Controlled application logic applies defined scoring, thresholds, status rules and workflow conditions.
The responsible user checks the inputs, rationale and resulting decision before approval.
RPG EXCELLENCE · AI ASSURANCE
Useful assistance. Visible limits. Accountable decisions.
RPG Excellence uses structured automation and decision support to help people organise evidence, identify priorities and prepare controlled outputs. Technology can assist the work; it does not inherit the authority or professional accountability of the user.
No automated or assisted output becomes an approved assurance conclusion merely because the platform produced it.
Approval remains attributable to an authorised person and must be supported by appropriate evidence.
What the platform does today—and where the boundary sits.
This register separates deterministic platform functions from future or provider-enabled generative capabilities. It will be revised when the processing method or customer data flow changes.
StatusCapabilityHow it worksHuman control
Approved platform records are assembled into structured reports and draft narrative using application rules and templates.
Named preparers, reviewers and approvers remain responsible for the issued report.
The platform highlights missing information, due activity, risk conditions and suggested next steps.
A prompt cannot approve an audit, accept a risk, close a finding or verify effectiveness.
RPG Excellence does not treat an external generative model as an autonomous assurance decision maker.
Any future provider-enabled feature must be documented, assessed and released under the controls on this page before customer use.
Eight guardrails applied across assisted features.
Defined purpose
Each assisted feature must have a stated user need, permitted use and prohibited use.
Minimum necessary data
Users should provide only information relevant to the task. Passwords, recovery codes, secret keys and unnecessary sensitive personal data must not be entered.
Evidence before output
Assisted wording does not become objective evidence. Conclusions must remain traceable to controlled records, observations, interviews, samples and applicable requirements.
Human review and authority
An authorised and competent person remains responsible for relevance, accuracy, context, approval and any action taken from the output.
Explainable calculations
Where the platform calculates risk, priority, progress or status, the relevant inputs, method and threshold should remain visible or documentable.
Role and tenant boundaries
Assisted functionality must not bypass subscription entitlement, organisation scope, user permissions, approval gates or administrative controls.
Testing before release
Material features are reviewed against intended behaviour, known failure modes, inappropriate reliance, access boundaries and representative use cases.
Challenge and correction
Users can reject unsuitable output and report material errors, privacy concerns or security issues for triage, correction and feature review.
Customer information is not an unrestricted AI input.
Purpose limitation
Information should be processed only for the documented feature purpose and within the user's authorised organisation.
Provider transparency
If an external model provider is introduced, RPG Excellence will identify the purpose, relevant data flow, provider role and available user controls before customer use.
No silent authority transfer
Introducing an assisted feature does not change user roles, subscription access, approval authority or accountable ownership.
PROHIBITED RELIANCE
What assisted functionality must not do.
- Issuing or representing an accredited certification decision
- Closing a nonconformity or CAPA without authorised verification
- Accepting residual risk on behalf of an accountable owner
- Replacing a task-specific safety, legal, medical or engineering assessment
- Creating evidence that was not observed, supplied or independently verified
- Overriding organisation access, approval or segregation-of-duty controls
From proposal to controlled operation.
Define
Purpose, users, inputs, outputs and prohibited reliance
Assess
Privacy, security, access, bias and failure scenarios
Test
Representative cases, boundary checks and human review
Release
Guidance, limitations, version control and monitoring
Review
Errors, user challenge, material changes and withdrawal
CHALLENGE, ERROR OR CONCERN
Report a concern →Users should question output that does not fit the evidence.
A material accuracy, privacy, security or inappropriate-reliance concern should be reported with the feature, record reference and reason for challenge. RPG Excellence can then determine whether correction, guidance, restriction or withdrawal is required.
Public control statement · Version 1.0 · Reviewed 17 October 2026. This page will be updated when material functionality or processing changes.