Skip to main content
Nexus

One workflow / 10 evidence questions

What evidence exists before an AI trial?

Choose one proposed workflow, not the whole organisation. The questions organise what is recorded about ownership, affected people, the current baseline, action boundaries and the process itself. The result points to missing evidence. It does not predict whether a system will work.

Note / how to read the result

This self-check borrows themes from the voluntary NIST AI Risk Management Framework, its Generative AI Profile and public management-system context from ISO/IEC 42001. It is not affiliated with those bodies and does not test conformance. Your answers never leave this browser.

0 / 10 answered

  1. 01 / Ownership and purposeIs one person accountable for the proposed workflow and its consequences?
  2. 02 / Ownership and purposeIs the intended outcome defined separately from buying or deploying an AI tool?
  3. 03 / Context and affected peopleHave you mapped the people who use, receive or may be affected by the output?
  4. 04 / Context and affected peopleIs personal, confidential or regulated data in the workflow documented?

    For personal data in Mauritius, review the Data Protection Act 2017 and obtain qualified advice where needed.

  5. 05 / Baseline and evaluationDo you have a measured baseline for the current process?
  6. 06 / Baseline and evaluationCan you test the workflow on representative cases before live use?
  7. 07 / Action and review controlsAre the system actions and human approval boundaries explicit?

    Payments, access changes, contract terms and external commitments deserve explicit gates.

  8. 08 / Action and review controlsCan a reviewer see evidence, escalate uncertainty and reverse a bad action?
  9. 09 / Process evidenceAre the normal steps and meaningful exceptions written down?
  10. 10 / Process evidenceIs the authoritative source for each input and decision known?
Answer every question to see your picture.

Framework context

  1. AI Risk Management Framework NIST, 2023Voluntary framework context. The self-check is not a NIST assessment.
  2. Generative AI Profile, NIST AI 600-1 NIST, 2024Risk and action context for generative AI.
  3. ISO/IEC 42001:2023 overview ISO, 2023Public overview only. This tool does not assess ISO/IEC 42001 conformity.
  4. Data Protection Act 2017 Mauritius Data Protection Office, 2017Primary legal source. Obtain qualified advice for a specific workflow.

Last reviewed: 2026-07-18

Not sure where agents would even apply?

The candidate-pattern explorer gives you a starting list of questions about source records, action boundaries and evaluation evidence for common workflows.

Explore candidate patternsBuild a value scenario