Forming a Scrum team should include all the skills necessary to develop a user story in order to deliver a potentially deliverable product increment with each sprint.
In traditional organizations, however, I always encounter a fundamental mistrust of the integration of testers in the Scrum teams. Instead, a separate test team is to remain, which is then responsible for regression tests, load and performance tests and the test automation. The rationale for this type of organization is the so-called independence of the testers.
I have several problems with this view of things. Scrum makes the team fully responsible for the results. The establishment of an "independent" test team assumes that the Scrum team does not live up to its responsibilities and would turn a blind eye to errors in the product increment.
Another danger associated with the independent test team is that the testers become test report reporters who are not involved in the elimination of the problem.
In the scrum sense, we prefer problem solvers. The tester in the Scrum team, as well as all developers responsible for the delivery of a flawless product increment and will make every effort to fix it or have it fixed when uncovering an error. Another advantage of the tester in the team is the simple possibility to develop automated tests in step with the implementation of the user stories.
the procedure described already shows a part of the problem: the lead time for a new Product Backlog Item increases to several Sprints: 1 Sprint implementation plus 1 Sprint deferred test (plus possibly another Sprint error correction, if unfortunately this is no longer possible, without the commitment to break the current sprint, and considered less important). This results in further problems: does one need 2 Definition of Done? When does the PO take the item? Does he take it off twice? How much buffer does the deployment team need to keep in fix to fix the returned bugs? Not to mention the context switch that becomes necessary. Pull testers and developers together and try to avoid mistakes instead of finding (with 1 sprint offset)? etc etc
How to change this problem?