Skip to content

Certification Runs in VB-OS

Certification extends single evaluation to composite compliance verification. A certification model defines the boundary version and evidence definitions required for verification, and a certification run executes the evaluations and produces a composite result.

A certification model is versioned. Each model version defines:

  • Boundary version: the specific compiled boundary version (boundary_version_id) that evaluations run against, with one boundary version per certification model version
  • Evidence definitions: a list of evidence definition objects (evidence_definitions: list[dict]) describing the evidence fields required for each evaluation

A certification run executes a certification model version:

  1. The platform runs evaluations against the boundary version specified in the model, using the provided evidence
  2. All evaluations are recorded as individual immutable evaluation records
  3. The run produces a composite status: RUNNING while in progress, then one of the terminal statuses

Run statuses:

  • RUNNING: evaluations are in progress
  • PASSED: all evaluations produced ASSERT
  • FAILED: at least one evaluation produced DEFER
  • ERROR: a system error prevented completion
  • CANCELLED: the run was cancelled before completion
run = client.certify(
model="payment-compliance",
model_version=2,
environment="production",
evidence={
"transaction_amount": 15000,
"account_balance": 42000,
"risk_score": 35,
"kyc_status": "verified"
},
metadata={"initiated_by": "release-pipeline"},
)
print(run.status) # "PASSED", "FAILED", "RUNNING", "ERROR", or "CANCELLED"
print(run.total_evaluations) # Total number of evaluations in the run
print(run.passed_evaluations) # Count of evaluations that produced ASSERT
print(run.failed_evaluations) # Count of evaluations that produced DEFER

Certification runs produce structured reports containing:

  • Each boundary evaluated and its verdict
  • Failure reasons for any DEFER verdicts
  • Evidence used for each evaluation
  • Timestamps, binary versions, and integrity hashes

Reports can be exported as PDF for auditor review. Each report is traceable back to the individual evaluation records.

Certification runs are immutable after creation. The run record, the individual evaluation records, and the report are all preserved as tamper-evident artifacts. The certification model version used is recorded, so the run can be audited against the model that was active at the time.

Before deploying a release, run a certification model that verifies all production readiness boundaries. Only deploy if the certification passes.

Define a certification model matching regulatory requirements. Run it periodically and produce auditor-ready reports showing continuous compliance.

Verify that third-party integrations meet your organization’s verification requirements before activation.

  • Evaluations: the individual records within a certification run
  • Boundaries: what certification models reference
  • Projects: where certification models live