Explicit construction
Known mathematical structure should be represented directly rather than hidden inside an embedding or inferred again for every downstream task.
About
An open research project for deterministic scientific data analysis.
FeatureGraph turns declared analytical decisions into repeatable computations, inspectable results, and reproducible study records.
Mission
Make the reasoning behind scientific data analysis explicit.
A scientific dataset contains more than values. Its files, columns, timestamps, labels, and annotations have a declared structure, and an analysis makes decisions about what to group, measure, compare, and check.
FeatureGraph records those decisions and executes them deterministically so results can be inspected, reproduced, and tested across records or datasets. Scientific and domain experts remain responsible for interpreting what the results mean.
Principles
How the project is being built.
Deterministic semantics
The same observations and configuration should produce the same states, events, objects, measurements, and queries.
Inspectability
Construction features and object tables remain available so results can be traced to the procedure that produced them.
Domain independence
Behaviors are defined through mathematical structure rather than terminology tied to one industry or dataset.
Reproducible releases
Code, data access, environments, figures, tables, and manifests are released together and archived with stable records.
Machine consumption
The long-term interface is intended to be precise enough for other software to construct and reason over behaviors without natural-language mediation.
Project status
Beta released, research active.
The first public release includes oscillation objects, wave-derived accumulation objects, object and feature tables, dataset loaders, deterministic querying, automated tests, and reproducibility infrastructure. Transition objects are now implemented and Oscillation composes them without changing the Accumulation contract. Current work centers on formal semantic alignment, broader validation, a consumer demonstration, and the first paper.
Start with the release