Concepts

These pages explain the ideas behind Starfold, one idea per page. Read them in this order, or jump to the one you need. If you are short on time, read How Starfold chains the tools first: it names every stage and says which page covers it.

Page	Read it to learn
UMAP: a map of your table	Why a map of your table distorts distances, and which map settings Starfold searches
HDBSCAN: dense clumps and noise	How dense clumps become groups, why some rows stay noise, and which clump settings Starfold searches
Optuna: searching the settings	How the search proposes settings, and why its score must never see labels
How Starfold chains the tools	How one call chains the tools in six stages, and where each output is explained
Copies and gaps	Why a score alone is not evidence, what a structureless copy is, and how the gap differs from the score
Refits: reproducibility, not truth	What a refit measures, what co-association is, and the words Starfold allows for the result

To run Starfold before reading these pages, start with Your first run.

• UMAP: a map of your table
• HDBSCAN: dense clumps and noise
• Optuna: searching the settings
• How Starfold chains the tools
• Copies and gaps
• Refits: reproducibility, not truth