How we use AI.

Every figure on this site — every breadth percentage, put/call ratio, indicator reading, percentile, and historical episode — is computed by our own data pipeline from sources we name on each page: our maintained ~5,500-stock daily price database, FRED, FINRA, the OCC, and SEC EDGAR. The pipeline runs after every close, verifies completeness before computing, and publishes the raw series for download. No number on this site is generated by a language model.

We use AI (currently Anthropic’s Claude) as an engineering and writing assistant: building the tools and charts, analyzing our own datasets, drafting and editing the plain-English readings and Insights essays, and pressure-testing arguments before they publish. It is a powerful collaborator, and pretending otherwise would be silly.

A human (the founder) reviews what publishes: the analytical stances, the claims, the thresholds in every “regime” rule, and the editorial calls. Insight essays go through external review before and after publication. When we update an analysis, we bump its modified date; when we get something wrong, we correct it on the page rather than quietly rewriting history.

The protections aren’t promises — they’re structural. Every claim is checkable against data we publish ourselves (the catalog). Every reading is dated to the close it was computed from (live status). Every methodology is documented, including where we deviate from classic specifications (methodology). And our analytical essays state in advance what would change our mind — so you can hold us to it.

Nothing on this site is financial advice; see the disclaimer.

Each Insights essay is engineered to answer one real question, and it carries a named human byline because a person stands behind the analysis and the call. The argument, the stance, and the thresholds in every “what would change our mind” section are written and owned by us; AI assists the drafting and the data wrangling, never the judgment.

The original work is the data. Every chart, scorecard, percentile and historical table is computed live from our own CSVs and labelled with its method — each post carries a “How we checked it” section and links the primary FRED/BEA/BLS/OECD series, not just our copies. We publish falsifiable claims with dated checkpoints, correct errors on the page, and ship the raw data so anyone can reproduce or refute the numbers.