Weekly Market Pulse by Anthony Rosenthal
How this letter builds its numbers, scores its calls, and decides what it will not print. Everything here is standing method. It changes rarely, and when it changes the date below moves.
Last updated: 7 September 2026
Every week the Weekly Market Pulse publishes a crash gauge, a running score in an argument about artificial intelligence, a set of break conditions on the compute trade, a scoreboard, and a public verdict on its own calls. This page explains how each of those is built, once, so the edition itself can carry this week's numbers rather than a reprint of the manual.
The test that governs all of it is simple. A reader holding the edition and this page should be able to take the published inputs, do the arithmetic themselves, and arrive at the same figure the letter printed. If they cannot, the number is an assertion rather than a measurement, and it should not have been published.
A weekly score from nought to one hundred describing how many of the conditions that usually precede a large market fall are currently in place.
It is not a forecast, and it is important to be exact about that. The gauge measures preconditions, not outcomes. Conditions assemble and then dissipate without anything happening, which is the normal case. What the score answers is a narrower and more useful question: do the next ninety days deserve more caution than the last ninety did?
Eight signals are read every week. Each is scored nought if it is benign, five if it is unsettled, and ten if it is stressed, against thresholds fixed in advance so the judgement cannot be made after the reading is known. Each signal carries a weight, because a widening in corporate credit tells you more about approaching trouble than a quiet week in oil does.
| Signal | What it watches | Weight | Benign (0) | Unsettled (5) | Stressed (10) |
|---|---|---|---|---|---|
| High-yield credit spread | The extra interest riskier companies must pay to borrow, compared with government bonds. It widens when lenders get nervous. | 15% | under 350bp | 350 to 450bp | over 450bp |
| MOVE index | Expected turbulence in government bonds. The bond market's equivalent of the VIX. | 15% | under 110 | 110 to 130 | over 130 |
| Yield curve, ten-year minus two-year | Whether long-term borrowing costs more than short-term, as they normally should. The signal fires as the curve returns to normal after being inverted, which is historically when trouble arrives rather than during the inversion itself. | 15% | under −0.25 | −0.25 to 0 | above 0 |
| ISM new orders | New orders placed with American manufacturers. A reading under fifty means order books are shrinking. | 12.5% | above 50 | 48 to 50 | under 48 |
| VIX, level and trend | Expected turbulence in American shares over the next month. | 12.5% | under 20, stable | 20 to 28, or rising | over 28 |
| Share of the S&P 500 above its 200-day average | How broad the market is. A rising index carried by a handful of names is a narrower market than the headline suggests. | 10% | over 60% | 40 to 60% | under 40% |
| Insider selling clusters | Sectors where company executives have been selling their own shares together over the past four weeks. | 10% | no sectors | one sector | two or more |
| Energy shock | Crude oil, read at two speeds at once. A sharp move feeds through to costs everywhere. The worse of the one-week and four-week readings is the one that counts, and falls score nought, because an oil collapse shows up in the credit signals already being watched. | 10% | 1wk under +5% and 4wk under +10% | 1wk +5 to 10%, or 4wk +10 to 20% | 1wk over +10%, or 4wk over +20% |
Each signal's score is multiplied by its weight expressed as a fraction, the eight results are added together, and the total is multiplied by ten. If every signal were stressed, that is ten multiplied by one point nought nought multiplied by ten, which is one hundred. If only the yield curve were stressed, it is nought point one five multiplied by ten multiplied by ten, which is fifteen.
The edition prints the eight readings, the score each one earned, and that arithmetic, every week. It is checked before publication by running the printed sum exactly as a reader would and confirming it returns the published figure.
A running score in the central economic argument about artificial intelligence, settled one week of evidence at a time.
Both sides of this argument believe the technology is powerful. They disagree about what it does to an economy.
The Displacer holds that machines substitute for people. Work that used to be done by salaried humans gets done by capital instead, the gains concentrate among the owners of that capital, wage income falls relative to profits, and eventually there is nobody left with enough income to buy what is being produced. On this view the danger is a demand shock.
The Augmenter holds that machines make people more productive rather than unnecessary. Output rises, costs fall, new categories of work appear that nobody had thought of, and the gains reach ordinary people through lower prices and higher wages. On this view the effect is a supply expansion, and it is a good one.
Four questions are asked each week, and whichever side the week's evidence favours takes that point.
| Pillar | A point for the Displacer | A point for the Augmenter |
|---|---|---|
| Adoption speed | Capability jumping faster than forecast; the technology visibly improving itself. | Real friction slowing deployment: regulation, liability, cost, organisations that will not change. |
| Labour dynamics | Jobs going, with the technology named as the cause. Whole roles replaced, not tasks. | Output rising without headcount falling. New categories of work appearing. |
| Type of shock | Wages falling behind profits. The consumption gap widening. | Real incomes rising. Gains reaching people through prices or pay. |
| Compute constraints | The cost of computation falling below the cost of the person it replaces. | Energy, chips and financing costs forming a ceiling that slows substitution down. |
Two disciplines keep the score honest. One fact cannot win two pillars in the same week, and it certainly cannot win two in opposite directions, so before a score is published the four pillars are checked to be resting on four different observations. And every score ends with a sentence naming the specific evidence that would move next week's points the other way, so the running total stays falsifiable rather than becoming momentum.
The running total was reset to nought at Week 20, in May 2026, when the framework took its current form. Scores from before that date used a different method and are not comparable.
A weekly reading on whether the scarcity underneath the artificial intelligence trade is actually breaking.
A toll bridge is only worth something while there is no second bridge. That is the whole idea. Every layer of the computing stack, the chips, the memory, the power, the data centres, earns a premium because it is a bottleneck, and the premium starts to fall on the day the bottleneck clears. This reading exists to say plainly when that is happening, and just as plainly when it is not.
It reports rather than scores. There is no composite figure and no band, deliberately, because averaging several weak indicators into one confident number is how a measurement acquires more authority than its inputs deserve.
Each condition carries the date it was last verified. If that date is older than the interval at which the underlying data refreshes, it is published as not observed and counts for nothing. A condition is never carried forward silently as though it were current.
Reported alongside, and deliberately not folded into the count: how far the seventeen listed companies in this book sit below their recent peaks, and the worst single drawdown among them. A thesis can be right while a position is painful, and the two facts are kept apart so that neither disguises the other. Also reported, and explicitly not scored, are the rental rates for high-end computing hardware and an operating cost floor that has not yet been supported by a full cost build.
Each week the reading names the constraint currently binding and whether it has moved, because capital attacks a bottleneck, capacity arrives, and the bottleneck relocates to somewhere the original list did not cover.
Every company named in On the Radar is scored in public, on rules fixed before the result is known.
The problem this is built to solve is the oldest one in financial commentary. Anyone can be right in hindsight if they are allowed to decide afterwards what counted as being right. So four things are written down and locked at the moment a call is first published, and none of them can be edited later.
Performance is measured against that benchmark and never against zero. A share up fifteen per cent while its sector rose nine and a half is up five and a half, not fifteen. A call that beats nothing it could have been held instead of is not a success.
Each call is marked at four weeks and again at its thesis horizon. The four-week mark is timing information and is kept internal, because a four-week reading on a twelve-month thesis says almost nothing about whether the analysis was correct. The thesis horizon is the verdict, it is published, and it is a minimum of twelve months, because a shorter window sets an analytical view up to fail for reasons that have nothing to do with whether it was right.
The single exception is the pre-registered invalidation event. If that fires, the call closes immediately, at any age.
| Score | Meaning |
|---|---|
| +2 | Vindicated. Beat the benchmark and the named catalyst arrived. |
| +1 | Working. Beat the benchmark and on track. |
| 0 | Open only. Behind the benchmark, no invalidating event, and inside the single non-renewable four-week grace period. This is never a closing verdict. |
| −1 | Failed softly. Behind the benchmark past the time stop, with no catalyst. |
| −2 | Failed hard. A named invalidation event fired. |
Beating the benchmark means beating it by more than one percentage point. That margin is fixed in advance, so a photo finish cannot be talked into a win after the fact.
The benchmark decides the sign; the thesis decides the depth. A call that beat its benchmark can never be scored minus one, whatever happened to the story, because minus one is defined as being behind. If a call beat its benchmark but a named invalidation event fired, it scores minus two. The worked example is a rare-earths position closed in August 2026. It fell almost twenty-four per cent while its sector fell over twenty-seven, so it did beat its benchmark. It still scored minus two. Outperforming a sector that fell twenty-seven per cent does not vindicate a thesis whose whole claim was that the company would stop being priced like that sector.
Whenever a running success rate is published, it is published with the number of calls it is drawn from, so a reader can reconcile it rather than take it. A track record that cannot go down is not a track record.
Every company that has appeared in On the Radar continues to be tracked until its thesis horizon, at least a year from the first mention. A company moves out of the main section and into this table when there is no new catalyst that week: the analytical view is intact, but there is nothing fresh to say.
A new catalyst means something specific. It is a named, company-level event from the previous seven days, an earnings release, a regulatory decision, a contract, a filing, that has not already been used to justify an earlier appearance. A share price moving in a helpful direction is not a catalyst. That is exactly what a Portfolio Watch company looks like.
Twenty-six assets, tracked from their opening prices on the first trading day of 2026. Those starting values are locked and are never recalculated, because a year-to-date figure whose starting point moves is not a measurement of anything.
Only verified closing prices are published. Where a verified close is not available, the previous week's figure is carried forward, the edition says so, and the gap is logged rather than filled with an estimate. An estimated price is never printed as though it were real.
Two of the twenty-six are not on the automated feed and are handled by stated convention. The Baltic Dry Index is taken each week from a named, dated source. MSCI Emerging Markets is published on the index scale, derived from the verified close of a tracking fund at a fixed, locked ratio, and that derivation is stated in the edition.
Alongside each asset the Scoreboard carries its realised volatility over the trailing eight weeks, annualised, as a measure of how violently it has been moving rather than merely where it has arrived.
These four explanations used to be rebuilt inside the letter whenever the mechanism came up, which meant a weekly reader met the same picture many times over. They are set down here once instead. Nothing in a given week's edition depends on having read them.
Why a bottleneck is worth money, and why that stops. A toll bridge earns its money from the absence of a second bridge, not from the quality of its tarmac. Every layer of the artificial-intelligence build-out carries a premium only for as long as it is the narrow point: the moment rival capacity arrives, the premium starts to go, however good the asset remains. This is the idea underneath the Stack Inversion, and it is why that section watches supply arriving rather than demand growing.
Why longer borrowing usually costs more. Lending money for ten years is riskier than lending it for two, so ten-year debt normally pays more, in the way a savings account pays more when the money is locked away for longer. When that relationship breaks and short-term borrowing costs more than long-term, something unusual is being priced, which is why the shape of the yield curve is watched as closely as its level.
Why interest rates reach ordinary life slowly. A change in the central bank's rate does not arrive in a household's budget on the day it is announced. It arrives when a fixed-rate mortgage ends, when a business refinances, when a lease is renegotiated. That delay, usually measured in quarters rather than weeks, is why this letter treats today's rate decision as a statement about next year rather than this month.
Why the financial system is described as plumbing. Money moving between banks, funds and central banks runs through pipes that nobody notices while they work. Most of the time the pressure in them is irrelevant to anything you own. Occasionally it is the only thing that matters, and it moves faster than any published economic figure, which is why a handful of funding-market measures sit in the crash gauge at all.