Owned evidence. Reproducible learning.
From sources to
MABOS training data.
Acquire admitted sources, prepare native MABOS corpora, and inspect the evidence behind each release.
Your data workspace
Connect a data credential
Use the dedicated credential issued for your workspace. Search preview access and reader credentials do not grant dataset or training access.
The credential is exchanged for a short-lived browser session. It is not saved in browser storage.
Add the selected search result
Save this discovery to an existing recipe. Its source policy still determines whether content can be acquired.
Recent builds
Discovery
Candidates retain their originating search and source relationship. Discovery is separate from collection and release.
Admitted sources
Use rights and source revisions are configured on the worker. Each build checks the current policy again.
Dataset recipes
Create a recipe
Dataset builds
A queued build has durable custody. A release appears only after acquisition, verification, and publication finish.
Examples and quarantine
Dataset releases
Inspect the manifest and export an admitted view. Sealed test content requires separate evaluation authority.
MABOS Data Foundry
Acquire the sources in a configured profile, compile role-specific corpus products, and verify their native admission. Numerical model training remains a separate stage.
Research training runs
Use configured research trainers for dataset experiments. Native MABOS numerical training is a separate stage.
Research model artifacts
Inspect experimental artifacts and their evaluation evidence here. Native MABOS deployment follows its model-specific runtime validation and installation contracts.
Research model evaluation
Compare a research candidate with its base and current model on reserved structured-output checks. Native MABOS qualification also requires task, formation, and runtime evaluation. Results stay within the evaluation workspace.