Every engagement asks the same three questions, examines the same ten assurance areas, and ends in a decision that leadership can act on. Only the scope changes.
Can it be trusted not to cause harm or break under pressure?
Does it actually do the job — on real tasks, not a scripted demo?
Is it worth what it costs?
Graded separately — never blended into one number that could hide a serious gap.
Every control is rated on what can be proven, not what is intended. “Practised but not documented” is treated as a gap, because that is how an auditor or a buyer will read it.
An agent does not just answer — it acts. We test every stage.
Interprets the request, the goal and the constraints.
We test: Ambiguous, adversarial and out-of-scope requests.Gathers context: conversation, permissions, system state.
We test: Missing, stale or manipulated context — including hidden instructions.Finds the right data, document or tool for the next step.
We test: Wrong-record retrieval, permission leaks, tool selection.Calls the tool, writes the record, sends the message.
We test: Irreversible actions, argument correctness, least privilege.Checks the result; recovers or hands over when it fails.
We test: Tool failures, silent errors, loops and escalation behaviour.Every dimension Good or better, no open critical issue, major issues within agreed limits.
No open critical issue; named issues fixed and re-tested before scale-up.
Any open critical issue, or any dimension graded Poor. Blocked until re-evaluated.
See every deliverable →Methodology in detail →The report & scorecard →
If the vendor says 5% errors and your tasks produce 12%, the business case is wrong from day one. Try your own numbers.
1,400 more bad outcomes every month than the business case assumed.
An independent evaluation measures the real rate before you commit, rather than after the invoices arrive.
Our method is aligned with and informed by these frameworks. We do not assess conformity against them. Standards alignment →