PRISMA 2020 asks for the eligibility criteria, every information source consulted, and the full search strategy for at least one database, reproduced verbatim — the actual query string, not a description of it.
That last requirement is stricter than it looks, and it is the one that separates a review from an opinion with references. A prose summary ("we searched PubMed and Embase for relevant terms") cannot be re-run. A verbatim query with its date and result count can, by anyone, including you in six months when a referee asks.
The reason to write criteria before screening is mechanical rather than moral: criteria written afterwards describe the papers you kept. They will feel principled and they will have been fitted to the data, and PRISMA's excluded-with-reasons requirement is what exposes it — you end up unable to state a reason that does not amount to "it did not fit".
When reality forces a change mid-screening, amend the note in place, date it, and say what triggered it. Do not silently rewrite. The amendment is itself part of what makes the result trustworthy, it costs nothing to keep, and a reviewer who can see when a criterion moved can judge whether it moved for a good reason.