Audit Package
Audit Package
A bounded replay-risk audit packages intake, model, verifier result, trace evidence, remediation, retest, and a written Bounded Replay Verification Report.
The audit produces a decision artifact: what bad path exists, where the guardrail is missing, and what changed in the retest.
Audit Flow
What you provide vs. what the audit models
Plain-language input is acceptable. You do not need to arrive with a formal Boolean model. The intake stage accepts agent prompts, tool/function schemas, MCP/server tool lists, approval rules, workflow diagrams, runbooks, or product specs. The modeling stage converts that material into state variables, actions, preconditions, effects, a forbidden condition, a bound, and explicit assumptions and exclusions.
See the Service page for the starter package and the intake CTA, and the customer input template for the intake format.
Inputs Needed
- workflow description
- actions
- required approvals/confirmations
- dangerous outcome
- assumptions
- bound
- exclusions
Outputs Returned
- workflow model
- verifier result
- UNSAFE trace if reachable
- SAFE_WITHIN_BOUND result if no violation is found within model/bound
- remediation notes
- safe-variant retest
- assumptions and exclusions
Evidence Quality
Trace steps include state_before, state_delta, and state_after, making the counterexample reconstructable.