Distributed compute operators
Coordinate providers across locations while preserving job identity, Runtime Execution Evidence requirements, and recovery controls.
The strongest fit is not “any AI workload.” It is a workflow where multiple parties need a bounded execution record, a clear verification decision, and safer conditions for payment or recovery.
Coordinate providers across locations while preserving job identity, Runtime Execution Evidence requirements, and recovery controls.
Turn verification policy, workload commitments, and result data into a repeatable platform boundary.
Give reviewers a clearer basis for release, dispute handling, provider payout, and future audit preparation.
SDDT is most relevant when execution is distributed, verification cost matters, and the organization needs more than a provider’s self-reported completion signal.
Customers, schedulers, providers, verifiers, and settlement systems each need a consistent record of what happened.
Checkpoint and failover paths matter because an interrupted job should not silently become a new or duplicate job.
Different workloads require different evidence. The verification path should be explicit rather than assumed.
No use case bypasses provenance, hardware-to-workload binding, independent review, or the current release gates.
A useful evaluation can begin with the workload, evidence required, provider boundary, failure path, and settlement decision that matter most to your team.
The recorded Google Confidential Space + Intel TDX + NVIDIA H100 run is PASS_NON_PROMOTABLE, classified as ATTESTED_WORKLOAD_AND_GPU_CORESIDENCE. Verified signature and identities plus an observed CUDA probe support that recorded co-residence claim, not authenticated execution/output binding.
Strict binding is BLOCKED_BY_VENDOR_PRIMITIVE_AVAILABILITY: no supported primitive/API was identified in documented public NVIDIA and Google interfaces reviewed as of October 2026. Additional internal testing alone cannot establish Tier 3. A vendor-supported mechanism or newly documented supported path must be separately qualified; it would not automatically open the gate.
These use cases describe potential controlled evaluations, not customer adoption or production authorization. Agent Trust Runtime v1.0's 13 local/synthetic scenarios and the public continuity/provider-failover demonstration are separate from H100 evidence.