, 2026, down from the October 2025 peak of 532 Moz and below the 5Y average of 346 Moz. Registered coverage ratio at 13.4%, below the 15% stress threshold for six consecutive months. A break of total stocks below the November 2023 cycle low of 266 Moz signals the bleed is accelerating; the squeeze is reasserting itself through physical flow rather than just paperwork. A rebuild back above the 5Y average plus registered coverage above 16% signals the squeeze is healing.

Q1 2026 primary silver producer cost prints are the first concrete test of the Section IX operating leverage framework. Framework depends on AISC dispersion across the silver producer universe remaining stable enough that low-cost names retain meaningful margin expansion at $80 silver. Q1 prints so far mixed. Pan American’s $6.63 anomaly is gold credits, not silver cost discipline. Q2 prints will be more informative once gold normalizes its byproduct-credit contribution.

Chinese export licensing implementation is the wild card. Regime took effect January 1, 2026. Operational impact will only become clear over the next two to three quarters as approved exporters work through the 45-day-plus regulatory process. If Chinese silver exports to the West fall meaningfully below 2025’s 5,100-tonne 16-year high, the supply side of the deficit tightens. If exports continue at 2025 pace despite the new framework, the policy ended up more symbolic than operational.

Explicit falsification criteria. If within ninety days the following three conditions are simultaneously met, the structural tightness thesis is wrong and the trade should be exited: the 12-month silver lease rate falls below 1.2%, AND the COMEX coverage ratio rebuilds above 16%, AND the Shanghai premium compresses to its post-2020 normal range on a like-for-like basis. If two of three trigger, the thesis is meaningfully weakened and position sizing should be cut by half. If one of three triggers, the thesis is on watch but intact. The combination matters because the three signals come from independent participant sets in different venues. All three softening simultaneously requires either coordinated relief across the bilateral lending, exchange-cleared delivery, and Chinese physical venues, or a genuine resolution of the underlying scarcity. Either way, the bear case is clean.

Ambient context, less directly relevant: the gold-silver ratio (currently approximately 47), PSLV premium-to-NAV (currently negative 4.52%, dropped from the diagnostic per Section VI), and Shanghai Gold Exchange silver premium to London (basis to be locked before publication, directional signal aligned with the bilateral lending and exchange-cleared scarcity readings per Section III).

XI. What Comes Next

Three claims.

First, the squeeze that drove silver from $80 to above $120 in two months discharged its acute phase but didn’t resolve its structural conditions. Futures market is pricing convenience yield near zero. Lease market is pricing convenience yield at approximately 1.97% at the 12-month tenor. COMEX warehouses have bled 220 Moz of total stocks from the October 2025 peak to May 2026, with registered coverage below the 15% stress threshold for six consecutive months. Shanghai Gold Exchange silver contract is trading at a premium to London well above its post-2020 norm. Three of these four signals are aligned. The futures curve at the risk-free rate is the outlier, and the outlier is the signal.

Second, the inference chain from that divergence to a trade is multi-step but not multiplicative. The four steps are conditional views of one underlying state of the world (structural silver tightness thesis being correct), not independent draws. The 40% Scenario 1 probability is the right integration. Trade is asymmetric rather than high-conviction. Soft spot: transmission from lender pricing to spot price within the twelve-month window.

Third, the operating leverage math at the category level shows miner equities have absorbed substantially less of the underlying unit profitability expansion than AISC arithmetic implies they should, with the realized SIL-to-XAG return ratio of approximately 0.95 to 1.2x and a trailing 52-week regression beta below one sitting well below the 3x to 8x that fixed-cost leverage theory predicts. The 2009 to 2011 gold analog argues most of that gap is structural rather than a mispricing waiting to be corrected. Remaining asymmetry lives in specific names, not the basket, because AISC dispersion across the silver producer universe runs from approximately $11 to $36 on forward guidance (or $7 to $36 if you include Pan American’s Q1 2026 anomaly) and basket weights don’t differentiate.

The next piece is the company-level work. Framework is three-dimensional: operating leverage at $80 silver versus consensus model embedded silver price (which names are still being modeled well below realized spot), country risk concentration (Mexico, Peru, Bolivia, and the smaller number of safer jurisdictions), and operational risk (production trajectories, grade decline, hedge book exposure, byproduct credit dependency). Names that score well on all three dimensions are the trade. Names that look cheap on operating leverage but score poorly on country risk or operational reliability are value traps.

I’ll cover primary producers, royalty and streaming companies, and a small set of developer and junior names. I’ll identify which names within the SILJ basket are vulnerable to consensus revisions rather than candidates for further upside, because category-level exposure to junior silver miners conflates very different risk profiles. Whether any of those names rises to a structural short call versus an “avoid” call depends on the specific operating and balance sheet details. Next piece.

Three independent measurements of physical silver scarcity say silver is structurally short. The futures curve says it isn’t. Pick a side. Pretending the gap isn’t there is the worst trade in the room.

Methodology Appendix

The convenience yield decomposition in Section IV depends on specific market data inputs that aren’t publicly reproducible. Inputs and assumptions:

Data source. All futures, lease, and SOFR readings sourced from Bloomberg terminal on the May 11, 2026 session. The lease rate series is the standard Bloomberg silver SOFR-basis lease curve; the futures curve is COMEX silver standard 5,000oz contract.

Tenor matching. The 12-month convenience yield comparison is constructed by interpolating between the Dec 2026 ($82.29) and Dec 2027 ($85.58) futures contracts to a synthetic May 2027 forward price of approximately $83.66, which produces an annualized carry from SI1 ($80.40) of approximately 4.05%. The 12-month lease rate of 1.63% is matched against this 12-month synthetic forward.

Storage assumption. LBMA-quality silver storage at major commercial vaults runs roughly 0.2 to 0.6% per year. Midpoint of 0.4% used in the headline decomposition, with sensitivity acknowledged across the range.

SOFR reference. 3.6% per NY Fed reference rate, May 8, 2026 reading.

Convenience yield identity. Futures decomposition uses the standard storable commodity cost-of-carry identity: F = S × (1 + r + s - c), solved for c. Lease decomposition uses the equilibrium identity for storable commodity lending: L = r - c, solved for c. In a frictionless integrated market, the two c values should converge.

Historical context. The 200 basis point figure is compared against an informal sample of approximately fifteen trading sessions across 2019, 2021, and 2023 outside acute stress windows. Sample produced convenience yield gaps clustering in the 30 to 80 basis point range. Sample too small to support a percentile claim about the current reading; the directional claim that 200 is much wider survives. A continuous decomposition across the post-2018 window is planned as a follow-up.

Reproducibility caveat. The specific 200 basis point figure is a Bloomberg-terminal result. The directional claim (gap materially wider than the post-2018 normal range, with the lease venue capturing physical scarcity that the futures venue isn’t absorbing) is independently supportable from public industry reporting and structural deficit data. The precise magnitude requires terminal access to verify.

Iranian officials called the leaked U.S. text a “wish-list” inside hours. Iran fired on three U.S. destroyers transiting the Strait on May 7. CENTCOM responded with strikes on Iranian military facilities. UAE absorbed Iranian missile and drone strikes across May 4-5 and again on May 8, including the major May 4 wave that hit the Fujairah Petroleum Industries Zone.

CENTCOM disabled three Iranian-flagged tankers across May 6 and May 8 under blockade enforcement. Project Freedom, the U.S. escort operation, was effectively halted May 5 by Saudi and Kuwaiti access restrictions, then publicly paused by Trump on May 6, with restrictions reportedly lifted by May 8 after a second Trump-MBS phone call. Framework is no longer the active diplomatic track.

"Late to Hormuz" (May 6) merged a 6-12 month structural fair value with a 48-hour tactical scenario set and arrived at $85-93. The structural floor held. The tactical premise didn't. The two have no business sharing a number.

Architecture below replaces the broken methodology. Two layers, separately specified, with separate kill switches at each layer. Tactical layer governs position sizing on a 4-8 week horizon. Structural layer governs fair value on a 6-12 month horizon. They move independently.

Marks across the back catalog. Cash Not Bonds carry leg cut in full. Palladium SBSW thesis back on track on the May 5 COT print. BOJ amplifier trimmed pending June. Buy The Miners working at $4,800+ realized gold. Short The Packagers thesis intact. Tactical entry wrong. Mosaic pending Q1.

Brent Dec 2026 at $87. Floor at $83. Structural EV at $108. The macro instrument is partially priced. Cleaner alpha sits in equities. Part three develops it.

Framework on probation. Next 10 kill switches resolve status. Probation timer starts now.

Below: full architecture, marks against the active book, watchlist, fair value computation.

Subscribe now

Read more

1. The Hormuz oil thesis is now consensus. The FT carries the headline. The trade got harder.

2. I’m revising the Dec 2026 Brent structural fair value down 7-12%, from $95-100 to $85-93. About $7-10 of the prior range was war premium I conflated with structural repricing. The directional thesis on inventory mechanics stands.

The April 17 low of $77.32 is the level Dec 2026 has held against. Public forecast dispersion runs $83 (Goldman) to $100 (Barclays) for 2026 Brent, so the floor is a trading reference, not an equilibrium.

3. Three resolution paths over the next 48 hours: Iran accepts the framework (35-50%, mid 42%), Iran stalls and the April 8 ceasefire holds without a signed deal (30-45%, mid 38%), deal collapses and escalation resumes (15-25%, mid 20%).

4. Kill switches across the active book remain operative. None have fully triggered. The Cash, Not Bonds amplifier legs are on watch into the BOJ next read and the FRBNY custody Thursday print. The exit is mechanical. Triggered by a kill switch print, not a headline.

5. The petchem leg (DOW down 5.78% today) is the cleanest live read on Short The Packagers. The cyclical war premium pricing out of DOW is consistent with the structural long thesis intact on Permian ethane. Mosaic prints May 11.

Six weeks ago I said the back end of the Brent curve was mispriced because physical engineering constraints don’t unwind on a signature. That’s now priced.

The front of the curve has caught the inventory drawdown story (FT lead today: 200mn barrel decline in April).

Reuters this evening: oil supply shock set to worsen as inventories fall further even if the conflict ends. Same physical-reality-from-below argument the back-end thesis was built on.

Ken Griffin warned on recession risk at the Semafor World Economy Forum in mid-April. When your framework reaches the lead headline, it stops paying. The job changes from being right to managing what’s left. If you’re reading the consensus from the FT lead, you’re late to the trade and you’ll be late to the unwind.

Subscribe now

Read more

They’re reading the rating. The bears are reading the contract.

The contract leaves nothing f…

Read more

Most of that move is the leverage senior gold producers carry to the gold tape, which fell 5% from its mid-April peak. That part is loud, mechanical, and not the question worth asking.

The question worth asking is why one fell four points harder than the other on the same gold move. The Q1 prints answer it. So does what I am doing about it.

If you read my post “Gold Is Not Failing” three weeks ago, the macro framework still holds. The structural bid is intact. The wash-out has cleared. Today’s post is the operational follow-up: how the framework survives contact with two earnings prints, two corrections from the original call, one explicit kill switch I owe subscribers, and one number that decides whether the better of the two names gets fresh capital next week.

Subscribe now

Read more

RTX has agreed to rebuild the magazine. Five framework agreements with the Department of War in February: Tomahawk to 1,000 per year, AMRAAM to 1,900, SM-6 to 500. Sell-side desks size the deals at $7 to $8 billion of incremental annual revenue at full ramp. Morgan Stanley called the upside “underappreciated” on April 22. Wolfe Research said the same in February.

None of it is in the numbers.

I pulled the Bloomberg consensus build for RTX. The Raytheon segment is modelled at 7.34% organic growth in FY27 with operating margin expanding modestly to 12.28%. FY28 consensus EPS sits at $8.28 across nineteen estimates with a standard deviation of $0.37, clustering tight against a base business compounding path. FY29 mean steps to $8.91, FY28 plus 7.6% growth, what base business produces. Consensus isn’t wrong on the framework calendar. It doesn’t have one.

The reconciliation vote that funds the rebuild lands in December. Seven months from now. The supplier base has already committed more than $5 billion of capex to deliver against framework volumes, on the assumption that vote clears. Williams International is building a billion-dollar engine plant in Florida. L3Harris is breaking ground on a $1.27 billion expansion in Virginia. The Pentagon committed $1 billion to anchor an L3Harris Missile Solutions IPO planned for the second half of this year.

If the December vote clears, the 2028 print is mechanical. If it scales the production lines down materially, RTX trades back to peer multiples and the asymmetry collapses. That’s the trade.

Base case on RTX direct: 25 to 35% total return through mid-2029, 8 to 10% annualised with dividends. Dead-money floor 6 to 8% annualised. Bull case closer to 12% annualised if framework revenue prints earlier than the supplier base implies. Patient capital with embedded optionality, not a coiled spring.

TSCS is read at 400+ investment firms. Subscribe for the full trade architecture, kill switches, and pre-committed thresholds. Free posts cover the thesis. Paid posts cover the execution.

Subscribe now

What consensus actually models

Walk the Bloomberg FA Segments build forward.

Raytheon segment FY25 actual revenue was $28,043 million at 11.5% adjusted operating margin. Bloomberg consensus FA Segments shows Raytheon growing to $29,895 million in FY26 and $31,981 million in FY27. Organic growth of 7.92% in FY26 and 7.34% in FY27. Operating margin modeled at 11.89% in FY26 expanding to 12.28% in FY27. That is the model.

Run the framework math against it. The five framework agreements size at $5 to $7 billion of incremental annual revenue at full ramp on my build, $7 to $8 billion on sell-side builds. Even $500 million of framework contribution into FY27 would push Raytheon segment growth above 8%. Consensus sits at 7.34%.

The base business growth rate this assumes is consistent with what RTX is actually printing on current programs. Raytheon segment sales grew 10% in Q1 2026 with the segment operating margin up 150 basis points to 12.2% on favorable mix and productivity. The trailing twelve-month organic on Raytheon’s existing program book sits at 6 to 7%. The 7.34% consensus sits one point above base. That’s normal program escalation, not framework conversion. The math doesn’t need framework revenue to balance.

BI’s RTX Equity Outlook on April 22 is the cleanest in-terminal check. Ferguson and Lee, the BI primary and secondary analysts on RTX, frame the thesis as: “Defense gains could extend on robust demand for air defense and missiles, efficiencies and higher-margin foreign sales.” That’s the entirety of it. No conversion year. No UCA timeline. No framework revenue contribution to FY27 or FY28.

BI’s Pentagon FY27 budget piece, dated April 15, addresses the framework procurement directly: “Procurement for RTX’s four critical weapons programs (Tomahawk, SM-3, SM-6 and AMRAAM) should widen 3.6x to $16.3 billion, underscoring the Pentagon’s efforts to rebuild inventory for a potential China conflict.” That’s the procurement request, not the revenue forecast. No analog to $16.3 billion in any consensus model for FY27 or FY28. Procurement requested, revenue not modelled.

The dispersion picture

Dispersion confirms the segment build. FY27 EPS consensus across 23 of 25 expected analyst estimates sits at $7.57 with a standard deviation of $0.178. The high estimate is $7.90, the low is $7.03. Tight clustering, no meaningful upside outlier. FY28 consensus across 19 of 21 expected estimates sits at $8.28 with a standard deviation of $0.37. The clustering is tight against a base business compounding path.

FY29 is the most informative pull. It’s where framework loading would first show in the mean if anyone had modelled it. The mean sits at $8.91 across 9 of 10 expected estimates with a standard deviation of $0.596. FY28 plus 7.6% growth. Not framework. Base business compounding into the year framework revenue should land.

Here’s the counter to take seriously. Sell-side doesn’t extend full models three and four years out, as a process matter. FY29 estimates are thin (9 of 10) because the buy-side does not pay for FY29 modelling depth at this distance from print, and analysts respond to that demand structure. Read that way, the absence of framework loading isn’t an information asymmetry. It’s how the revision cycle works. Models extend forward when management guides them forward or when contemporaneous data forces the issue.

I think it’s right and it strengthens the trade. The thesis isn’t that consensus is wrong about framework economics. It’s that the revision cycle hasn’t engaged with framework revenue yet, because the conversion mechanics aren’t contemporaneous (UCA awards, definitization, percentage-of-completion recognition). When framework revenue physically prints in late 2028 in Raytheon segment reporting, consensus will revise. The trade is being positioned ahead of that mechanical revision, not ahead of an analytical disagreement. Structurally easier to defend. Doesn’t require analysts to be missing anything. Just requires the calendar to do what calendars do.

LMT PAC-3 backs this. Consensus FY26 and FY27 EPS revised up across the two quarters after framework conversion in early to mid-2026. Analysts weren’t wrong before the revision, they were waiting for the print. Same mechanism here. The asymmetry is timing, not insight.

Bloomberg doesn’t aggregate FY28 or FY29 segment-level revenue for RTX in FA Segments. The model doesn’t extend that far. That’s the finding. The conversion year isn’t modelled because the model doesn’t reach the year conversion lands.

Caveat. Dispersion tells me what the consensus aggregate is doing. It doesn’t tell me what individual broker models contain privately. Some sell-side desks may be carrying framework revenue in unpublished internal models without it showing in their consensus submissions. I do not think the dispersion supports that interpretation. If even two or three of nineteen FY28 contributors had loaded $1 to $2 billion of framework, std dev would widen past $0.37 and at least one outlier would appear above $8.80. Neither is there. But I cannot verify this absent access to individual broker models, so the claim is “consensus has not loaded framework” rather than “nobody anywhere has modeled it.”

The mechanics of consensus model extension are well-documented in defense. When LMT printed PAC-3 framework conversion ahead of consensus expectations in early to mid-2026, consensus FY26 and FY27 EPS revised meaningfully upward across the following two quarters. That is the directional pattern. The trade is the gap between when supplier capacity comes online, when consensus extends models forward to engage with it, and when the political calendar locks the production targets in.

The empirical picture is clean. Consensus is positioned as if framework doesn’t deliver in FY27, FY28, or FY29. Suppliers say it delivers in 2028 and ramps through 2029 to 2030. That last condition has a date.

The December 2026 reconciliation binary

The framework targets are RTX press release figures from February 4, 2026. Tomahawk 1,000+ per year, AMRAAM 1,900+ per year, SM-6 500+ per year. Not contractually committed. Not in the $271 billion backlog. Not in any DoD primary document I can find. They depend on FY27 reconciliation to fund.

The FY27 budget request asks for 727 of 785 Tomahawks and 434 of 540 SM-6s through reconciliation. Those numbers are the framework production targets translated into FY27 procurement asks. The first reconciliation bill passed earlier this year. The second is pending. Speaker Johnson called the second reconciliation bill a “tough sell” in mid-April.

The conference report markup runs September through December 2026. Pass-fail on framework production lines is a specific test against the appropriations bill text. If the conference report includes 727 Tomahawks and 434 SM-6s in framework-funded form, the calendar locks and the 2028 print is mechanical. Definitization runs on the historical 9 to 18 month base rate. UCAs land Q1 2027 to Q2 2027, sequentially after the December 2026 reconciliation print clears. Definitization runs into Q1 to Q2 2028. Material revenue recognition under percentage-of-completion accounting begins two to three quarters after definitization, into late 2028 or early 2029. The trade architecture below is sized for that path.

If the conference report scales the production lines down materially, the targets become aspirational. RTX retains the existing program book (Tomahawk Block V most recently definitized in January 2026 on a 13-month framework-to-definitization timeline, SM-3 IB at $1.366 billion for 78 rounds, AMRAAM Lots 39-40 still working through definitization), which is a real business at a real margin. The upside surprise that makes the multiple work disappears. RTX trades back to peer multiples of 17 to 21x on a consensus path with no framework. Asymmetry gone.

This is where the genuine information asymmetry lives. Defense appropriations watchers (DefenseScoop, Breaking Defense, BGOV) cover markup at granularity institutional investors don’t read. Committee assignments matter. HASC and SASC rosters have shifted. Senator Wicker as SASC chair has been explicit on munitions stockpile rebuild as a priority. The Speaker’s whip count on the second reconciliation matters more than any sell-side note. The previous markup on PAC-3 came through largely intact with marginal trims. Useful base rate, not dispositive for a different program in a different fiscal cycle.

Pre-committed read: if the conference report preserves more than 80% of the framework production asks (more than 580 Tomahawks, more than 350 SM-6s), the trade is on. If between 50% and 80%, the trade is alive but compressed and position size cuts to half. If below 50%, the trade is off and the position closes against the print.

That binary lands in about seven months.

The chokepoint timeline

Five programs. Framework-implied volumes. The supplier constraint that gates each. Earliest physical delivery year. Capacity is real, publicly funded, under construction. Revenue isn’t modelled.

Tomahawk: 1,000+ per year, gated by Williams International

Framework target more than 1,000 per year. Current baseline approximately 169 deliveries in CY25 against an FY25 maximum facilitization of 530. The 1,000+ target is an 11x step from the FY25 minimum sustaining rate, a 6x step from current deliveries, and a 2x step from documented maximum capacity.

The binding constraint is Williams International, the private Michigan company that sole-sources the F107 turbojet. The expansion is funded and underway. December 2024 Defense Production Act award of $253.7 million for F107-WR-105 capacity. May 2025 announcement of a $1 billion investment in a new one-million-square-foot manufacturing complex at Shoal River Industrial Park in Okaloosa County, Florida. Groundbreaking on the 250,000-square-foot Phase 1 (Module 8) in November 2025, completion targeted late 2026. Phase 2 begins 2028. Final 500,000 square feet planned for 2035 to 2036. Combined investment exceeds $1.25 billion.

The targets are physically planned for. Williams doesn’t commit $1 billion to a new facility without demand visibility. The timeline is what constrains the ramp. Phase 1 delivers turbojets in late 2027 at earliest, factoring tooling installation and qualification after construction completes. The 1,000+ Tomahawk per year target depends on Phase 1 plus the existing Walled Lake campus. Earliest physical delivery year for framework-rate Tomahawk volumes: 2028.

The diligence gap is real. Bloomberg’s entity database, SPLC, and financial data functions return zero results for Williams International. A company committing $1.25 billion to the largest private defense expansion in the US has no terminal footprint. Public records that would tighten this timeline meaningfully (Okaloosa County permits, USA Spending obligations, DPA Title III congressional reporting) are available but not pulled at time of writing.

The warhead chain is equally concentrated. Tomahawk warhead fill is RDX/HMX from the Holston Army Ammunition Plant in Kingsport, Tennessee, government-owned and operated by BAE Systems Ordnance Systems on a contract through 2033. Holston is mid-build on a multi-year nitration and melt-cast capacity expansion, timeline 2027 to 2028.

AMPAC in Cedar City, Utah is the sole North American producer of rocket-grade ammonium perchlorate, the oxidizer in every solid propellant in the US defense inventory. Acquired by NewMarket Corporation (NYSE: NEU) in January 2024 for approximately $700 million. NEU announced a $100 million capacity expansion in April 2025 targeting a 50%+ AP capacity increase scheduled for completion during 2026. NEU is treated separately in a follow-on piece on the supplier-side trade architecture.

AMRAAM: 1,900+ per year, the most defensible ramp

Framework target at least 1,900 per year. Current production approximately 1,200 globally per Raytheon CEO Phil Jasper at FEINDEF in April 2025, who said Raytheon doubled AMRAAM production in 2024 to 1,200 and was doubling again in 2025. The Tucson plant has headroom to 2,400. The framework target sits inside the band Jasper defined publicly. This is a 1.6x current-production step, the smallest multiplier in the framework.

Supplier landscape is the cleanest. Nammo Perry, Florida facility for AMRAAM solid rocket motors targets 2027 operational date. Northrop Grumman has invested over $1 billion in SRM capacity since 2018 and is doubling tactical SRM capacity by 2027. The gating risk is Nammo’s 2027 timeline slipping, which compresses the ramp by 6 to 12 months but does not break it. Earliest physical delivery year for framework-rate AMRAAM volumes: 2027 at the soonest, 2028 base case.

AMRAAM is the program where FY27-FY28 consensus could plausibly carry framework revenue without breaking the math. If consensus were loading framework anywhere first, it would be here. It hasn’t.

SM-6: 500+ per year, gated by the Mk 72 transition

Framework target more than 500 per year. Current capacity approximately 165 per year on one shift, 200 per year fully funded for FY26. SM-6 Selected Acquisition Report (December 2023) shows ramp to 200 by FY26 and 300 by FY28. Framework target is 1.7x the FY28 plan-of-record and 2.5x the funded FY26 rate.

The binding constraint is the Mk 72 first-stage booster, sole-sourced to L3Harris (legacy Aerojet Rocketdyne, now L3Harris Propulsion). Aerojet president Ross Niebergall acknowledged in late 2024 that the company was thousands of motors behind contract across 15 to 20 programs. RTX management described Aerojet output as a “mixed bag” in July 2024. Calio Q1 2026: “Rocket motors, given the concentrated supply base, it’s going to require a step change in the supply chain.”

The framework volume implies roughly 2 to 3x what L3Harris is currently shipping. RTX awarded second-source Mk 72 development contracts to Northrop and Nammo in April 2025. Northrop static-fired its prototype Mk 72 in December 2025. That is a development milestone. Production qualification is on a 24 to 30 month timeline at minimum. Production-rate second-source Mk 72 supply is a 2028 event in best case, 2029 to 2030 base case.

L3Harris has responded with the largest non-Pentagon defense industrial capital program in the United States. Camden, Arkansas: more than $400 million committed, six-fold large-SRM capacity increase, broke ground November 2025. Orange County, Virginia: $1.27 billion committed, announced April 15, 2026. Huntsville AMF-South: 379,000 sq ft, opened August 2025. Pentagon committed $1 billion as a convertible-preferred anchor investor in the L3Harris Missile Solutions IPO planned for 2H 2026. Earliest physical delivery year for framework-rate SM-6 volumes: late 2028, with full ramp 2029 to 2030.

This is where the gap between supplier capacity and consensus is widest. SM-6 can’t ramp to 500 per year without second-source Mk 72 capacity, which doesn’t exist. Consensus hasn’t engaged with that.

The Mk 72 bottleneck generates one of the watches this piece builds: the L3Harris Missile Solutions IPO.

SM-3: $400 million to $1.2 billion run-rate plus the sustainment lift

RTX did not numerically disclose framework production targets for SM-3 Block IIA or Block IB. Defense Archives puts SM-3 IIA current production at 24 per year ramping toward 36, and SM-3 IB at 3 to 4 per month with maximum facilitization of 5 per month. Block IIA is gated by Mitsubishi Heavy Industries supplying the third-stage rocket motor and nose cone under the US-Japan workshare. Block IB was definitized in March 2026 at $1.366 billion for 78 rounds.

The Missile Defense Agency awarded an $8.41 billion contract modification on March 18, 2026 (HQ0851-21-D-0001) lifting the SM-3 sustainment IDIQ ceiling from $3.33 billion to $11.74 billion through October 29, 2029. That is the near-term print trade and it deserves its own section below.

Definitization base rates

Historical base rate for framework authorization to definitized production contract is 9 to 18 months for missile programs. Lockheed PAC-3 MSE framework signed January 6, 2026 converted to a $4.7 billion undefinitized contract action on April 10 (94 days framework to UCA). RTX’s own patterns show 13 months on December 2024 Tomahawk Block V (definitized January 2026), approximately 10 months on May 2025 SM-3 Block IB UCA (definitized March 2026), and a $3.5 billion AMRAAM Lots 39-40 UCA from July 2025 still open at 9 months.

Apply that base rate to the February 2026 frameworks. First UCA award between Q1 2027 and Q2 2027, sequentially after the December 2026 reconciliation print clears. First definitization 12 to 15 months after that, into Q1 to Q2 2028. Material revenue recognition under percentage-of-completion accounting begins two to three quarters after definitization, into late 2028 or early 2029. That is the supplier-deliverable timeline against a consensus model that does not extend that far.

The aggregate

Full-ramp incremental Raytheon segment revenue brackets $5.0 to $7.0 billion annually on my build. Seven-year cumulative ramp-weighted contribution is approximately $38 to $52 billion.

The capacity investment is sized for those numbers. Williams’s $1.25 billion. L3Harris’s $2.3 billion across Camden and Orange County. Northrop’s $1 billion of cumulative SRM investment since 2018. NEU’s $100 million AP expansion. Holston’s multi-year nitration and melt-cast modernization. Combined private and government industrial capital deployed across the framework supply chain exceeds $5 billion. That is a real industrial mobilization.

It does not show up in consensus revenue models for FY27, FY28, or FY29.

The SM-3 sustainment trade

The MDA contract modification on March 18, 2026 lifts the SM-3 sustainment IDIQ ceiling from $3.33 billion to $11.74 billion through October 29, 2029. The ceiling went from $722 million in 2020 to $11.74 billion now, a sixteen-fold expansion in five years on a single IDIQ.

Structurally different from framework production. The work runs against the existing program of record, not new capacity. No Williams International, no Mk 72, no second-source qualification gating recognition. Sustainment task orders definitize on 60 to 180 day timelines versus the 9 to 18 month base rate for new production UCAs. Sustainment runs higher-margin than new-build. Recurring engineering, software updates, kit upgrades, and field service run 200 to 400 bps above new-build margin, per comparable LMT and NOC structures.

The math on EPS contribution. Sustainment IDIQs typically run at 70 to 90% of ceiling over their performance period, so the $8.41 billion lift translates to roughly $5.9 to $7.6 billion of incremental sustainment revenue across the four-year window, or $1.5 to $1.9 billion per year. Apply Raytheon segment operating margin in the 12 to 13% range for sustainment-weighted work, RTX’s effective tax rate of approximately 18%, and a share count of roughly 1.33 billion. Annualised SM-3 sustainment: $0.11 to $0.15 of EPS per year. Cumulative FY26 to FY29: $0.45 to $0.60 of EPS not visible as a discrete line in consensus.

Directional, not precise. But $0.11 to $0.15 against a $7.57 FY27 consensus base is large enough to matter. Task orders against the lifted ceiling will show in 10-Q backlog disclosures within two quarters of March 18, so the Q3 2026 print in October. The first quarter where SM-3 sustainment is visible above noise is Q1 or Q2 2027 reporting. That is the near-term print catalyst, well ahead of the 2028 framework recognition, and it tightens the floor on the dead-money case.

What the multiple is paying for

RTX trades at 25.13x forward earnings on the May 1 close, against $6.91 FY26 consensus EPS at $173.67. Lockheed at 17.3x forward, Northrop at 20.5x, General Dynamics at 20.9x. The premium to the defense peer set is structural and persistent. Opened in October 2023, held through 18 plus months and multiple guide revisions.

The implied long-term EPS growth rate from Bloomberg EE is 10.12%, which sits on a base business path: Pratt commercial aftermarket growing 19% in Q1 with MRO output up 23%, Collins growing low-single-digits on rising Boeing and Airbus build rates, Raytheon growing 7 to 9% organically on strong global defense demand. That is the model the multiple is paying for. It does not require framework revenue to be embedded in any line item.

The previous version of this piece said 2 to 3 turns of multiple were paying for framework optionality. That would have set up multiple compression hitting before framework revenue. The data doesn’t support that read. The premium versus peers is paying for Pratt aftermarket quality, which is real and is in the numbers, plus a generic defense premium, which the entire sector has earned through 2024 to 2026.

Cuts both ways. If framework isn’t in the multiple and isn’t in the EPS, framework optionality is unpriced. That’s the opportunity. It also means there’s no multiple expansion catalyst on entry. Upside comes from EPS revisions when framework revenue prints, not from a re-rating. This is a compounding position with embedded optionality, not a coiled spring. The analytical work in this piece is proportioned to the size of the information asymmetry, not to the velocity of the payoff.

The defense sector is down 8% since mid-April. Lockheed’s structural problems ($4B+ classified program losses, heroic 2H ramp required) and Northrop’s $157 million B-21 EAC charge. RTX has neither. Short interest 13.3 million shares (0.99% of float), declining, days to cover 3.78. Nobody’s betting against RTX.

One popular retail argument is that buying Wolfspeed (NYSE: WOLF) or Qorvo (NASDAQ: QRVO) plays the framework upside through GaN substrate exposure. It is factually wrong. RTX manufactures military-grade GaN MMICs in-house at its Andover compound semiconductor foundry at Manufacturing Readiness Level 9, with throughput sufficient for AMRAAM AESA seekers, LTAMDS T/R modules, Patriot upgrades, and SPY-6. Wolfspeed filed Chapter 11 on June 30, 2025; its RF GaN business had already been acquired by MACOM (NASDAQ: MTSI) in late 2023, well before the bankruptcy. The retail market bought the wrong supplier derivatives.

The trade architecture

One position sized today, two binary watches.

Trade 1: RTX direct, gated on the December 2026 reconciliation print

RTX at $173.67. Half size at entry. Add to full size on confirmation that the FY27 reconciliation conference report preserves the framework production lines.

The structure recognises that the framework thesis depends on reconciliation clearing. Half size today buys SM-3 sustainment recognition and the dividend, with optionality on reconciliation converting it to a full framework trade. Full size on confirmation, ahead of supplier capacity coming online.

Base case walks like this. RTX prints FY26 in line with the $6.91 guide. The Q3 2026 print in October shows SM-3 sustainment task orders building backlog, which provides the first visible evidence above consensus. The reconciliation conference report in December 2026 preserves the framework production lines at 80%+ of request. Consensus FY27 grinds higher through 2027 on Pratt aftermarket strength, Raytheon base programs, and SM-3 sustainment recognition. Ends $7.65 to $7.80 versus current $7.57. First framework revenue shows in Raytheon segment reporting in late 2028 at a $200 to $400 million quarterly run-rate, AMRAAM and Tomahawk components ramping first. Consensus FY29 EPS revises up from current $8.91 to $9.40 to $9.80 over the following two quarters as analysts extend models forward and load framework. The stock trades at 22 to 24 times the revised forward number, which puts it in the $210 to $235 range by mid-2029. Total return through 2029: 25 to 35% on entry, plus dividends.

Bull case: framework revenue starts earlier (Q3 2028 or Q2 2028 if definitization runs faster than base rate) and FY29 EPS revision running closer to $10. At 24x, $240. Total return 40%+ plus dividends.

Dead-money case: framework revenue delayed to 2029 by supplier slippage or definitization friction. RTX prints in line through 2027 and 2028, framework doesn’t appear until Q2 2029. Stock drifts $170 to $190 for two years. Annualised return 6 to 8% with dividend. Thesis isn’t broken, asymmetry compressed.

The position is sized for entry around $170 to $175 with an add window into weakness toward $160. Weakness toward $160 represents either a sector rotation accelerating (which is the current pressure from LMT and NOC issues) or a Q2 or Q3 2026 print that disappoints on Pratt durability or Collins build rates. Neither of those is the framework thesis. Neither breaks the trade.

The kill switch on RTX direct isn’t framework arriving late. It’s framework not arriving at all. That requires the December 2026 reconciliation conference report to scale framework production lines below 50% of request, which is the falsifiable trigger detailed below, or the supplier capacity expansion to fail (Williams Module 8 not completing, Mk 72 second-source not qualifying, Holston not completing nitration expansion).

Watch 1: L3Harris Missile Solutions IPO (2H 2026)

Most actionable binary on the framework supply chain. Only watch in this piece with a compressed timeline. L3Harris has disclosed Missile Solutions generates approximately $4 billion of revenue at roughly $620 million EBITDA, with the Pentagon committed as a $1 billion convertible-preferred anchor investor. Reuters reported on April 30, 2026 that L3Harris had confidentially submitted a draft registration statement for the Missile Solutions IPO, which is the procedural precursor to the public S-1 filing. Zero publicly-traded SRM pure-plays. The IPO creates one, into a market where every defense prime and framework depends on SRM supply that’s structurally constrained.

This is not actionable today. It becomes actionable on public S-1 filing.

What the S-1 must confirm: revenue $3.5 billion or above (materially below $4 billion suggests segment cherry-picking), EBITDA margin 14%+ ($620 million implies 15.5%, which would be the highest-margin SRM business in public markets), backlog 3x+ revenue, Pentagon $1 billion convertible-preferred preserved.

Walk away on: revenue below $3 billion, EBITDA margin below 12%, backlog concentrated in a small number of programs with near-term completion risk, Pentagon anchor restructured, reduced, or conditioned on milestones not yet met.

Comp set is essentially empty for SRM pure-plays. Closest analogues: Mercury Systems (MRCY) at 14 to 18x EV/EBITDA (turnaround floor), Curtiss-Wright (CW) at 16 to 20x (defense-industrial mid), HEICO (HEI) at 25.3x current with a 5-year average 24.6x (aerospace parts ceiling). L3Harris Missile Solutions is structurally closer to Curtiss-Wright than HEICO: a sole-source defense supplier with pricing power and concentrated customer exposure, not an aerospace aftermarket monopolist with hundreds of SKUs.

At $620 million EBITDA: 14x implies $8.7 billion enterprise value (Curtiss-Wright floor pricing). 16x implies $9.9 billion (mid-range, sole-source position with discount for IPO illiquidity and customer concentration). 18x implies $11.2 billion (full Curtiss-Wright pricing, justified if the S-1 confirms the Pentagon anchor and 15%+ EBITDA margin). 22x implies $13.6 billion (HEICO-adjacent, only achievable if the market prices SRM monopoly and framework optionality together).

Pre-S-1 range 14 to 17x, $8.7 to $10.5 billion EV. Below 14x on the IPO is a gift. Above 20x on day one means the market’s already priced scarcity, and the trade moves from “buy the IPO” to “wait for the first miss and buy the dip.”

Size: starter on IPO allocation, add to full size on first earnings print that confirms S-1 numbers. High-conviction trade gated by one binary (S-1 filing and pricing). Risk: IPO pulled or delayed, Mk 72 monopoly locked inside L3Harris at a conglomerate discount.

L3Harris is a watch, not a trade, because entry is gated on a document that hasn’t publicly filed. Calling it a trade today is the loose framing this piece avoids. The public S-1 turns it into a trade. Until then it’s a binary worth tracking.

Watch 2: NewMarket Corporation (NEU)

NewMarket Corporation (NYSE: NEU) is the publicly-traded wrapper around AMPAC, the sole North American producer of rocket-grade ammonium perchlorate. The capital allocation signal is real. Specialty Materials capex from $4.1 million in FY24 to $50.7 million in FY25, 12x year-over-year, against a segment generating 6.7% of revenue. The mispricing isn’t a valuation disagreement, it’s an absence of sell-side coverage on a $6.2 billion market cap company that sole-sources oxidiser for the entire US solid propellant base.

I am not building the NEU trade in this piece because the structure depends on options market verification I have not completed. The OMON pull on NEU LEAPS, the implied volatility check at 18 to 24 month expiries, and the open interest assessment at relevant strikes are prerequisites for sizing this position as anything more than half-size in common stock with 5.6% probability-weighted upside, which is below the equity risk premium. NEU is a follow-on piece when that work is done.

Pre-committed thresholds

The kill switch architecture states the trade in falsifiable form. Each line has a value, a date, and a position adjustment. If the trigger fires, the action is the action, not a re-evaluation.

The reconciliation binary. If the FY27 reconciliation conference report does not include at least 580 Tomahawks and 350 SM-6s in framework-funded form by December 20, 2026, RTX position cuts to zero. If between 580 and 727 Tomahawks (or 350 to 434 SM-6s) clear, the position holds at half size and the bull case is off the table. If at least 727 Tomahawks and 434 SM-6s clear at full request, the position adds to full size on the first trading session after the conference report publishes.

The consensus revision check. If Bloomberg EEB FY28 mean reaches $8.65 by November 30, 2026 absent a Q3 2026 EPS beat above $1.85, the consensus gap has closed prematurely and the RTX direct position cuts to quarter size. If the FY28 mean reaches $8.85 absent any earnings catalyst, the position closes against the print.

The supplier execution check. If RTX Q3 2026 (October) Raytheon segment organic prints below 6.0% with no SM-3 sustainment commentary and no framework definitization update, the capacity ramp is running into demand problems consensus is correctly modelling. Thesis weakens. Position cuts to quarter size pending Q4 2026.

The SM-3 sustainment check. If the 10-Q for Q3 2026 (October 2026) does not show meaningful SM-3 sustainment task order backlog accumulation against the $8.41 billion ceiling lift, the IDIQ is being filled slower than the contract structure implies and the near-term print catalyst is delayed. The RTX direct base case timeline extends by approximately 6 months, but no position adjustment is required.

The Williams Module 8 check. If Williams International Phase 1 (Module 8) does not announce or confirm completion by Q2 2027, the Tomahawk framework ramp slips by 6 to 12 months. No position adjustment on RTX direct (the framework print catalyst extends but the trade is preserved).

The L3Harris IPO check. Reuters reported on April 30, 2026 that L3Harris had confidentially submitted a draft registration statement. The actionable trigger remains the public S-1 filing. If the public S-1 has not filed by January 31, 2027, the IPO is pulled or delayed materially and the Watch 1 position closes. If the public S-1 reveals revenue below $3 billion or EBITDA margin below 12%, do not participate. If the public S-1 reveals Pentagon convertible-preferred restructured or withdrawn, do not participate.

These are the triggers. They are dated. They have specific values.

Kill switches at the framework level

Framework-level kill switches ordered by assessed likelihood.

Framework definitization slips materially or scales down. Highest-likelihood negative outcome, and the binary detailed in the thresholds above. The December 2026 reconciliation conference report is the trigger. Speaker Johnson has called the second reconciliation bill a “tough sell.” Monitor: Pentagon contract announcements for multi-year awards at framework-implied scale, FY27 appropriations markup September to December 2026, RTX Q2 earnings July 22 definitization commentary, Lockheed PAC-3 framework follow-through.

The industrial base hits a hard ceiling. Williams International, Holston AAP, and Mk 72 risk. Any one of the four major supplier expansions (Williams F107 Module 8, Holston RDX/HMX modernization, AMPAC 50% AP expansion, Mk 72 second-source qualification) slipping by a year compresses the framework ramp to 60 to 70% of stated targets, moving full-ramp incremental revenue from $5.0 to 7.0 billion to $3.5 to 4.5 billion. The thesis still works at those numbers. Monitor: L3Harris Propulsion earnings for on-time delivery, Northrop tactical SRM capacity announcements, GAO SRM industrial base updates, Williams International Module 8 construction progress.

Consensus loads framework before capacity comes online. If sell-side analysts revise FY28 EPS estimates above the pre-committed thresholds before framework revenue physically prints in 2H 2028, the upside surprise narrows or disappears. Monitor: Bloomberg EE FY28 mean and dispersion trajectory through 2026 and 2027.

FMS political risk. Japan (400 Tomahawks), Australia (200), Netherlands (175) face domestic friction. Slippage compresses the international component, doesn’t break the thesis.

Tariff escalation. RTX has $850 million gross tariff exposure for 2026. Net P&L after mitigations is a $75 million tailwind YoY. Trade survives unless tariffs go up an order of magnitude.

Capital allocation during build-out. RTX’s $3.1 billion 2026 capex is up nearly 20% from 2025. If frameworks finalize faster than expected and trigger additional capacity investment, FCF could disappoint relative to the $8.25 to 8.75 billion guide even as the underlying business strengthens. Raytheon margin compression below 11% would signal the build-out is harder than communicated.

Geopolitical resolution softens demand. Low likelihood. Operation Epic Fury reportedly expended approximately 850 Tomahawks in February to April 2026 (per CSIS), which alone exceeds twelve years of FY26-rate procurement.

What I’m watching

Reconciliation calendar is the sequence.

July 22, 2026 (expected): RTX Q2 print. Watch for follow-on order commentary on SM-3 IB beyond the March 2026 definitization, AMRAAM Lots 39-40 progress, and any framework-related management language. Raytheon segment organic margin above 12% confirms base business strength.

September to October 2026: HASC and SASC markup of FY27 NDAA. Watch the Tomahawk and SM-6 line items in subcommittee bills. Reconciliation language on framework production funding starts here.

October 2026 (expected): RTX Q3 print. The first quarter where SM-3 sustainment task order backlog should be visible against the $8.41B IDIQ ceiling lift. Watch 10-Q backlog disclosures.

November to December 2026: Conference report on the second reconciliation bill. The framework binary lands here. The pre-committed thresholds above govern position adjustment.

January 2027 (expected): RTX Q4 print and FY27 guide. Watch for any explicit framework revenue commentary or Raytheon segment FY27 organic guide above 8%.

2H 2026: L3Harris Missile Solutions IPO S-1 filing.

Through 2026 to 2027: Bloomberg EE FY28 EPS dispersion trajectory. Bloomberg FA Segments extension to FY28 estimates (when consensus first publishes FY28 segment-level revenue, the framework conversion year becomes visible in the model).

Williams International Module 8 construction progress at Shoal River Industrial Park (Phase 1 completion targeted late 2026). Northrop Defense Systems segment commentary on Mk 72 qualification milestones.

LMT Q2 late July as the parallel framework-to-backlog test on PAC-3, the cleanest read on whether the LMT framework conversion mechanics that this trade extrapolates to RTX are firming or slipping.

What this piece doesn’t cover

Three topics sit adjacent to this thesis and deserve separate treatment.

The first is NewMarket Corporation as a standalone supplier-side trade. The capital allocation signal is real and the AMPAC monopoly position is structural. The trade requires options market verification and a more developed comp set than the Olin Winchester analogy supports. Follow-on piece.

The second is the broader DOGE and defense budget politics. Reconciliation risk on framework funding is the central trade frame in this piece, but the wider political landscape, including the tension between budget efficiency rhetoric and the largest munitions procurement request in a generation, is a structural question this piece touches but does not develop.

The third is Northrop Mk 72 second-source qualification as a standalone optionality trade. The qualification is a real binary on a 2027 to 2028 timeline that could re-rate Northrop’s SRM franchise. I am not building the trade because quarter-size on a binary that pays $15 to 30 per share if it qualifies is rounding error against a $573 stock. It belongs in any portfolio that already owns Northrop on broader merits, but it does not justify a standalone position.

Caveats

Framework production targets that anchor the headline math (Tomahawk 1,000+ per year, AMRAAM 1,900+ per year, SM-6 500+ per year) are RTX press release figures from February 4, 2026. They are not contractually committed, not in the $271 billion backlog, and not in any DoD primary document I could locate. RTX’s spokesperson declined to comment to Breaking Defense on current annual production levels or contract definitization timing. Sell-side modeling that implies the dollar figures are independently DoD-attested overstates source quality.

The sell-side framing referenced in the introduction is anchored on secondary reporting. The $7 to $8 billion headline figure and the “underappreciated” framing from Morgan Stanley’s April 22 note are consistent across multiple financial news summaries and confirmed via Bloomberg News quotation. The underlying Wolfe Research February 8 note was sourced from descriptions on Investing.com, Yahoo Finance, and Insider Monkey rather than the primary note; the methodology is not verified. The magnitude is what matters for the thesis and the magnitude is not in dispute across sources, but I am not representing direct access to the sell-side research.

The PAC-3 revision pattern referenced as the analog for consensus model extension is directional. The exact magnitude and timing of the LMT FY26 and FY27 EPS revisions following framework conversion in early to mid-2026 should be charted from Bloomberg before publication; the directional pattern is what the trade depends on.

The supplier timeline analysis is grounded in publicly disclosed data: DPA awards, capacity expansion announcements, groundbreaking dates, qualification milestones. Williams International is private and its production capacity, current output rates, and Module 8 completion timing are not independently verifiable beyond the trade press and DoD press release record. Public records that would tighten the timeline meaningfully (Okaloosa County permits, USA Spending obligations, DPA Title III congressional reporting) are available but not pulled at time of writing. Holston AAP is government-owned and its expansion timeline is similarly tied to DoD disclosure rather than independent audit.

The SM-3 sustainment EPS contribution is built from publicly disclosed IDIQ ceiling figures and assumptions about utilization rate, margin, and recognition timing. The $0.11 to $0.15 per year EPS contribution is a directional estimate. RTX does not break out sustainment versus new-build margin in segment reporting, so the margin assumption is anchored on comparable structures across LMT and NOC sustainment work. Actual contribution may run higher or lower depending on task order timing and mix.

The L3Harris Missile Solutions IPO valuation framework uses EBITDA figures L3Harris has disclosed publicly but that have not been audited in a public S-1. The $620 million EBITDA figure may change when the public S-1 drops. Reuters reported on April 30, 2026 that L3Harris had confidentially submitted a draft registration statement, which is the standard procedural precursor to the public filing. Pentagon convertible-preferred terms have not been publicly detailed beyond the $1 billion headline. The trade is not actionable until the public S-1.

Operation Epic Fury munition expenditure figures (~850 Tomahawks of approximately 3,100, 50% of THAAD inventory, 50% of Patriot interceptor stockpile) are from CSIS’s April 21, 2026 reporting. State Department and White House releases confirm timing but do not publish exact stockpile figures.

Position sizing reflects the kill switch stack, supplier-specific risks, and the patience requirement on the RTX base. One position today (RTX direct, half size). Two binary watches (L3Harris IPO on public S-1, NEU as follow-on). Three catalysts at three horizons: SM-3 sustainment recognition in 2026 to 2027 prints, the December 2026 reconciliation binary, the 2028 to 2029 framework print.

The framework is not in the numbers. The trade is the print.

Share

This post is for informational and educational purposes only. It does not constitute investment advice, a recommendation, or an offer or solicitation to buy or sell any security. The author may hold positions in the securities discussed. Readers should conduct their own research and consult a licensed financial advisor before making investment decisions. Past performance does not guarantee future results.

URA launched in November 2010 to give investors diversified basket exposure across the uranium value chain. That thesis was the entire product. It lost essentially everything in the 2011-2015 stress window despite holding positions chosen to express different theses across different parts of the universe.

Caveat. URA’s trigger was Fukushima, an exogenous accident, not the endogenous narrative-policy break I’m arguing the current cycle is exposed to. Fukushima isn’t the same mechanism.

But once the demand narrative collapsed (for whatever reason), the value-chain diversification investors thought they had provided essentially zero protection.

Conversion didn’t decorrelate from miners. Developers didn’t decorrelate from producers. Industrials didn’t decorrelate from spot.

The cover story for diversification was the value chain. The reality was a single demand-narrative bet expressed five different ways. Two cleaner cases produced the same outcome through endogenous mechanisms closer to the current cycle.

The 2008-2015 Chinese steel buildout produced correlated multi-factor breaks across mining, processing, equipment, and downstream construction equity that “diversified” portfolios couldn’t absorb. The trigger was demand-narrative compression as the post-stimulus credit cycle turned. Not an accident. The 2020-2024 offshore wind cycle is mid-stress, with Ørsted, Siemens Gamesa, and Vestas all impaired together.

The trigger was cost-curve assumptions colliding with rate normalization. Policy aggressiveness compressed in response rather than supporting. Both cycles had the architecture the nuclear renaissance now has: a demand thesis driving policy aggressiveness, supply chain capex sized to that thesis, and an investor base holding “diversified” positions across the value chain.

Both produced correlated breaks that the value-chain diversification could not absorb. The five names everyone owns (Cameco for fuel cycle, Constellation for fleet, Centrus for enrichment, BWXT for construction, NexGen for developer optionality) aren’t five bets. They’re one bet, made three times, dressed as diversification. The market hasn’t noticed yet because the cover story is convincing.

Here’s the framework for why the consensus is sized as if buildout cycles don’t produce correlated breaks, when the historical record is meaningfully populated with cases that did. I’m not claiming the current cycle will break the same way.

I’m claiming the consensus is sized for an outcome more confident than the case-study evidence supports. Call the mechanism narrative-policy beta.

Bearish in framing, net long in recommendations. The Nuclear Primer and the February uranium piece are assumed as foundation. Cameco, Constellation, Centrus, and BWXT are USD on NYSE. NexGen is CAD on TSX with NYSE dual listing. Spot uranium and SWU references are UxC weekly. All references are late April 2026 unless otherwise noted.

Paid subscribers read the rotation. Everyone else reads the framework.

Behind the paywall: a forensic read of Cameco’s contract book that resolves the equity-versus-fundamental correlation question. Four scenarios with specific portfolio weights for each. Rotation principles for when one factor fires versus when two correlate. Pre-committed falsification triggers across all three factors. A binding exit condition.

The triggers are dated. NRC commissioner composition through June 30, 2026. CNSC Port Hope renewal the week of November 3, 2026. PJM September 2026 emergency auction. Paid subscribers see the position rotate as signals fire.

Subscribe now

New to TSCS? The Nuclear Primer is the 45,000-word free foundation this piece builds on.

What you actually own

Read more

Two days later, the Wall Street Journal ran a story arguing the money was necessary because the American magazine was too empty to defend Taiwan.

On the same day the story ran, the Pentagon closed a $1 billion strategic investment in L3Harris, the solid rocket motor supplier that bottlenecks every munition the story was about.

Nobody structures, negotiates and closes that kind of deal in response to a news story that broke hours earlier.

The sequence matters.

On Tuesday, Admiral Samuel Paparo, commander of US Indo-Pacific Command, submitted written testimony to Congress stating that “current production timelines are misaligned with operational expenditures and the threats we face in today’s global security environment.” The munitions he named in the statement were JASSM-ER, LRASM, Maritime Strike Tomahawk, Precision Strike Missile, SM-3 and SM-6. These are exactly the weapons a Pacific fight turns on.

Thursday, the Wall Street Journal reported anonymous administration officials putting the replacement window at six years.

Within hours of the WSJ story, White House Press Secretary Karoline Leavitt called it “false,” saying the military is “fully equipped with enough weapons to handle any operation ordered by the president.”

All of this inside a single news cycle. Two days after the Pentagon unveiled a $1.5 trillion defense budget, the largest in American history. Three weeks before Trump meets Xi in Beijing.

Someone is being walked through this.

It is worth understanding who, and how.

What follows is the answer to two questions the WSJ piece does not ask. Why did the story run this week and not six months ago. And what does the American munitions gap look like if you compare it to the Chinese readiness gap the Pentagon is simultaneously describing.

Subscribe now

I. The Timeline

Before the leak, the plan. Before the plan, the war. The order is the argument.

January 6, 2026. The Department of War and Lockheed Martin announce a framework agreement for the Patriot PAC-3 MSE, targeting production expansion from approximately 600 per year to 2,000 per year under a seven-year agreement. Lockheed delivered 620 PAC-3 MSEs in 2025, more than 20 percent above the prior year.

January 29, 2026. Lockheed Martin and the Department of War (the Trump administration’s “Department of War” secondary title for the Department of Defense, authorized under the September 2025 executive order) announce a framework agreement to quadruple annual THAAD interceptor production from 96 per year to 400. Lockheed simultaneously breaks ground on a new Munitions Acceleration Center in Camden, Arkansas.

February 4, 2026. RTX (the old Raytheon) announces five multi-year framework agreements with the Department of War. Tomahawk annual production is to exceed 1,000 units, up from roughly 60 per year, which RTX itself describes as more than a tenfold increase. SM-6 is to exceed 500 per year. AMRAAM is to reach 1,900. “Many of these munitions will grow 2 to 4 times their existing production rates,” the company said. Tomahawk, specifically, is the tenfold outlier.

February 28, 2026. Operation Epic Fury begins. The USS Frank E. Petersen Jr. fires a Tomahawk Land Attack Missile that day, a fact confirmed by a Navy DVIDS photo caption and subsequent Naval News reporting; Reuters separately confirms that “the first shots fired were Tomahawk missiles.” By late March, the U.S. has fired more than 850 Tomahawks in a single campaign, a figure that exceeds multiple prior years of Tomahawk procurement combined.

April 1, 2026. Boeing and Lockheed sign a seven-year framework to triple PAC-3 MSE seeker production. The seeker is the single most-cited industrial bottleneck in the Patriot supply chain.

April 9, 2026. The Department of War awards Lockheed a $4.761 billion PAC-3 MSE contract, with work at Camden, Arkansas, and estimated completion by June 30, 2030.

April 21, 2026. Secretary of War Pete Hegseth unveils the FY27 defense budget at $1.5 trillion, a 42 percent year-over-year increase and the largest single jump in American defense spending since World War II. Pentagon officials acknowledge, on the record, that the budget “was drawn up prior to” Epic Fury. Defense press reporting puts the missile procurement line at a 188 percent year-over-year increase. A separate roughly $200 billion supplemental is signaled for the Iran operation and stockpile replenishment.

April 23, 2026. The WSJ piece runs. “The U.S. has burned through so many munitions in Iran that some administration officials increasingly assess that America couldn’t fully execute contingency plans to defend Taiwan from a Chinese invasion if it occurred in the near term.”

April 23, 2026 (same day). L3Harris closes a $1 billion Department of War strategic investment in its Missile Solutions business, structured as convertible preferred security and warrants. This isn’t a standard procurement contract. The Pentagon’s taking an equity-adjacent position in the solid rocket motor bottleneck. Nobody structures, negotiates and closes that kind of transaction in response to a news story that broke hours earlier.

May 14-15, 2026. Trump is scheduled to meet Xi in Beijing.

The timeline breaks the official narrative. The WSJ piece asks readers to accept this causal chain: U.S. fought Iran, ran out of munitions, therefore must surge production and budgets. The actual chain runs the other way. The U.S. planned to surge production and budgets, then fought Iran, then leaked a story calibrated to make the surge politically unopposable and to shape the negotiating posture of a summit it had been planning for weeks.

The frameworks of January 6th, January 29th and February 4th, together, committed the American industrial base to a scale-up that makes the six-year WSJ figure close to meaningless for the munitions that drove the headline. The officials aren’t lying. The ramp-up schedule matters. Signing a framework agreement in January does not mean the missiles exist in July. Lockheed’s own filings describe the PAC-3 agreement as seven-year, not one-year, and the binding industrial constraint (solid rocket motors at L3Harris) takes years to expand regardless of how much paper flows from the Pentagon. There’s a transition window, two to three years, running on accelerated legacy production, depleted stockpiles, and whatever the new facilities deliver. Inside that window, Taiwan is harder than outside it. Fair point.

But “transition window” isn’t the story being sold. The story being sold is “Taiwan is vulnerable because Iran drained us,” which invites “spend more, faster.” If the story were “Taiwan is vulnerable during the industrial ramp we committed to in January,” the response would be “don’t start more wars during the ramp.” Different stories. WSJ tells the first. We’re telling the second.

II. The 2027 Category Error

Every urgency claim in American defense writing about Taiwan flows through 2027. The FY27 budget is sold against it. The WSJ piece’s tempo of alarm depends on it. Davidson Window citations are default in Congressional testimony, think-tank reports, cable-news defense segments. Western strategic planning treats it as settled.

It isn’t settled. It’s an analytical artifact, the product of an inferential leap from a Chinese readiness milestone to a Western operational deadline.

The phrase “Davidson Window” entered the strategic lexicon in March 2021, when Admiral Philip Davidson, then commander of U.S. Indo-Pacific Command, testified before the Senate Armed Services Committee that the threat of a Chinese invasion of Taiwan was “manifest during this decade, in fact, in the next six years.” That sentence put 2027 at the center of Western defense planning. His successor, Admiral John Aquilino, maintained the urgency. CENTCOM used the timeline to justify Middle East posture drawdowns. The Director of National Intelligence cited it in public threat assessments. The FY27 budget is, in several places, sold as a response to it.

Follow Davidson’s testimony back to the Chinese source material and you don’t find an operational deadline. You find the “Centenary Goal of the PLA,” 建军一百年奋斗目标, a domestic directive from Xi Jinping calling for the People’s Liberation Army to accelerate the “integrated development of mechanization, informatization, and intelligentization” by the hundredth anniversary of the PLA’s founding. Every PLA document that mentions 2027, including the translated corpus held at the China Aerospace Studies Institute, frames the date as a readiness milestone, a point by which the force is supposed to possess a particular set of modernized capabilities. No invasion window. Taiwan isn’t named. The target is what the military should be, not what it should do. The U.S. intelligence community’s own March 2026 assessment makes the same point, concluding that Beijing has not committed to a 2027 invasion timetable.

Western intelligence heuristically translated “possess the capability by 2027” into “willing to use the capability by 2027.” That translation, in the loose sense of “inferential leap” rather than a linguistic mistranslation, was not malicious. It was lossy. And the loss has compounded across four years of defense commentary until a domestic Chinese readiness goal has become, in the American discourse, an intercepted war plan. Critics of the framing, notably Oriana Skylar Mastro at Stanford, Michael Beckley at Tufts, and Jude Blanchette at CSIS, have argued the same point in less blunt language for years: 2027 is a bureaucratic milestone that has morphed into an analytic Rorschach test, and most of the American discourse around the “window” is testing the analyst, not the adversary.

The WSJ article’s urgency depends on accepting the 2027 framing. Strip the frame, the “six years to replace” figure loses its weight. The six years were never anchored to anything specific. The transition window, the industrial ramp, the supplemental, the summit, all of it gets re-pegged to a date that was always softer than the reporting suggested.

Intent exists. Xi’s personal ownership of PLA modernization, his invocations of Taiwan reunification as “great rejuvenation,” accelerating PLA exercises, all real. None of it produces an operational deadline. Intent without a deadline requires sustained preparation. Intent with a deadline requires emergency supplementals. American defense politics is funded on the second framing. It should be funded on the first, which is harder to sell and cheaper to execute.

The 2027 category error is the single most consequential analytical move in the American Taiwan discourse, and it is baked into almost every piece of defense journalism that uses the “window” framing. Correcting it doesn’t make Taiwan safe. It flattens the urgency gradient, which changes the rational policy response, which changes the defensible supplemental. That’s why the correction doesn’t appear in the WSJ piece.

III. The Wargame Symmetry

Second analytical move. This one’s load-bearing.

Any serious argument about American munitions inadequacy for a Taiwan fight runs through CSIS’s wargaming work. The Center for Strategic and International Studies ran nearly two dozen iterations of a Taiwan invasion scenario between 2022 and 2023, and has updated the outputs since. Those wargames produced the consumption rates that populate every piece of defense writing on the subject, including this one. LRASM global inventory exhausted within three to seven days. JASSM stocks depleted around day nine. Long-range munitions consumption rates of more than 5,000 missiles in three weeks of a Taiwan Strait invasion fight. These numbers are the analytical currency of the discourse. They appear in CSBA briefings, AEI commentary, Congressional testimony and, implicitly, in the WSJ piece’s “could not fully execute contingency plans” formulation.

Here is the part almost nobody flags. Those consumption rates are calibrated to a PLA that runs the scenario correctly. Synchronized launches across multiple brigades. Accurate targeting. Sustainable logistics. Functional command and control under stress. If the actual PLA is worse than the wargame PLA, American magazines last longer. The missiles don’t change. Chinese fires are less accurate, less synchronized, less sustainable than the simulation assumes.

Is the actual PLA worse than the wargame PLA?

The evidence, accumulated between 2023 and early 2026, is unusually rich and unusually bad for Beijing. Xi Jinping removed much of the senior leadership of the PLA Rocket Force across 2023 and 2024. Li Yuchao, the commander, and his deputies Liu Guangbin and Zhang Zhenzhong, were detained and investigated for corruption. Two consecutive defense ministers (Wei Fenghe and Li Shangfu) were removed over related investigations. In November 2024, Admiral Miao Hua, a member of the Central Military Commission, was suspended. In December 2023, the National People’s Congress Standing Committee had dismissed nine senior military officers from the legislature, with further removals following in February 2026. In June 2024, the Central Politburo formally expelled Wei and Li from the Party with language stating that they had “severely contaminated the political environment of the equipment sector,” a phrase which only appears when the Party is communicating industrial-scale rot.

What the corruption looked like, per U.S. intelligence assessments cited by Bloomberg in January 2024, was extraordinary. Missiles filled with water instead of fuel. Silo fields in western China built with lids that did not open. Funds allocated to modernization systematically diverted. Ranks and promotions sold. These specific claims come from anonymous U.S. intelligence through a single outlet. Credibly reported, not independently verified. Bloomberg reported that U.S. officials had concluded Xi was “less likely to contemplate major military action in the coming years” precisely because of what the intelligence community had found.

Then it got worse. In late January 2026, forty days before Operation Epic Fury began, Chinese authorities publicly announced that General Zhang Youxia, vice chairman of the Central Military Commission and China’s most senior uniformed officer, was under investigation. The accusations, according to Guardian and SCMP reporting that both outlets explicitly flagged as unable to independently verify, were twofold: accepting bribes in exchange for promotions, and leaking China’s nuclear weapons data to the United States. The second accusation, if true, is the more significant. It would imply, at the highest level of the CMC, that U.S. intelligence had either a direct source or a stream of product on Chinese strategic forces. Whether Zhang was the source or was framed as one to protect another is unknowable outside Langley. Reuters’s more conservative framing describes the investigation as “serious violations of discipline and law,” standard CCP language that doesn’t confirm the nuclear-leak allegation. We hold both versions. The direction of damage is visible either way.

Put these facts together. What fraction of the PLA Rocket Force is operationally degraded? Three scenarios. The percentages below are our estimates, weighted against the confirmed public record. Not figures from any published intelligence assessment.

In the low scenario, roughly 20 percent probability, corruption is largely ring-fenced to financial graft and older liquid-fueled systems (legacy DF-4 and DF-5s). The modern solid-fuel deterrents, DF-41 and DF-26, remain operationally insulated. The detentions reflect Xi consolidating political control, not terminal rot. Roughly 10 to 15 percent of the force degraded.

In the central scenario, roughly 60 percent probability, hardware defects (targeting gyros, failing seals) extend into the conventional IRBM and SRBM inventory due to a decade of procurement skimming. More critically, the purges have paralyzed operational readiness at the institutional level: the trust required for synchronized launch protocols and complex joint operations is functionally shattered by political terror. Roughly 30 to 45 percent of the force degraded.

In the high scenario, roughly 20 percent probability, corruption extending to physical tampering of strategic deterrents implies total quality-assurance breakdown, and the fall of Zhang Youxia implies the nuclear command-and-control apparatus is compromised by factional infighting. 60 percent or more of the force degraded.

We weight the central scenario at 60 percent because it is the only one of the three consistent with both available evidence streams simultaneously. The low scenario requires dismissing the Bloomberg intelligence reporting about hardware defects; the reporting is not independently auditable, but it is specific and sourced, and a 10 to 15 percent degradation assumes the defects are not real. The high scenario requires believing Chinese strategic weapons capability is compromised at the top end; China has publicly demonstrated functioning strategic systems through ICBM tests, hypersonic glide vehicle deployment, and recent Rocket Force exercises, which cuts against the high case. The central scenario reconciles both: quality issues exist, and strategic weapons still work. That is why it carries the most weight.

The central scenario is the one to hold. Not the maximalist case. It doesn’t require believing every silo is empty or every launcher is broken. It requires believing a force with a decade of senior leadership purges, equipment ministers formally expelled for contaminating “the political environment of the equipment sector,” and a vice chairman under investigation, is less operationally capable than its order of battle suggests. Low bar.

Re-integrate with the wargames. If the central-scenario PLA is the actual PLA, CSIS consumption rates overstate the threat. American magazines last longer in the fight because Chinese fires are less accurate, less synchronized and less sustainable than the wargames assume. The airframes and launcher counts are real, so LRASM and JASSM-ER still exhaust. But the pace is slower, the window of American inadequacy is narrower, and the strategic picture inside the transition window is meaningfully less dire than the raw wargame outputs suggest. Put crudely: the munitions gap is calibrated to a scenario that is becoming less likely precisely because of the Chinese weakness the symmetry argument describes. The wargames are simultaneously the evidence for the gap and the reason the gap may be narrower than the wargames imply.

Almost nobody makes this move. Hundreds of pieces cite CSIS consumption rates as evidence of American inadequacy. Dozens cite the PLARF purges as evidence of Chinese degradation. Nobody adjusts the first set for the second. The two half-truths sit in separate documents. Integrate them, you get where the Pacific actually is.

Chinese mass does compensate for Chinese incapacity. Beijing has built hundreds of silos, thousands of ballistic missiles, the world’s largest navy by hull count, and more air-launched missiles than it knows what to do with. A 30 percent operational degradation rate on a large force is still a large effective force. The PLA isn’t a paper tiger in absolute terms. But Pentagon officials describe American capacity as compromised “in the short term” during a specific scenario. The same qualification applies to the PLA, more so.

Epistemic asymmetry matters here. American production data in Section V comes from SEC filings, DoD budget documents, GAO assessments, named-analyst CSIS work. Chinese capability data comes largely from Bloomberg’s reporting on anonymous U.S. intelligence, plus confirmed Chinese state announcements of the detentions. First set: auditable. Second: not. U.S. intelligence could have the Rocket Force degradation scope wrong. The leaks could be selective, optimized to justify Pentagon posture. The critique we level at the WSJ story applies to the Bloomberg one too.

That said, the scale of the confirmed detentions is evidence independent of any intelligence leak. A defense minister. A Central Military Commission vice chairman under investigation. Much of the top of the Rocket Force.

The scale reads as systemic. An alternative reading frames the purges as factional consolidation, not operational failure. Under that reading the symmetry argument still survives, weaker: American weakness is well-documented, Chinese weakness is inferred, the symmetry case attenuates. We hold the stronger version.

IV. A Brief Interlude on What The Story Leaves Out

Before the arithmetic: what Sections II and III together do to the WSJ framing. The piece the Journal published, and every downstream piece that quotes it, treats the American side of the ledger as the variable and the Chinese side as the constant. China is coming. The question is whether America can hold them. The answer depends on American missile counts, American production schedules, American industrial capacity.

That framing only works if the Chinese side is a constant. It is not. It is a variable, and the direction of variance over the past three years is downward. The same intelligence community that provides the order of battle used to calibrate the wargames provides the corruption and purge assessments used to describe Chinese degradation. Both streams are selective. Both streams are auditable against confirmed public events. Neither stream is being integrated by the outlets that use the wargame outputs to sell supplementals.

A rigorous WSJ piece would ask: American magazines are tight, Chinese readiness is also tight, what does the balance look like during the transition window? That version doesn’t get written, because it doesn’t support a $200 billion supplemental.

It supports a more careful posture, a narrower set of authorities, and a prudence framing that is harder to translate into line items. The version that does get written is the one where the Chinese side is held constant at wargame-readiness levels, the American side is described as depleted, and the resulting asymmetry is presented as urgent.

The urgency is the product. The arithmetic below is what the product is defending against.

V. The Numbers

The six-year figure and “munitions gap” framing don’t survive arithmetic. Baseline numbers below, sourced from DoW budget documents, CSIS, Lockheed and RTX SEC filings, and Congressional testimony. Variance cited explicitly. Where math rests on CSIS wargame outputs, the PLA-degradation argument above matters.

Tomahawk. Current Tomahawk production is not disclosed in RTX’s 10-K. The framework target is more than 1,000 per year, a figure RTX itself describes as a more than tenfold growth from existing rates of roughly 60 per year. The FY26 combined request, before the war, was for 79 missiles (57 Navy, 22 Marine Corps). The FY27 request, which the Pentagon confirms was drawn up before Epic Fury, asks for 785 Tomahawks split 58 in the base budget and 727 in reconciliation, an 894 percent year-over-year increase. Section 20004 of the One Big Beautiful Bill Act of 2025 separately appropriated $250 million specifically for Tomahawk production capacity expansion, targeting a rate of 800 per year. Roughly 55 percent of the Pentagon’s $40 billion munitions request is structured as reconciliation rather than discretionary funding, because reconciliation dollars remain available over a longer horizon than year-by-year appropriations. The design of the ask is pre-commitment against a seven-year industrial horizon, not one-year procurement.

Big caveat. Tomahawk, PAC-3, LRASM, JASSM-ER and THAAD all compete for solid rocket motor supply out of L3Harris (the old Aerojet Rocketdyne), which is the genuine industrial bottleneck. Todd Harrison at AEI, the same institution that has called the overall defense proposal unbuildable, described the mechanism on the record: “They can’t buy that many in a single year, and Raytheon can’t produce that many at once. But the intent is they will have the money there so that they can go to Raytheon and say, ‘Hey, we want to place a big order.’ It’s even better than a multiyear procurement, because in multiyear procurement, you still have to wait for future appropriations.” Framework agreements aren’t missiles. The 785-unit FY27 request isn’t a one-year delivery schedule. It’s a funded pre-commitment against bottleneck-constrained capacity. Honest Tomahawk replacement range is two to four years, depending on how fast the SRM constraint unwinds.

THAAD. Baseline production entering 2026 was 96 per year. FY26 base budget requested only 25 interceptors, plus 12 under reconciliation, for 37 total. The January framework targets 400 per year. Estimated stockpile as of December 2025 was 534 interceptors. CSIS estimated between 100 and 150 were used in the 2025 Twelve-Day War, plus more in Epic Fury. At 400 per year, a 150-interceptor drawdown is a five-month replacement. At 96 per year, it was eighteen months. Six years only works if you freeze production at the FY26 request level, which is not the plan. The downstream math rests on a CSIS usage estimate with a wide range. Higher actual usage, the replacement window extends. Lower, it compresses. Arithmetic directionally correct. Precision softer than the confident phrasing suggests.

PAC-3 MSE. This is the single most-used interceptor in the arsenal and the most important in any ground-based terminal defense of forward airbases, which is what a Taiwan scenario mostly requires. Lockheed delivered 620 PAC-3 MSEs in 2025, more than 20 percent above the prior year. The January framework, stated in Lockheed’s annual-report chair letter as a “seven-year agreement,” targets expansion from approximately 600 per year to 2,000 per year. The April 9th contract obligated $264.96 million in FY26 Army procurement plus $4.496 billion in FY26 Foreign Military Sales funds. Boeing’s seeker capacity is being tripled in parallel. The scale is large. The ramp is slow. The binding constraint is again solid rocket motors.

SM-3. The most constrained interceptor. Stockpile December 2025 was 414 per CSIS. The U.S. used roughly 80 in the Twelve-Day War. SM-3 Block IB unit cost has risen from approximately $9 million in FY21 to roughly $12.5 million in FY24, reflecting the expiry of the FY2019-FY2023 Multi-Year Procurement contract. The larger Block IIA variant runs closer to $24 million per round. The FY26 request proposed ending IB production in favor of IIA, trading quantity for capability. This one is tight. If there’s a real ballistic-missile-defense gap, it’s here, and frameworks alone won’t solve it.

SM-6 and AMRAAM. Both tracking the same pattern. SM-6 baseline 100 to 125 per year against a 500-plus framework target. AMRAAM baseline 800 to 1,000 against 1,900-plus. Neither is binding in a Taiwan scenario, but both are on the ramp.

LRASM and JASSM-ER. These are the munitions that matter in a Taiwan scenario, and the WSJ piece does not mention them. CSIS’s wargames, run roughly two dozen times between 2022 and 2023 and updated since, found that the U.S. global LRASM inventory of roughly 450 missiles was expended within three to seven days in every direct-invasion scenario. JASSM stocks were depleted around day nine. Budget reality puts these wargame rates in context. Air Force JASSM base-budget procurement was 550 units in FY24, 450 in FY25, and 389 in FY26 (discretionary plus reconciliation combined); Ukraine-era supplementals added units in FY24 that are not repeated in the outyear baseline. Combined Navy-plus-Air Force LRASM procurement was 118 in FY24, 115 in FY25, and 238 in FY26 (with Navy driving most of the FY26 increase). One technical point that matters: LRASM shares more than 70 percent of its hardware with JASSM-ER and uses the same production assembly line, so the two should not be read as independent capacity pools. Lockheed’s announced combined ramp target of 1,000 per year is still well short of the 5,000-long-range-missile consumption rate CSIS forecast for a three-to-four-week invasion fight. That consumption rate is itself calibrated to an optimistic PLA assumption, so the real gap is narrower than the raw comparison suggests. It remains the tightest binding constraint in any honest accounting.

L3Harris (the old Aerojet Rocketdyne). Worth naming specifically because the bottleneck sits here rather than at prime final assembly. Camden currently produces more than 115,000 solid rocket motors per year across a 2,000-acre site with over 100 buildings. In November 2025, L3Harris broke ground on an Advanced Aerojet Production Facility at Camden designed to increase large-SRM manufacturing capacity six-fold, with lease commencement in January 2028 and a 20-year term. A 2023 DPA Title III agreement initially funded $215.6 million of expansion. Subsequent DoD SRM awards through September 2025 totaled more than $120 million across nozzle production, insulation materials, metallic cases, rayon precursor and carbon phenolic ablatives, attacking the input layer of the bottleneck rather than final assembly. The April 23rd $1 billion strategic investment then added an equity-adjacent layer on top. The Bloomfield, Indiana munitions campus anchored by Prometheus Energetics is designed to support Tomahawk, Javelin, Stinger and GMLRS, with phased operations beginning in 2027. The bottleneck is real. The response to the bottleneck is already substantial, pre-committed, and structured to run for a decade.

If you wanted to write a serious piece about American munitions inadequacy for a Taiwan contingency, you would write about LRASM, JASSM-ER, SM-3 Block IIA, solid rocket motor capacity at L3Harris, and the seeker chip supply chain. The WSJ piece writes about Tomahawks and Patriots, which are the wrong weapons for the scenario being invoked but the right weapons for the political argument being made. Tomahawks and Patriots have enormous Congressional-district footprints across Tucson, Huntsville, Andover, Camden and Troy. LRASMs and solid rocket motors do not have the same district-level political purchase, at least not yet, even though Camden itself straddles both categories.

The six-year replacement figure is not defensible against the post-February industrial plan. It is defensible only if you freeze production at the FY26 request level, which would be to ignore the largest missile production expansion since the 1980s, announced during the same administration, roughly eighty days before the article ran.

One honest qualification about what the arithmetic in this section actually measures. The munitions analysis above is calibrated primarily to an invasion scenario, where LRASM, JASSM-ER and SM-3 Block IIA do the binding work. Our own probability estimates place blockade as the more likely Chinese action by a factor of roughly one-and-a-half to five, depending on where inside the ranges the true values sit. A blockade scenario is not fundamentally a munitions problem. It is a surface fleet sustainment problem, a logistics resilience problem and a signaling problem, with Chinese Coast Guard and maritime militia (rather than the PLA Rocket Force) as the primary instruments. The industrial arithmetic above still matters for deterrence credibility in a blockade, because what restrains Beijing from escalating a blockade into a strike is the American magazine visible behind it. But the reader should hold in mind that the piece is arguing rigorously about the less likely of two plausible Chinese actions. A full treatment of the blockade case, with a different industrial ledger behind it, is work for a follow-up.

VI. The Pattern, The Actor, The Prediction

Every leak has a beneficiary. When the Wall Street Journal prints an anonymous-officials story about a munitions gap two days after the largest defense budget in American history and three weeks before a summit with China, ask: who wins?

Every Pentagon leak benefits the Pentagon. Every defense industry story benefits the primes. If “someone gains” is the test for manufactured narrative, almost nothing in Washington reporting survives it. The distinction needs drawing sharply.

Three features separate the April 23rd story from routine pre-markup messaging.

First, its timing relative to the Beijing summit, not merely the budget markup. Pentagon leaks ahead of House and Senate Armed Services markups are a recurring pattern, roughly annual, and expected by anyone who watches the cycle. Pentagon leaks three weeks before a presidential summit with a strategic rival, calibrated to shape that summit’s negotiating posture, are structurally different.

Second, the specific numerical claim (six years to replace) is mathematically inconsistent with publicly-announced framework agreements signed by the same administration roughly eighty days earlier, which makes the story an assertion of capability gaps that the administration’s own industrial policy is built to close.

Third, and most telling, the munitions featured in the story (Tomahawk, PAC-3) are not the munitions most binding in the scenario invoked (LRASM, JASSM-ER, SM-3 Block IIA). A leak optimised for Congressional district politics foregrounds Tucson, Huntsville, Andover and Troy. A leak optimised for genuine strategic warning would foreground the solid rocket motor supply chain at Camden and the seeker chip supply chain. The WSJ piece foregrounds the first set. None of these features is a tell alone. Combined, they distinguish this leak from baseline pre-markup Pentagon messaging with enough clarity to warrant the reading.

There’s a useful historical template, recent enough the mechanism should be visible. Through 2022 and 2023, a visible pattern of anonymous-officials reporting emerged around HIMARS rocket depletion, Stinger stockpile gaps, and solid rocket motor bottlenecks in the Aerojet supply chain during the Ukraine drawdown. The scarcity was real. Ukraine had exposed single points of failure that the Pentagon had allowed to atrophy for two decades. What the stories accomplished was not to create the gap. The gap already existed. What the stories accomplished was political. They secured unprecedented multi-year procurement authorities, Defense Production Act investments, and long-term recapitalization contracts that had been impossible in peacetime. The real shortage was leveraged into structural capital recapitalization that serves the primes for a decade.

The 2026 leak is the same mechanism with a different adversary. Real shortage. Authentic capacity gap. Legitimate concern about a Taiwan contingency inside a finite window. All of it true. All of it also being used to sell a more than $3.2 trillion (per CRFB analysis) decade-long expansion that was approved in outline before the first Tomahawk flew, in a political environment where any serious “no” vote can be framed as a vote for Chinese invasion. The HIMARS pattern is the pattern. The WSJ piece is its 2026 repetition.

If there is a single institutional actor worth understanding in this pattern, it is the Pentagon’s Office of Legislative Affairs, whose documented function is to coordinate messaging between the building and Congress in service of the building’s resource asks. That is OLA’s job, it does it well, and it does it by a set of methods refined over decades of practice. The HIMARS cycle of 2022 and 2023 is the clearest recent case study in how the function operates: real shortage, authentic gap, selective leaks to credible reporters, multi-year procurement authorities secured as the downstream result. The April 23rd story matches that template closely enough that a reader shown the two cycles side by side, without dates attached, would struggle to tell them apart. Whether any specific individual at OLA had any specific conversation with any specific WSJ reporter isn’t a question we can answer. It’s also the wrong question. The mechanism is institutional, not personal.

It may also be emergent rather than coordinated. Pentagon officials leak on individual incentives, reporters print on individual incentives, the primes benefit downstream. No central coordination required for the pattern to exist. The observable features of this leak (pre-summit timing, numerical inconsistency with announced frameworks, wrong-munition selection) hold either way.

What we can say: the template exists, has been used before, has worked before. What it produces looks observationally identical to what we read on April 23rd. A reader who wants to call the resemblance coincidental is free to. Explaining why it’s coincidental, given the timing (three weeks pre-summit, forty-eight hours post-budget-release, coincident with the L3Harris $1 billion close), is work we haven’t had to do.

The Pentagon wins. The story frames the $1.5 trillion FY27 request and the forthcoming roughly $200 billion supplemental as not merely reasonable but urgent. The American Enterprise Institute, a conservative think tank, has already called the proposal unbuildable, arguing “we lack the manufacturing ability and supply chain to produce weapons” at the envisioned scale. The Committee for a Responsible Federal Budget has called it “a major driver of a dangerous fiscal trajectory.” The story neutralizes both critiques. It makes a “no” vote on the supplemental a vote for Chinese invasion of Taiwan.

The defense primes win. Lockheed Martin, RTX, Boeing and L3Harris all have framework agreements or equivalent industrial-base commitments that pay out over seven years. The larger the framework, the larger the pull-through for components, sustainment, upgrades and foreign military sales. The WSJ story materially increases the probability that the frameworks are fully funded and that additional supplementals follow. Same dynamic the HIMARS stories produced in 2023, at larger scale.

The hawkish wing of the administration wins. The story gives Trump leverage in Beijing. “We have a munitions gap” is dual-use rhetorically: it signals to Xi “do not provoke us while we are rearming,” and it signals to Congress “do not block the supplemental.” The former is a risky signal to send. Advertising weakness to a strategic rival, even with a planned ramp behind the weakness, carries tail risk. That the administration is willing to send it anyway is consistent with the domestic budget calendar weighing heavily in the calculation, relative to the strategic one. The story ran two days after the budget and three weeks before the summit. Readers can draw their own conclusions about whether that timing is coincidence.

The media wins. Taiwan-fear stories are among the most reliable traffic drivers in American defense journalism. The WSJ writers earn internal credit for a scoop, Bloomberg earns terminal reads for the summary, X earns engagement, and Substacks earn subscribers. Including this one. We are not pretending otherwise.

The following people pay. American taxpayers, for a CRFB-estimated $3.2 trillion decade-long defense expansion on top of a roughly $39 trillion national debt that Jerome Powell has described as being on an unsustainable fiscal path. Beneficiaries of the $73 billion in domestic discretionary cuts used to partially offset the defense increase, which hits education, housing, agriculture and health. Gulf allies waiting in line for Patriot refills they have been told will not come. Ukraine, which has now been forced to substitute for the U.S. in Gulf air-defense markets, signing deals with Saudi Arabia, the UAE and Qatar for cheaper anti-drone systems. And, downstream, American bond markets, which have to finance all of this at yields that continue to drift higher in a decoupling from the Treasury curve we have written about before.

We’re not arguing the spending is unnecessary. We’re arguing it’s being sold through a narrative that hides half the ledger and leverages an industrial shortfall, real but bounded, into a structural recapitalization that outlasts the shortfall by a factor of ten. The honest version would say: “China is also weak. Our industrial base can be rebuilt. A $3.2 trillion ten-year expansion has large opportunity costs. Here is the trade-off.” The version being sold says: “China is strong, we are weak, please approve the budget.” The first version is harder to write. The second version writes itself. The WSJ piece is the second version, executed with precision, in line with a template the Pentagon has been refining for four years.

VII. Probabilities and Falsification

Probabilities over certainties. Here’s where we stand as of April 24th, 2026.

Subjective estimates, pattern recognition and historical analogues weighted against available evidence. Not a formal model. Calibrated priors for what to watch, not point forecasts. Each anchored to a small set of driving facts.

Probability that China militarily invades Taiwan between now and end of 2027: low, perhaps 5 to 10 percent. The combination of Rocket Force purges, the Zhang Youxia investigation, the ongoing summit diplomacy, the Wang Yi-Pyongyang trip, the Busan truce framework and Xi’s own stated 2026 preference for “mutual respect, peaceful coexistence” all cut in the same direction. China’s timeline, to the extent one exists, is longer than the invasion-is-imminent framing suggests.

Probability that China imposes a Taiwan blockade or grey-zone escalation between now and end of 2027: medium, perhaps 15 to 25 percent. Blockade is cheaper, less risky, and does not require the Rocket Force to work as advertised. CSIS’s 2025 blockade wargames concluded that the American response is harder to calibrate than in the invasion case, which paradoxically makes blockade more, not less, attractive to Beijing. If we are wrong on the invasion probability being low, we are more likely wrong about blockade than about assault.

Probability that the “munitions gap” narrative dominates FY27 budget debate: very high, perhaps 80 percent or more. This is already happening. The narrative is sticky, bipartisan and industry-supported. Even if Congress trims the supplemental, the framing survives.

Probability that the roughly $200 billion supplemental passes substantially intact: medium-high, perhaps 60 to 70 percent. Slender Republican majorities, midterm-year politics and CRFB opposition cut against, but the WSJ-style messaging campaign and summit urgency cut for. The base case is a modestly trimmed but directionally intact supplemental by late summer.

The named prediction. We expect the next iteration of this leak cycle to land between mid-September and late October 2026, timed to the FY27 continuing resolution negotiations and ahead of the winter NDAA conference. The featured munition will shift from Tomahawk and PAC-3 to LRASM, JASSM-ER, or SM-3 Block IIA, because those are the weapons whose actual constraints most align with the scenario being invoked, and the Pentagon’s messaging operation will eventually run out of rhetorical mileage on Tomahawks. The headline will cite “officials familiar with the assessment” or equivalent. The framing will contain the phrase “cannot sustain” or “cannot support” in relation to a Pacific contingency. If that story does not run, on roughly that timeline, against one of those munitions, this piece’s pattern-recognition framing is weaker than we believe, and we will say so.

What would falsify the “manufactured scarcity” thesis beyond the named prediction. Three things would force us to reconsider. First, if the supplemental is defeated or deeply cut (sub-$100 billion final), it would suggest the messaging campaign is less effective than we estimate, and therefore less deliberate than we argue. Second, if the Trump-Xi summit produces a genuine Taiwan de-escalation, it would suggest the summit was the point, the budget was secondary and the WSJ story was pre-summit theater rather than post-summit budget armor. Third, and most importantly, if China launches a visible military action against Taiwan (blockade, quarantine, missile strike) before the end of 2026, it would mean the threat was real, the warning was accurate, and we were wrong to read the timing as manipulative.

What would falsify the PLA-degradation correction, specifically the claim that CSIS consumption rates overstate the threat: the first serious PLA rocket-force exercise in the Western Pacific that demonstrates synchronized launches from multiple brigades without detected command failures. That would reset the readiness estimate upward, narrow the transition window’s protection, and put more weight back on the alarmist reading of the WSJ piece. We are watching for this specifically.

Any of these should update readers, including us, toward the straightforward reading of the story. In the absence of these, the manufactured-scarcity reading and the degraded-PLA reading together remain the more parsimonious pair.

One caveat to our own probabilistic claims. When we describe the leak as “consistent with coordinated messaging,” we do not claim to read the minds of anonymous officials. We cannot. We can only note that the pattern of timing, numerical choice, and munition selection matches what coordinated messaging would look like better than it matches what independent reporting would look like, and that the historical template from 2022 to 2023 provides a prior. A coincidence interpretation is less parsimonious.

Conclusion

The Wall Street Journal didn’t write a fake story. It wrote a true story on a timeline set by the people who leaked it. The munitions are depleted. The industrial base is strained. The Taiwan scenario is harder in 2026 than it was in 2023. All correct.

What the story leaves out is that the industrial scale-up was committed to before the war, that the Chinese side of the ledger is in worse condition than the American side, that the six-year figure is mathematically incoherent against the production frameworks already signed, and that the 2027 date which anchors the whole discourse rests on an inferential leap from a Chinese domestic readiness directive to a Western operational deadline. The story as written asks readers to accept a framing in which Iran caused the crisis, China will exploit the crisis, and only money can solve the crisis. The more defensible framing is that the crisis was priced in before Iran, that China has its own crisis larger than the one being described, that the 2027 window is softer than the reporting suggests, and that money is one of several tools, used in a budget fight that is the point of the leak.

The transition window matters. For the next two to three years, while L3Harris solid rocket motor capacity is the binding constraint, the American munitions position is compromised for a peer fight in the Pacific. That compromise is real. It’s not fabricated. Any reader treating the underlying analysis as false instead of opportunistically timed is making the opposite mistake from the one the WSJ framing invites. The gap closes, but not by next Tuesday. The policy question inside the window (don’t start another war we can’t sustain) is a different question from the one the leak is engineered to ask (approve the supplemental).

The two countries facing each other across the Pacific are both less formidable than their budgets suggest, and both more formidable than their readiness reports imply. The wargames used to prove American inadequacy are calibrated to a Chinese force other analyses, from the same intelligence product, describe as degraded. Reconcile the two sets of numbers, you get where the Pacific actually is. The WSJ isn’t writing that piece. The Pentagon isn’t writing that piece. The defense primes aren’t writing that piece. No incentive to. The frameworks pay out regardless.

One thing to take from this. The next six months will see more leaks, more “officials said” stories, more urgent assessments, more Taiwan imagery in American defense reporting. Watch the timing. Watch who gains. Watch what is not being said about the other side of the ocean. And watch the ramp schedule on Tomahawks, PAC-3s and THAADs, because by mid-2028 the gap will close, the budget case will need a new crisis, and the story that replaces this one is already being drafted by people whose jobs depend on drafting it.

We will be here when it runs.

Subscribe now

Share

TSCS covers commodities, macro and capital flows for institutional readers. Our paid research includes bespoke commodity reports, reflexive analyses of industrial and financial cycles, and deeper work on the names behind the frameworks discussed above. Subscribe below.

This post is for informational and educational purposes only. Nothing in it constitutes investment advice, a recommendation to buy or sell any security, or a solicitation of any financial transaction. Readers should conduct their own research and consult a qualified financial adviser before making any investment decision. TSCS and its authors may hold positions in the securities or sectors discussed.

Before I update it, one piece of live business. I published a palladium kill switch on April 18. It’s one COT print from firing. Friday 3:30pm ET, the CFTC publishes April 21 data. The April 14 print was -1,808 managed money net short, down from -2,541 the week before. Twenty-nine percent of the short book covered in a week. One more print like that and I cut the futures leg in half, regardless of what Commerce does three days later on the AD final.

Why the switch fires before the catalyst: the trade was sized for a forced cover. If shorts are already running, the squeeze weakens. The AD print still moves spot. You just don’t get paid for it. Base case on Commerce: final at or near 132.83%, no postponement.

The actual price path of the last three weeks was violent.

Front-month crashed 9.1% on April 17 when Iran said the Strait was reopened, rallied 5%+ on April 20 when Iran closed it again, and has been bouncing around $95-99 since.

Back end drifted about $3 lower. Curve compressed from $27 backwardation to $16.

The curve hasn’t caught up to the physical evidence. It hasn’t run away from me either. Narrower claim than three weeks ago. Good enough reason to rebuild this piece around the scoreboard.

The next two weeks resolve more than the last twenty-two did. April 28 collapses BOJ, FOMC day one, and the Commerce AD final. Alamos and Agnico Eagle print. Mosaic on May 11. The palladium COT on Friday. Behind the wall: what I’m watching, what kills each thesis, and the full scoreboard of every call in the series.

Subscribe now

The Scoreboard

A reader asked for this and the ask was sharp enough that I’m making it permanent. A scoreboard that only shows the wins is marketing. The allocators reading this have the tools to mark me to market either way. Better to do it myself than make them do it.

A ledger of every substantive call, mostly from the last 90 days, plus older ones whose status has moved recently. Publication date, thesis, horizon, current status. Four buckets: Active (still in horizon, not falsifiable yet). Resolved correct (validated). Resolved wrong (falsified or tracking against me). Superseded (a later piece replaced the framing).

Read more

Managed money is net short 2,541 contracts of NYMEX Palladium.

A month ago they were flat. Three weeks ago they were net long.

They have been adding roughly 500 contracts per week to a losing position for three weeks running.

Into a four-catalyst regulatory stack that begins in seven days.

In a market one twenty-fifth the size of COMEX gold.

Where no Russian producer defended itself, no US OEM publicly opposed the petition, and no administrative off-ramp exists.

This is wrong-way positioning into a binary with no escape valve. It is the single most mispriced setup I have found in twelve months of reading trade-remedy dockets. No sell-side note has it.

That is the trade.

The 132.83% headline is the setup. Not the thesis.

The setup is a 241.93% combined duty wall. AD since February 19. CVD since March 11. Built on Adverse Facts Available because no Russian producer participated in the investigation.

The thesis is what happens between April 24 and June 19 when four separate statutory catalysts land into a market where the speculators are offsides going in.

The asymmetry is not in the margin. It is in who is wrong, and what they do when they realize it.

SBSW closed at $13.19 on April 15. Palladium spot sits at $1,572 at the April 17 close, up 71% year on year. All position and price data here are stamped to the April 17 close unless otherwise noted.

RBC is Outperform at $17 (cut from $18 on February 19). BMO is Market Perform at $18 (lifted February 3). HSBC carries the highest Street target at $24.80.

Every institutional desk covering PGMs has seen the 132.83% headline. The priced trade is “long the only US primary palladium producer into a favorable trade ruling.” That trade is available, directional, and mostly exhausted at current levels.

The trade that is not priced is the short-covering dynamic in a 15,000-contract market where hedge funds have been building into the wrong side of a stacked binary for three weeks. With no organized US downstream opposition and no administrative off-ramp for Russia.

That is what this piece is about.

A note on the apparent contradiction.

SBSW is crowded long. PA futures are crowded short. The reflexive read is that one cohort is wrong and the trade resolves through capitulation into the other.

That read is incomplete.

The two positions are different time horizons on the same metal. SBSW longs are pricing a 12 to 24 month rerating on tariff-protected US production with operating leverage to Pd above $1,500. Futures shorts are pricing six to twelve week macro: EV penetration, a thin contract that screens as a sell on momentum, and the secular thrifting story that has been right for three years.

Both can be internally rational. Neither is responsive to the catalyst window.

The window does not adjudicate which view is right on the metal in 2028. It adjudicates whether the next eight weeks contain a forced flow that runs against the futures position.

To be clear. If the AD final on or around April 24 prints affirmative at or near 132.83%, the macro short does not have a thesis-consistent reason to cover. The order does not change global Pd supply-demand math. Only US basis.

The cover comes from risk management, not thesis change.

Risk-managed covering in a 15,000-contract market with 46.6% of the short side held by hedge funds is the entire trade. The squeeze does not require the shorts to become bulls. It requires them to become smaller.

This also means the trade has a hard expiry. If managed money does not cover into the April 24 to June 19 window, the squeeze thesis is wrong and the position closes regardless of where SBSW trades.

The equity leg can survive on fundamentals past June 19. The futures leg cannot.

Sizing reflects this.

Who the trade is for

Read more