Traceable source
Each package identifies the systems, owners, time period, and export path behind the records.
AstralEdge dataset standard
Built for serious AI work.
STANDARD / 01
Every package is designed to answer the questions that matter before ingestion: where the data came from, what changed, what may be used, and where the limits are.
Clear licensing
The permitted use, recipients, duration, restrictions, and ownership position are defined before delivery.
Privacy-led preparation
Personal details, credentials, privileged material, and unrelated records are removed from scope.
Training utility
Records are filtered, structured, deduplicated, and documented without stripping away the context that makes them useful.
Delivery integrity
Stable identifiers, checksums, manifests, and version control keep every delivery inspectable.
Lifecycle control
Recipient and version records support corrections, updates, and targeted withdrawals when required.
Delivered with the data
The package explains itself.
Technical teams get the records and the material needed to evaluate, ingest, version, and govern them.
Source brief
Where the data came from, who controls it, and which period the package covers.
Data card
Purpose, contents, preparation, known limitations, and intended model use.
Schema
Fields, types, relationships, identifiers, and representative values.
Quality report
Filtering, duplication, language, depth, completeness, and review results.
License scope
Permitted uses, restrictions, duration, onward access, and deletion requirements.
Delivery manifest
Files, record counts, versions, and checksums for technical acceptance.
Source a dataset
Tell us what the model needs to learn.
Share the domain, task, format, language, volume, and evaluation goal. We will determine whether an existing or purpose-built package fits.
Discuss your requirements