Section
Research
Two research engines, Market Cycle and Pre-Rally, study where the U.S. stock market stands in its cycle and what came before large rises in company shares. Every result here carries its date; where a method has been tested, it also shows how often it was right.
How this section is made
AI-made · Not investment advice
AI agents built and run these engines and wrote these texts and films; the figures come from the engines' own calculations, and mistakes are possible.
How this section is made
Prepared with AI. The research engines in this section were built and are run by AI agents under the direction of the site's owner, on the owner's own computer. AI agents also wrote these texts and made the short films from the engines' own records. The Market Cycle reading itself comes from a fixed statistical model, not from a language model. Mistakes are possible: every figure carries its date and its measured accuracy.
Measurements, not instructions. Each reading comes from a method fixed before it was tested and is shown with its measured error. It says nothing about what anyone should do with money. A method that was right in the past can be wrong now.
Market Cycle
Where is the U.S. stock market within its cycle of rising and falling phases?
How it works
Method and limits
Method
- The definitions and the test were registered before development; the model was frozen before its single test. The phase reading uses index closes only; the four-stage reading also uses volatility, breadth across industries, rotation across sectors and valuation.
- The method was chosen on data from before 1990 only and then frozen; the live models are refitted with that frozen method on all weeks known today.
- Each model was tested once, on weeks or markets it had never seen, and the result is published whether or not the target was met. Every reading since the engine went live on 1 October 2026 is stored unchanged.
What it does not show
- It reads the week assessed and says nothing about what comes next.
- The four-stage reading missed its accuracy target; the rising or falling phase has the stronger evidence.
- A turn-risk estimate showed no usefulness in testing and is not shown.
- Index levels and the other inputs are collected privately and are not shown.
Pre-Rally
What came before the largest one-year rises in the shares of companies that were already large, and can any set of conditions find those rises without also catching weak outcomes?
How it works
Method and limits
Method
- Find every rise of the defined size in the period and date the week it began.
- Search single conditions, pairs and triples on the earlier years for any that matched these rises without matching weak weeks, then check them once on later years kept aside.
- Measure ideas about causes against comparable weeks with the same volatility band, listing age and quarter, and repeat the search on shuffled histories to see what chance alone produces.
What it does not show
- Companies that left the exchanges are missing, so weak outcomes are undercounted.
- Rises cluster in a few years, so the sample is small and uneven.
- A higher rate for a group of weeks says nothing about any single company.
- Exploratory lists are not validated rules: a company in one met a condition and nothing more.
How these results stay current
How the results are kept current
The engines run on a private computer, not on this website. After each run, a checked summary of numbers and dates is published here; the page itself calculates nothing.
When that computer is off, the page keeps the last results with their dates. Each result carries the date by which a newer one is expected; a few days after that date passes without a new run, it is marked "Older than usual". Results that change only when a study is rerun are marked after several months. Nothing is estimated in between.
Never published: the engines' input data and the texts of their rules.