Fifteen indicators I track to judge how stretched the AI and market cycle has become — valuation, unit economics, leverage, positioning, and the triggers that would confirm a turn. Each is scored 1 to 10 against written thresholds and dated, so the record shows when my view changed and when it didn't.
The rows split into two groups that shouldn't be averaged together. Conditions measure how stretched the system is; triggers measure whether anything has actually broken. Extreme conditions alongside quiet triggers is a fragility reading, not a timing signal — and that combination can persist for years.
These are my own scores, not model output and not investment advice. Tap any indicator for its definition, sources and caveats. You can also download the whole scorecard as markdown and paste it into Claude or ChatGPT if you'd rather argue with it than take my word for it.
Equity Risk Premium
daily cadence
Measured as: ERP, percentage points
Direction: lower = riskier
Benchmark: Historical average around +4pp. Negative means equity earnings yield sits below the risk-free 10Y.
Source: Calc: S&P 500 earnings yield minus FRED DGS10
Notes: Trailing vs. forward earnings changes this by more than a point. Pick one convention and never mix them, or the series is worthless.
Bands: -0.5→10, 0.5→9, 1.5→8, 2.5→7, 3.5→5, 4.5→3, 99.0→2
Buffett Indicator
quarterly cadence
Measured as: Market cap / GDP, %
Direction: higher = riskier
Benchmark: Long-run average near 100%.
Source: FRED: WILL5000PR / GDP
Notes: Domestic GDP denominator against a globally-earning numerator biases this upward decade over decade. The level is less informative than the trend.
Bands: 200.0→10, 180.0→9, 160.0→8, 140.0→7, 120.0→5, 100.0→3, 0.0→2
Market concentration
monthly cadence
Measured as: Top 10 share of S&P 500, %
Direction: higher = riskier
Benchmark: Dot-com peak concentration was roughly 29%.
Source: Slickcharts / S&P index weights
Bands: 38.0→10, 34.0→9, 30.0→8, 26.0→6, 22.0→4, 0.0→2
AI capex intensity
quarterly cadence
Measured as: $ capex per $1 AI revenue
Direction: higher = riskier
Benchmark: Sustainable range is debated but single-digit-to-one is extreme by any infrastructure-buildout precedent.
Source: Manual: hyperscaler 10-Q capex lines, lab revenue disclosures
Notes: Softest number on the board. 'AI revenue' has no agreed definition — write down what you are counting in the period note or the ratio drifts silently.
Bands: 9.0→10, 7.0→9, 5.0→8, 3.0→6, 2.0→4, 0.0→2
Model commoditization
irregular cadence
Measured as: Open-weight share of enterprise workloads, %
Direction: higher = riskier
Benchmark: Cheap open-weight models erode the pricing power labs need to service data center debt.
Source: Manual: enterprise survey data
Notes: Adoption share is not spend share. A firm can 'adopt' an open model for trivial workloads while paying frontier prices for everything that matters. Prefer inference dollars if you can ever find them.
Bands: 85.0→8, 70.0→7, 55.0→6, 40.0→4, 0.0→2
Hardware input inflation
monthly cadence
Measured as: HBM module pricing, YoY %
Direction: higher = riskier
Benchmark: Rising cost per unit of buildout compresses returns on capex already committed.
Source: TrendForce
Notes: Cuts both ways. Bad for buildout economics, good for memory suppliers — so it is not a clean short signal on anything.
Bands: 150.0→8, 100.0→7, 60.0→6, 25.0→4, 0.0→2
High-yield credit spreads
daily cadence
Measured as: ICE BofA HY OAS, %
Direction: lower = riskier
Benchmark: Below 3% is historic complacency. Record low 2.41% in June 2007.
Source: FRED: BAMLH0A0HYM2
Notes: NON-MONOTONIC. Low spreads mean complacency; very high spreads mean the event is already underway, which the bands will score as LOW risk. If this ever blows past 6%, ignore the suggested score and read the level directly.
Bands: 2.5→10, 3.0→8, 3.5→7, 4.5→5, 6.0→3, 99.0→2
Margin debt
monthly cadence
Measured as: Margin debt as % of GDP
Direction: higher = riskier
Benchmark: 50-year median near 1.5% of GDP.
Source: FINRA margin statistics
Notes: Enter the ratio, not the dollar total. Dollars grow with the market and tell you nothing on their own.
Bands: 4.0→9, 3.5→8, 3.0→7, 2.5→5, 2.0→4, 0.0→2
Private credit defaults
quarterly cadence
Measured as: Default rate, %
Direction: higher = riskier
Benchmark: Private credit typically cracks before public equity does.
Source: Fitch private-credit monitoring
Notes: Marks are manager-supplied and lag reality. Treat improvements with more suspicion than deteriorations.
Bands: 12.0→9, 9.0→7, 7.0→6, 5.0→4, 0.0→2
Circular financing
irregular cadence
Direction: qualitative
Benchmark: Hyperscaler-to-lab compute credit loop holds until a lab misses revenue.
Source: Manual: earnings disclosures, deal reporting
Notes: Qualitative, so write the escalation rule down now while you are calm: one lab restructuring a compute commitment moves this to 8, a default moves it to 10.
Bands: none — qualitative, scored by judgment
Berkshire cash position
quarterly cadence
Measured as: Cash and T-bills, $B
Direction: higher = riskier
Benchmark: Rising cash signals inability to find value at prevailing prices.
Source: BRK 10-Q
Notes: Filing-driven, so it will always lag the fast rows. Also confounded by T-bill yields — cash costs less to hold when short rates are high.
Bands: 350.0→9, 300.0→8, 250.0→7, 180.0→5, 120.0→3, 0.0→2
VC funding velocity
quarterly cadence
Measured as: AI VC funding, $B per quarter
Direction: lower = riskier
Benchmark: A retreat in private capital would be early confirmation. Record inflows are the opposite.
Source: PitchBook / Crunchbase
Notes: Inverted: a LOW number is the risk signal here, which is why the bands compare downward.
Bands: 60.0→9, 100.0→7, 150.0→5, 200.0→3, 9999.0→2
IPO exit window
monthly cadence
Direction: qualitative
Benchmark: Exit liquidity for private marks. Open, narrowing, or shut.
Source: Renaissance Capital IPO monitor
Bands: none — qualitative, scored by judgment
Unemployment rise off low
monthly cadence
Measured as: Rise off 12-month low, percentage points
Direction: higher = riskier
Benchmark: A 0.5pp rise off the trailing 12-month low is the Sahm-style signal.
Source: FRED: UNRATE
Notes: Enter the RISE, not the level. 4.1% flat tells you nothing; 4.1% against a 12-month low of 3.6% tells you everything. This is the one row where the original scorecard was measuring the wrong quantity.
Bands: 0.5→9, 0.4→7, 0.3→5, 0.2→3, 0.0→1
Hyperscaler capex guidance
quarterly cadence
Measured as: Guided capex change, YoY %
Direction: lower = riskier
Benchmark: The bubble pops the day a major cuts spend and the stock is rewarded for prioritizing profit.
Source: Manual: quarterly guidance
Notes: Highest-information row on the board and the only one that is close to binary. Watch the market REACTION as much as the number — a cut punished by the market means something different from a cut rewarded.
Bands: -10.0→10, 0.0→8, 5.0→5, 15.0→2, 9999.0→1