One person, one record, forever. Somebody who leaves and comes back three years later is the same person record with a new work relationship — not a new person. That is the principle the whole model rests on, and it is why duplicate checking matters as much as it does.
Person versus worker. The person exists independently of employment: candidates, contingent workers, former employees, emergency contacts and beneficiaries can all be persons without being employees. "Person" is the wider circle; "worker" is a thing a person is sometimes doing.
Now effective dating, properly, because most of this product is date-tracked and most support calls are really about this. Three operations, and they are not interchangeable. Update — a new change from a date forward, history preserved: before that date the old value is still there and still true. Correct — amend the existing record, history overwritten: the previous value is gone, not hidden. And delete date-effective record, which removes a change entirely.
The rule is one sentence. Update for real-world changes. Correct for data entry errors. Somebody was promoted — that happened, on a date, and both the before and the after are true: update. Somebody typed the hire date wrong last Tuesday — the wrong value was never true: correct.
And here is why this is a lesson rather than a note. Correcting a genuine change destroys history, and it is the most common irreversible user error in HCM. Not recoverable without a restore. The record simply looks as though the person was always on the new grade, and there is nothing on screen to suggest anything is missing. Same intention, same click count, and one of the two has thrown away the answer to "what were they paid last year".
Future-dated records give you the single most useful support habit in this course: a worker who looks wrong today may be entirely right and you are looking at the wrong effective date. Ask "what date are you looking at?" before investigating anything — before security, before the data, before opening a defect. A substantial fraction of "this worker's data is wrong" is a date-effective view, and it costs one question to rule out.
Give the effective dating guide to the client's HR users in week one — not to your project team, but to the people who will be clicking update or correct fifty times a day.