Access data without moving it
Some of the most valuable data for research and AI is also the hardest to share. It carries regulatory and contractual weight that data holders, the institutions that hold the data, won’t waive, and much of it is too large to move. The usual choice is a full copy or no access at all.
This paper sets out a middle path. It describes the architecture behind Platform 1.0, the controls that hold on every engagement, and how to choose an access pattern for your data and your compute.
White Paper 00 · Technical Overview · Available now
What’s inside
- Executive summary: The problem, the approach, and what data holders and buyers get, on one page.
- Why valuable data stays out of reach: Regulation, contracts and sheer size, and why handing over a copy is the wrong default.
- The architecture: Catalog, attestation and audit log: what each part does and what each one can see.
- A governed read, step by step: From approved request to logged read, with a sequence diagram.
- Six controls that hold everywhere: Copies, verification, credentials, revocation, logging and governance.
- Choosing an access pattern: Cloud storage or appliance, container or Peer-Collaboration Mode (PCM), and what changes at scale.
- Security and compliance considerations: How the controls map to the questions a security review asks.
- Glossary and references: Terms, standards such as GA4GH DUO, and further reading.
Figure 1, from the paper
LooporaData sits in the middle of every engagement for discovery, verification and logging. The read itself runs between the data holder’s storage and the buyer’s verified compute. The custody boundary stays with the data holder. Access controls and monitoring apply for the life of the analysis session. Controls are designed to prevent copying or export, and LooporaData never receives a copy.
Written for the people who sign off
- Security and privacy reviewers: What each part of the architecture can see, and where the controls sit.
- Data and IT teams at data holder institutions: What registration asks of your storage and network, and what it doesn’t.
- Buyer data science and platform teams: How your pipeline reaches a dataset, and what your team needs to get started.
Read White Paper 00
Coming next
- Coming next Cloud storage access: Where the data holder’s data lives · The data already sits in an object store the data holder controls. Verified compute reads it in place, with a credential scoped to one approved request.
- Coming next Instrumented container: How the buyer computes · The default way buyers compute. A trusted entrypoint inside a normal container sets the rules, on infrastructure the buyer already runs.
- Coming next Attestation: Trust and verification · The check every request passes before a credential exists: is this exactly the pipeline that was approved?
- Coming next Output review: Trust and verification · Two independent checks, one at registration and one on every run: what a pipeline is built to do, and what it actually tries to release.