Authority in force
The approved mandate, limits, conditions, and version applying when the action occurred.
Independent decision-and-execution assurance
Assess the decision. Verify the execution.
QAi verifies whether a consequential action followed the authority in force, received the required human decision, and matches what actually executed. It identifies missing evidence and population gaps without entering the live decision path.
Out-of-path assurance · SIMBA deterministic engine · Findings, not commands
CASE / REF-0427Synthetic banking case
01
AI recommendationApprove complaint refund
EUR 480 · 16:42:08 UTC
02
Authority at action timeREFUND-POLICY / v3.4
Complaints L2 · limit EUR 500
03
Human decisionApproved · Complaints L2
Decision owner recorded · 16:47:31 UTC
04
Execution evidenceNo receipt observed
Payment cannot be established
SIMBA findingInsufficient evidence
05Visibility across systems
Deterministic assessment
Attributable human control
Defensible evidence
The problem
One action · several systems
The decision trail is fragmented. Accountability is not.
Authority may sit in a policy repository, the AI recommendation in an application log, human review in case management, and execution in a separate operational system. Each system shows only part of the action.
When the action is challenged, teams reconstruct the chain manually. They may know what was recommended without knowing which authority applied, who decided, or whether the intended outcome actually executed.
01Was the action permitted?
02Who made the decision?
03What actually executed?
04Is the evidence complete?
Observation completeness matters. “No action occurred” is not the same as “no action was observed.” QAi keeps that uncertainty visible at case and population level.
How QAi works
Flight recorder + evaluator
One action. Four connected records. One explicit finding.
QAi establishes what happened across the workflow. SIMBA resolves the authority in force, checks evidence completeness, and evaluates the action using declared rules in a fixed order.
AI recommendation
What the AI proposed or did, with its source and event time preserved.
Human decision
The accountable person, role, decision, and time wherever human control was required.
Executed outcome
Independent evidence from the system that ultimately performed the action.
Outside the live path. QAi observes and evaluates. The institution’s systems continue to recommend, decide, approve, and execute independently of QAi availability.
Examine security & deploymentWhat QAi returns
SIMBA deterministic findings
A clear answer, including when the evidence is not enough.
Every evaluation resolves to exactly one of four terminal states. Missing evidence is never interpreted as success, and model-assisted checks may escalate a case but cannot close it.
Within boundary
Complete evidence supports the action within the authority that applied.
Outside boundary
The evidence shows that the action crossed a declared limit or condition.
Insufficient evidence
A required fact is missing, so a defensible conclusion cannot be reached.
Unobserved
An expected action or outcome was not seen in the relevant source.
Routing condition · not a resolution state
Human decision required determines who must act and what decision evidence must be recorded. It does not replace the four-state finding.
Synthetic example · REF-0427 · illustrative data
The recommendation was permitted. The payment remained unproven.
Authority and human approval were present, but no independent execution receipt was observed. QAi therefore refused to treat the overall action as complete.
QuestionEvidenceFinding
Was the decision permitted?Policy v3.4 · EUR 500 limitDEC.* · Within boundary
Was the refund executed?No independent receipt observedTRN.* · Unobserved
Overall case stateExecution remains unresolvedInsufficient evidence
Review the evidence
Download the pack Interrogate the complete synthetic case.
The downloadable pack contains the evidence inventory, seven institutional questions, findings, declared gaps, and reviewer challenges.
Why it matters
Operational assurance
Find gaps sooner. Reconstruct cases faster. Support a stronger review.
QAi should be assessed on practical outcomes: whether it reduces reconstruction effort, improves evidence coverage, identifies previously unknown gaps, and makes findings easier to reproduce and defend.
Operational value
Evidence that holds together under scrutiny.
The value is not another governance dashboard. It is a reliable connection between the decision, the applicable authority, accountable human control, and the executed result.
01Find hidden gapsSurface missing decisions, broken correlations, and unobserved outcomes.
02Reconstruct fasterBring the records for one challenged action into a coherent case.
03Review with confidenceTie each finding to the authority, evidence, limitations, and evaluation logic.
04Discover failures earlierExpose control and evidence failures before formal challenge or remediation.
Where the problem appears
One assurance pattern across regulated workflows.
Initial focus: banking. The product boundary is broader: consequential workflows where authority, human accountability, and execution evidence must remain connected.
Regulation raises the evidence standard.
QAi is not presented as required by a specific framework. Institutions and their advisers determine legal sufficiency. The product must earn its place through operationally useful findings and evidence.
Start with one workflow
Focused Evidence Sprint
Test QAi on a real reconstruction problem.
Choose one consequential action and a defined population. Connect representative records, run the deterministic assessment, and compare the result with the institution’s current review process.
Define the action
Agree the workflow, authority owner, source systems, population, and measures of value.
Connect the evidence
Map representative records, apply SIMBA, and identify incomplete or conflicting evidence.
Measure the difference
Compare reconstruction time, coverage, manual effort, identified gaps, and reproducibility.
hello@qai-global.com
Start with the action and current evidence problem. Share case data only through an agreed secure channel.