A focused, evidence-backed check for consequential workflows where approval, evidence, policy or execution state may change before a tool action completes.
$1,000 introductory check · one protected action · two scenarios · one concise findings report.
Tell us which tool action changes data or system state and what approval it needs. No credentials in the first email.
Agree a staging or synthetic fixture, two concrete transitions, access conditions and explicit exclusions.
Exercise the agreed transitions and retain the available evidence, including effect data only when independently observable.
Receive evidence references, PASS / FAIL / BLOCKED / UNVERIFIED outcomes and a call to discuss next steps.
The assessment focuses on the interval between an initially valid decision and a later consequential effect. It asks whether stale authority, stale evidence, uncertain execution or recovery logic can still create an effect that should no longer be allowed.
Revoked or superseded authority versus the eventual consequential effect.
Material basis changes and selective invalidation of dependent state.
Retries or replacement actions when the prior consequence may already have committed.
Whether recovery need is incorrectly treated as fresh permission to execute.
The protected effect, authority sources, enforcement point and explicit exclusions.
Agreed cases, observed behavior and retained artifacts for each tested transition.
PASS / FAIL / BLOCKED / UNVERIFIED results with evidence references and claim boundaries.
Prioritized control changes plus one bounded retest where included in scope.
One consequence-bearing action and two agreed transitions. Written findings and one technical review call.
Target 2–3 business days after access and prerequisites are ready, with schedule confirmed during scoping.
Request scopeMore transition families, additional evidence collection, remediation priorities and a retest if contracted.
Timeframe, access and price depend on the agreed workflow and evidence boundary.
Discuss extended scopeNot included in the $1,000 check: production changes, penetration testing, custom adapter development, remediation implementation, a retest or certification. Tests require explicit authorization and a suitable non-production environment. No payment is taken on this website.
One agreed, consequence-bearing AI-agent tool action and two concrete changes in authority, evidence or execution state. We observe whether the intended control holds under the exact test conditions.
A fixed-scope check of one action, two agreed transition scenarios, a concise report with evidence references and limitations, and one technical review call.
No. Testing requires an explicitly authorized staging or synthetic environment and agreed evidence access. Custom integrations and production changes are not included.
No. PASS applies only to the tested scenario and observable evidence. Unknown outcomes are marked BLOCKED or UNVERIFIED as appropriate; no general security certification or compliance attestation is offered.
For verification boundaries and reproducible public research, review Technology & Evidence and the RAB-1 controlled benchmark.
RAB-1 is the current public example of a bounded EvidenceBound result: frozen scenarios, deterministic gates, retained evidence and exact-tag reproduction on NVIDIA OpenShell.