Open research software · Beta
Scientific data analysis that is deterministic, inspectable, and reproducible.
A researcher writes a short specification describing what the data contains and what the study should calculate and check. FeatureGraph carries out the repeatable work, preserves how each result was produced, and leaves scientific interpretation with the people who know the domain.
Publications · start here
See the work, run the evidence.
Start with the BIDMC workflow, follow its frozen transfer to TEP, then examine interoperability with CLaP.
INTEROPERABILITY STUDY
Read the CLaP object study
See detector output become nine explicit temporal objects.↗
PREPRINT
Read the BIDMC multiscale preprint
Envelope scale as part of the specification: 53 records, a
contract frozen before the held-out run, and a cardiac-phase
result on 20 held-out records.↗
BIDMC STUDY
Read the current BIDMC study record
See the declared analysis, automated outputs, validation
evidence, and limitations.↗
The work FeatureGraph automates
Two layers of scientific data analysis, made repeatable.
Structural understanding identifies what each file, column, timestamp, label, and annotation represents. Analytical understanding defines what should be grouped, measured, compared, and checked; FeatureGraph carries out those declared steps and preserves the evidence, while researchers and domain specialists decide what the results mean scientifically.
Evidence · CLaP interoperability study
CLaP labeled the signal. FeatureGraph turned those labels into a study.
CLaP detected three recurring states and eight changes in the 20,700-measurement Crop time series. From a short researcher specification, FeatureGraph automatically grouped the unchanged labels into nine stretches, located every start and end, calculated their durations, and recorded the order in which they occurred.
Result: expanding the nine stretches back to individual measurements reproduced every CLaP label exactly. CLaP remained responsible for detecting the states; FeatureGraph carried out and checked the declared analysis.
Read the experiment recordOne concrete example
Watch measurements become an inspectable analysis.
The demonstration below shows which measurements belong together, where each interval starts and ends, and how its duration and amplitude were calculated.
Raw observations are shown. Reveal the object boundaries and measurements.
| Segment | Start | Peak | End | Duration | Amplitude |
|---|---|---|---|---|---|
| Objects not yet constructed. | |||||
Start with the research record
Inspect the objects, then test the contract.
DocumentationInstall, learn, and use the released beta.→
Research assistantDescribe a study, review the contract a model proposes, and
approve it before anything runs.→
Beta releaseSource, release notes, and reproducible artifacts.→
CLaP studyDetector output, object materialization, and exact
reconstruction.→
Research updatesShort notes on how the framework is evolving.→
The research program
Describe the analysis.
Run it reproducibly.
FeatureGraph is an open scientific data-analysis framework. Each study states what the source data and methods provide, automates the declared grouping, measurement, comparison, and validation steps, and records which conclusions still require scientific judgment.
01
Representation
Turn state and event sequences into bounded objects that people and software can inspect, measure, relate, and query.
- Reproducible studies and evaluations
- Open software and runnable demonstrations
- Explicit boundaries, properties, and provenance
- Citable datasets and research artifacts
02
Interoperability
Connect external detectors to a shared object representation without taking over their scientific role.
- Cross-detector and cross-domain evaluations
- Technical pilots and proof-of-concept studies
- Independent validation and failure analysis
- Research collaboration and integration