You are asked to warrant Arabic quality on work your client will review in a language half your team does not read. These are the checks that let you show the work rather than describe it, and what to hand a client at the end of it.
The client's Arabic reviewer finds problems your QA pass could not see, late, and the conversation becomes about whether they are real. A finding with a screenshot and a selector is not an opinion.
Arabic quality is warranted in a sentence and audited in a meeting. Anything that cannot be re-run cannot be checked later, which means it cannot be defended later either.
A physical margin in a shared component library is one mistake that arrives in every project built on it. Scanning each project finds it three times; finding it once is cheaper.
Re-running a manual QA pass produces a second opinion, not a diff. Without a stable identity per finding, fixed and never-detected look identical.
Each project is its own target with its own history, so a client's site is scored against itself rather than against a benchmark nobody agreed to.
Findings carry a stable signature, so a re-scan says what was fixed, what is new and what did not move. That is the artefact a handover needs.
Expiring, not indexed, and it says how many pages the score covers. A client reading a score with no scope has no way to tell a five page crawl from a fifty page one.
Every finding is a statement about a rendered page. It cannot tell you your component library is wrong, only that three of your projects show the same defect, and it will not review a design before it is built. A scanner that guessed at intent would be a scanner that argued with you.
Or run a free scan on a live client project first. It is one page and needs no account.
Deciding between approaches first? Compare the four.