End of each iteration
At the end of each iteration, the team revisits the requirements, stakeholders, data representation, risk assessment and evaluation procedure.
What clause 9.3.4.3 asks
At the close of every cycle the team looks at five parts of the profile and changes any that need it: the bias requirements, the stakeholder set, the data record, the risk review and the evaluation method, including the list of what is watched. Those five are the whole record in small form. The cycle is not finished until each has been considered.
Why it matters
The five-part review is what keeps the profile alive. A monitoring finding can change the requirements, add a stakeholder, change the data map, change the risk assessment and change the monitoring list. If the team updates only the monitoring list, the rest of the record drifts out of date. The clause makes the review explicit so that a quiet iteration is a recorded decision, not an omission.
What good looks like
Use a five-row checklist at the end of every iteration. For each part, record unchanged or changed, with the reason. Where a part changes, create a new version and link it to the finding. Then confirm that the next iteration's inputs are the new versions. Keep the checklist with the evaluation document.
Evidence an assessor asks for
The assessor will ask for the checklist from the last iteration. A missing checklist suggests the iteration updated only the metric it was watching.
- Five-part checklist Each part marked unchanged or changed, with the reason
- New versions The updated profile entries linked to the finding
- Next-iteration inputs Confirmation that the next iteration uses the new versions
- Decision record The decision that no change was needed, where that is the case
Where it commonly fails
- Only the monitoring list is updated
- The checklist is completed from memory after the audit request
- A changed requirement is not versioned
What to do before publication
Add the five-part checklist to the iteration template. It is the simplest way to keep the record coherent.
Read across
- 5.2.1 Requirements for bias consideration
- 6.3 Outputs
- 7.3 Outputs
- 8.3 Outputs
- 9.3.3 Ongoing evaluation: outputs
← 9.3.4.2 Each iteration of the ongoing evaluation9.3.5 Ongoing evaluation: outcomes →