Epsilon: A Bring-Your-Own-Data Research Platform with Trusted Verification

We present Epsilon, an open-source platform for analyzing sensitive research data with per-execution verification. Epsilon addresses three practical barriers faced by researchers using Trusted Research Environments (TREs): (1) researchers experience challenges in using their own datasets without uploading data to platform infrastructure; (2) manual policy review creates bottlenecks in the submission workflow; and (3) existing TEE platforms provide only platform-level attestation with no per-execution proof binding specific code to specific datasets. Epsilon adopts a zero-trust model: no component outside the enclave is trusted, enabling secure cross-institutional data analysis without requiring institutions to share raw data. Epsilon introduces Archetypes, a bring-your-own-data mechanism by which researchers link their datasets without persistent data transfer. Only database credentials are stored, in an encrypted vault; data is not persisted on platform storage and is processed ephemerally within the TEE. An automated policy pre-filtering layer screens submissions before human review. A Verification Center extends standard TEE platform attestation with per-execution dataset and code binding, where the enclave passes a hash bundle (code, data, output, timestamp) to the Nitro Secure Module, which signs it as part of the hardware attestation document, producing portable verification receipts. Source code, SDK, and research templates are available on GitHub.¹