Client-provided
The client identifies the source and represents its authority to provide the material under the project agreement.
Provenance, consent & rights
A dataset is only useful when its use can be explained.
TrainLayer treats source, permission and permitted use as part of the dataset specification—not paperwork added after collection.
Rights lifecycle
From intended use to documented handoff.
01Define the use
Document the model purpose, intended users, geography, modality and whether the data supports training, evaluation, benchmarking or another approved purpose.
02Establish the source
Record whether data is client-provided, licensed, contributor-collected, publicly sourced under an assessed basis, or synthetically generated.
03Confirm permissions
Capture the relevant agreement, consent language, license or client representation and connect it to the permitted scope.
04Screen the content
Apply project rules for personal data, confidential material, minors, regulated categories, prohibited sources and other sensitive content.
05Preserve lineage
Track transformations, annotation, filtering, exclusions and version changes so the delivered dataset can be explained.
06Enforce limitations
Carry intended use, restrictions, retention and deletion expectations into dataset documentation and handoff.
Consent-aware collection
Consent must match the actual program.
Contributor-facing language should describe what is collected, why, how it may be used, whether participation is voluntary, applicable compensation, retention expectations and how questions or withdrawal requests are handled where relevant.
TrainLayer does not describe consent as universal or permanent. The applicable scope depends on the language agreed for that program and the governing contract or law.
Source categories
Different sources require different evidence.
Contributor-collected
Recruitment, notice, consent and compensation records are connected to the relevant collection protocol.
Licensed or public-source
License terms, access conditions and intended use are assessed; public accessibility alone is not treated as unrestricted permission.
Synthetic
Generation method, seed inputs, filtering, model dependencies and human verification are documented.
Restricted data
Some data should not enter the workflow.
Project rules may prohibit unlawfully obtained material, credentials, highly sensitive identifiers, unnecessary medical or financial information, exploitative content, data involving minors without an approved basis, or material whose rights cannot be reasonably established.
Where uncertainty cannot be resolved, exclusion is preferred over unsupported assumptions.
Return to security and governance ↗