Early
Recovery. Cheap money, few owners, nothing priced in.
Mid
Growth broadens. Profits rise faster than prices.
Late
Prices and ownership at records. Nothing broken yet.
Money tightens
Cash gets scarce. The cushion is gone and funding costs bite.
It breaks
Credit stress, then everything else.
0 of 3
Tripwires fired
Three things we decided in advance would mean the cycle has turned. Two firing
declares it, with no argument at the time.
15 of 54
At an extreme
Readings at or near the highest in their own recorded history. These say how
far things have gone, not that anything is wrong yet.
13
Worth watching
Stretched but not extreme. The middle ground, where most of the board sits.
Cloud capex cuts→
LEADS · quarters●●○the pre-committed tripwire
The five biggest AI spenders cutting their published plans. History says they cut together or not at all.
guidance · Q2
Credit speed→
LEADS · quarters●●●velocity leads level
Not the level of credit stress but how FAST it is changing. Speed matters more than level.
FRED · 11 Aug
Profit forecasts▲
COINCIDENT●○○watch-light: informs, cannot trigger
Watch-light only: it informs, it cannot trigger, because we cannot back-test it honestly.
EODHD · 10 Aug
Spending vs the economy▲
LEADS · quarters●●○1998-2001 crossed 7 qtrs before the top; n=1
How much of everything the country produces is going into computers and software, against its own ten-year average. Only five stretches this high since 1957; one of them was 1998 to 2001.
FRED NIPA · Q2
Borrowed share of the build▲
LEADS · quarters●○○~judgement; no prior AI cycle to calibrate on
How much of the AI spending is paid for with new debt rather than earnings. Above half means the build is being financed, not funded.
filings · TTM
The weakest borrower▲
LEADS · quarters●○○~the edge funds first and fails first
How much of its spending the most stretched buyer funds with debt. Above 100% means it is borrowing more than it spends. This is where the cycle breaks first.
filings · TTM
What is promised but not owed▲
LEADS · quarters●●○~$2.9trn across 5 filers; compounding faster than capex
Money these companies have contractually committed to spend, which does not appear as debt on their balance sheets. Microsoft's accounts show about six per cent of what it has actually contracted for.
SEC 10-K/10-Q
Interest burden→
LEADS · weeks●●○Minsky's own line, not a mined threshold
What share of the big spenders' cash flow goes on interest. This is the line that decides whether debt is a tool or a trap. Low means they can still stand still.
filings · TTM
Companies running out of cash→
COINCIDENT●○○~dominated by utilities; read the trend
How many of the 500 largest US companies spend more than they earn and would exhaust their cash within two years.
filings · TTM
Profits outside the AI trade▲
COINCIDENT●●○the circularity test; refuted 2026-08-14
How fast the other 400 companies are growing their cash flow. If this were flat, the profit boom would be one theme wearing a market's clothes. It is not.
filings · TTM
Leverage▲
LAGS●●○builds slowly; confirms rather than warns
Total debt against a year of cash flow. Under about two is comfortable for companies of this quality.
filings · TTM
The strongest borrower→
LAGS●○○~context for the line above, not a signal
The same measure for the best balance sheet in the group. The gap between this and the line above is the whole argument.
filings · TTM
Revenue per gigawatt of compute→
COINCIDENT●○○
OpenAI's revenue divided by the electrical capacity it runs on. If revenue grows faster than compute, the infrastructure is getting more productive and the boom is starting to pay for itself. If compute grows faster, each new data centre earns less than the last. This is the cleanest signal we have because OpenAI sits where the money from real customers enters the system, rather than circulating between the big companies. Flat so far: both have tripled together. ⚠️ The figures are reported rather than filed, so this is not yet citable in a published piece.
reported · 2025
Memory makers' gross margin▲
COINCIDENT●○○
What Micron keeps from every dollar of memory it sells, after the cost of making it. This is the scarcity end of the AI economy: prices so far above cost that they are guaranteed to attract new supply. The striking part is that cost of production has barely moved while revenue nearly quadrupled — the industry is not shipping much more memory, it is charging far more for the same.
filings · May Q
Model makers' gross margin→
COINCIDENT●○○
The other end of the same economy, and the number that would complete the picture. OpenAI and Anthropic are private and disclose nothing, so this stays blank rather than being filled with a plausible guess — a made-up figure here would close the gap on screen and hide the very thing worth watching. The tile below is the market-based substitute.
⛔ no primary exists
What a model charges over what it costs→
COINCIDENT●○○
Closed models' prices divided by the price of open models of the same age, served by competing hosts. Because anyone can serve an open model, its price sits close to the cost of running it — so this ratio is roughly what the big labs charge above cost. The letter calls this the master signal: if it widens, the labs are keeping the savings from cheaper computing; if it narrows, competition is handing those savings to customers and the model business is being competed away. Comparing models of the same age is what keeps it fair, since newer models are simply better.
marketplace · weekly
Memory makers' spending growth▲
COINCIDENT●○○
How fast Micron is increasing what it spends on new plant. High prices always call forth new supply, and this is that supply being paid for. It is the reason today's shortage is not a permanent condition.
filings · TTM
Memory capacity already signed for▲
COINCIDENT●○○
What SK Hynix has contractually committed to spend on new capacity but has not spent yet, from its half-year accounts. Unlike an announcement, this is a signature. It is tomorrow's supply, already paid for, arriving whether or not demand is still there when it lands.
H1 2026 accounts
The cash buffer▼
LEADS · quarters●●○~the drain sets when tightening bites
Money parked at the Fed overnight. It absorbed three years of tightening and peaked near $2.5trn. Once empty, further tightening hits banks directly.
FRED · 13 Aug
Central bank balance sheet▲
LEADS · quarters●●○policy acts with a lag
Whether the Fed is adding money to the system or taking it away, measured over the last quarter.
FRED · 12 Aug
Companies issuing shares▼
LEADS · quarters●●○2000 / 2009 / 2021 — n=3
Are companies BUYING BACK their own stock, or SELLING it to the public? Positive means selling. Since 2000 this has been negative 86% of the time; it turned positive in Q1. The only precedents are 2000, 2009 and 2021.
FRED Z.1 · Q1
Overnight funding▼
LEADS · weeks●●●funding stress shows here first
What banks pay to borrow cash overnight versus what the Fed pays them to hold it. While this is negative, cash is plentiful. It pushing positive is the first real sign money is getting tight.
FRED · 14 Aug
Bank lending▲
COINCIDENT●●○credit plumbing
Whether banks are actually extending credit. Credit growth is the plumbing beneath everything else.
FRED · 29 Jul
Household ownership▼
LEADS · quarters●●○records mark late cycle, not the top
How much of everything American households own is sitting in shares. When the public is already all-in, there is less money left to arrive.
FRED Z.1 · Q1
Semis in the index▼
LEADS · quarters●○○~2–5% normal, per one source
Semiconductor companies as a share of the whole US market. A normal reading is 2–5%. This is what the AI trade looks like measured as ownership rather than as a story.
EODHD · 14 Aug
Profit share of the economy▼
LEADS · quarters●●○mean-reverts over years
⭐ Company profits as a share of everything the country produces. Every valuation multiple divides price by earnings; this tells you whether those earnings can last.
FRED NIPA · Q1
How far above normal▼
LEADS · quarters●●○the reversion gap
The same measure against its own sixty-year average, in percentage points. Reverting this gap would cut aggregate profits by roughly a third.
FRED NIPA · Q1
Countries at record highs→
LEADS · quarters●○○~>60% line is Longview's, not ours
What share of major national stock markets are at an all-time high. Above 60% has been treated as a danger zone: when everything is at a record, there is nothing left to buy.
EODHD · 14 Aug
Giants vs the rest→
COINCIDENT●●○behaviour, not structure
Whether the ten biggest companies are beating everyone else. Near zero means the market is broad; strongly positive means a handful of names are carrying it.
EODHD · 14 Aug
Market trend▲
COINCIDENT●●●the oldest filter there is
How far the S&P 500 sits above or below its own one-year average. The simplest trend test there is.
EODHD · 14 Aug
Capex acceleration▼
LEADS · quarters●●○~2nd deriv turns first; trigger is 2 falling qtrs
Whether the big cloud companies are still SPEEDING UP their spending, not merely spending a lot. This is the trigger: two quarters of falling growth removes the funding pressure and the suppliers' revenue at the same time.
filings · calendar Q2
Capex vs cash flow▲
LEADS · quarters●●○~funding strain precedes cuts
How much of the building is paid for out of the business itself. Above 100% means they are borrowing to build.
filings · TTM
Funding ladder▼
LEADS · quarters●○○~private marks, single source
What private AI labs are paying to raise money. When the price they raise at stops rising, the money is getting tired.
private mkts · Jul
Data-centre supply→
LEADS · quarters●○○~manual capture, thin history
Whether there is spare capacity. Empty buildings and falling rents would mean the boom has overbuilt.
JLL/CBRE · Jul
Cloud growth→
LEADS · quarters●●○~the 2022 divergence template
Whether cloud revenue is still accelerating. The danger sign is spending going up while sales slow down.
filings · Q2
Return on new capex→
LEADS · quarters●○○~no baseline yet
Whether the money already spent is earning anything back. The question everyone avoids.
filings · 11 Aug
Capex growth▼
COINCIDENT●●○~is the suppliers' revenue, so coincident to them
How fast the big cloud companies are increasing what they spend. Their spending is the suppliers' revenue, so this number is the chip makers' order book.
filings · calendar Q2
Margin debt▼
LEADS · quarters●●●0–9m; 3 signals since 1997, no false positives
How much investors have borrowed to buy shares. Above +60% a year has marked three major tops since 1997 and nothing else.
FINRA · Jun
Small-investor mood▲
LEADS · weeks●●○contrarian, weekly since 1987
A weekly survey asking private investors if they are bullish. Contrarian: when everyone is optimistic, that is the warning. Only the extremes carry information.
AAII · weekly
Shares above trend→
COINCIDENT●●○participation, now
What share of the 500 largest US companies are trading above their own one-year average. Broad participation, not just a few giants.
EODHD · 2026-08-21
Market vs its own trend▼
COINCIDENT●○○
Where the US market sits against its own one-year average. The bluntest question there is: is the trend still up? It is allowed to disagree with everything else here.
EODHD · 2026-08-21
Countries at new highs→
COINCIDENT●○○
How much of the world is hitting new highs at once. Counter-intuitive: everything rising together is a late-cycle sign, not a healthy one. Above 60% is the danger line.
EODHD · 2026-08-21
Single-stock stress▲
LEADS · weeks●●○~led Jul-23 and Jul-24 pullbacks
How jumpy individual shares are compared with the index. A calm index hiding violent moves underneath is a classic late-cycle tell.
CBOE · 2026-08-21
Fear gauge (VIX)▼
COINCIDENT●●○contrarian, coincident
What options cost, which is what investors pay to protect themselves. Contrarian: cheap protection means nobody is worried, which is when trouble starts.
EODHD · 12 Aug
Cost of hedging→
COINCIDENT●○○
How many puts are bought for every call. Contrarian: a high number means people are paying up for protection, which is fear, which is usually the better moment. A low number is complacency.
CBOE · 2026-08-21
New highs vs new lows▼
COINCIDENT●●○35–50% preceded pullbacks
How many shares are near a one-year high against how many are near a low. 35–50% has preceded major pullbacks.
EODHD · 2026-08-21
Overbought share▲
COINCIDENT●●○mechanical, mean-reverts fast
What fraction of shares have run too far too fast. A low number means there is room left, not that something is wrong.
EODHD · 2026-08-21
Turbulence→
COINCIDENT●○○~2 readings of history
The average fall of the five worst shares each day. Persistent punishment beneath a calm surface.
EODHD · 2026-08-21
Liquidity→
LEADS · quarters●●○tightens before the economy rolls
How easy it is to borrow money in the system. This tightens BEFORE the economy slows, so it is the early warning.
FRED · 12 Aug
Yield curve→
LEADS · quarters●●●inverts 12–18m before recession
The gap between long and short-term interest rates. When short rates exceed long ones, a recession has usually followed.
EODHD · 12 Aug
Credit spreads▲
LEADS · quarters●●●widens before equity breaks
The extra interest risky companies must pay over the government. Almost every market break shows up here first.
FRED · 11 Aug
Real interest rates→
COINCIDENT●●○policy stance, now
The interest rate after inflation. Above zero means policy is squeezing the economy rather than helping it.
FRED · 1 Jul
Company profits▲
COINCIDENT●○○cannot be back-tested — watch only
Whether analysts are raising or cutting their profit forecasts across ~200 large companies.
EODHD · 10 Aug
Recession risk▲
LAGS●●●claims and unemployment lag
A scorecard of ten warning lights across borrowing costs, credit and jobs. Low means the economy is not signalling a downturn.
FRED · 12 Aug
Colour marks the exceptions only — a plain tile is a normal reading, not a missing one.
LEADS means the dial moves before the event; coincident means it moves with it.
Dots are how much evidence stands behind it: three is a tested threshold, one is judgement.
A reason beginning with ~ is our own estimate rather than an established finding.
Sparklines appear where we hold real history. Internal working document; it shows our
thresholds. Research, not advice.