01 / Statistical discovery engine
Define the population
Make features, outcome and time boundaries explicit.
A subgroup is a conjunction of feature conditions. The population and data preparation determine what its comparison means.
Statistical discovery engine
Find the segments that deserve a closer look.
Search for combinations of feature conditions whose outcomes differ from the population, then inspect quality, uncertainty and competing explanations.
Design and research · public preview
Subgroup / system sketch
How it works
Follow the system.
One part at a time.
Select a part to see its role, the decisions behind it, and what still needs evidence.
02 / Statistical discovery engine
Search the segments
Explore candidate conditions with indexed membership.
The engine includes bitmap-based indexing and optional GPU evaluation. Application-specific feature meaning remains with the caller.
03 / Statistical discovery engine
Score and test
Compare segment quality with statistical evidence.
A high score is not sufficient on its own. Hypothesis testing, multiple comparisons and calibration affect the interpretation.
04 / Statistical discovery engine
Validate elsewhere
Examine whether a discovered pattern survives.
Evaluation and held-out comparisons help separate a useful finding from selection effects. A subgroup difference does not establish causation.
Design principles
What guides the work.
General-purpose discovery
The algorithm is independent of trading or any one application.
Membership is inspectable
A segment retains the conditions and rows behind its statistics.
Selection-aware evaluation
Search and validation belong to the same interpretation.
Project status
Know where
the work stands.
Extracted algorithm and evaluation code. This site illustrates the workflow without querying a dataset or claiming predictive performance.
Source and project notesThe source repository requires authorized access. This public page contains a curated explanation and original schematic illustrations.