Two deployables that were never deployed separately

Forty-four releases over three years, of which forty-one changed both services in the same window.

the number that decided it:

  releases touching only service A     2
  releases touching only service B     1
  releases touching both              41

and of the three, two were dependency bumps.

so the deployment independence the split existed to
provide had been exercised once, deliberately, in
three years.

Counting releases is a five-minute query against the git history and it answers the question that the architecture discussion had been answering with opinions. One genuine independent deploy in three years is not a boundary that is earning two pipelines, two runbooks and a retry policy.