Keep the ending
out of the question.
We separate the rules used to select examples from the later outcomes used to assess them.
The sequence
- State a concrete question and comparison.
- Record the universe, dates, features and missing-data rules.
- Use only information available at the observation time.
- Compare against simple alternatives and suitable shuffled controls.
- Reserve fresh periods for fixed tests when possible.
- Publish the result, including failures and remaining gaps.
Different levels of evidence
Study 01 has a registered volatility-family check on later data. The TIA case search is exploratory: we inspected results and refined descriptive screens. Those are different evidence levels, and we label them differently.
Several coins in one crash are not several independent market events. Overlapping future windows and correlated assets can make a result look more certain than it is. Recognition accuracy is separate from forecast accuracy.
Implementation and costs
Numerical measurements come from reproducible code. AI can help organize, query and explain records; it does not replace arithmetic or transform an association into a cause. A forecast also needs realistic costs, liquidity and execution checks before any trading claim.
The website publishes a selected set of research snapshots rather than credentials, account records or operational services.
Open the glossary ›