Agent Assurance
Registered packs can be sampled for independent testing and blind second review. Accountable reviewers retain scope, conclusions, and remediation decisions.
Inventory
Each registered version links its manifest to available evaluation definitions, guardrails, and retained records from captured executions.
Independent assurance
Re-runs a supported pack's eval sample and packages observed findings and limitations for independent review.
- hands the case to Subject: sanctions decisions
- hands the case to Subject: credit decisions
Re-derives another agent's decision blind and concurs, dissents, or escalates (maker-checker).
- hands the case to Subject: sanctions decisions
- hands the case to Subject: credit decisions
Subjects under review
Cerberus under review — its evaluation sample is rerun and its dispositions are independently re-derived.
Spark under review — decision behavior is challenged from outside; narrow field-exposure checks are not a fairness conclusion.
Resolve
Automated challenge evidence, findings, and maker-checker dissents land with accountable reviewers.
Tracking findings to closure. Open work — no Cadre pack is assigned yet.
The walkthrough
- 1
The registry links pack versions to their available evaluation definitions, guardrails, and captured execution records.
- 2
The assurance pack re-runs a supported pack's evaluation sample and inspects its observed execution paths, packaging findings and limitations for independent review.
- 3
Quorum re-derives sampled decisions blind — concur, dissent, or escalate — while the committee retains accountability.
Open example capture - 4
Challenge records and maker-checker dissents land with accountable reviewers; remediation tracking is the next open slot.