Improve · Candidates
Keep a promising candidate separate from an active release
A candidate is an output proposed for further use, not an automatic replacement for a deployed model. SAVRN Cloud’s planned release workflow separates evaluation, approval and promotion so each decision has its own evidence. The public preview demonstrates this sequence with synthetic artifacts. Connected candidate releases are Coming soon.
Preserve the experiment behind the candidate
Open a preview candidate to examine the training context that produced its local record. The planned connected equivalent would identify a real artifact and its lineage, including the base model, dataset revision and recipe. A reviewer needs this context to understand what changed and to determine whether the comparison against an earlier release is meaningful.
Evaluate before seeking approval
The preview requires candidate evaluation using a different dataset identity from training. Its deterministic results must satisfy the selected rubric before the approval step becomes available. This demonstrates a release gate, not real quality assurance. Production review would need measured results, critical failure analysis and a documented rationale for accepting the remaining limitations.
Record the decision and its authority
A separate demo reviewer and a note accompany local candidate approval. This helps illustrate why a score and a release decision are different records. In a connected service, reviewer identity, authorization and any separation-of-duties requirement must be enforced. A typed name in a browser form cannot establish actual independent approval.
Promote with a recovery path
Promotion creates a named synthetic deployment in the preview. Rolling back retires that local deployment while leaving the candidate approved for inspection. The planned connected workflow would need to distinguish artifact readiness, deployment creation and traffic activation, with observed evidence for each. Recovery must identify what is restored and whether new work can safely resume.
Service maturity
Explore the workflow today
Evaluate, approve, promote and roll back synthetic candidates within the browser-local workflow.
Before connected service access
Connected release management requires verified artifacts, authenticated reviewer authority, controlled activation and tested rollback.
What this makes possible
The intended outcome is a release history that explains which evidence supported promotion, who authorized it and how the change can be reversed.
Build toward the work that matters.
Tell SAVRN what your institution needs to run, who reviews the results and where its data must stay. That workload defines the next service to qualify.
Discuss an AI project