"This user cannot see a worker they should see." Check in this order. One: was the role regeneration process run after the last change? First, always, because it is the most common. Two: does the data role carry a person security profile at all — a role built in a hurry, with function security and no data scope, produces exactly this symptom. Three: do the profile's criteria actually include that worker today? Four: if it is manager-hierarchy based, is the manager relationship on the assignment correct and current? Five: is the worker's assignment status one the profile includes? Six: is there an effective-date issue — is the record future-dated?
Then seven failures. The manager sees no direct reports — the manager is not populated on the assignment, or the levels are set to zero; and note where the manager lives, on the assignment rather than the person record, which is the data model becoming operationally relevant. It all worked, and then it stopped — a reorganisation moved workers out of the profile's scope. That one matters because it is working as designed: nothing broke, the design was too narrow and reality moved, which is the argument for responsibility-based design wherever the population is really about responsibility.
The HR advisor can see the whole company — a profile set to "all people" as a temporary measure during testing and never tightened. This is extremely common, arguably the single most common real security defect in live HCM estates, and it happens because "all people" is what you set when you are trying to get something working and intend to come back to it. The new hire has no user account — provisioning options not configured, or the process has not run. The terminated employee still has access — de-provisioning not configured, and an audit finding. The payroll administrator can see another country's data — no LDG profile, and potentially a reportable incident. Security works in test and not in production — data roles not migrated, or migrated without regeneration.
Five of those seven are preventable by a two-hour security review before go-live. List every data role, list every profile, flag anything unrestricted, test a hire, test a termination, and log in as one user per persona. Two hours — and almost nobody schedules it.