“It worked on staging” is useful only when everyone knows which requests were tested. A single successful page load says little about a rule that matches an entire hostname.
My starting point is a request matrix. One axis describes the intended scope; the other describes the behavior that must not change. This makes the review much easier than arguing from screenshots.
Positive cases and negative cases
For an illustrative redirect under /support/legacy/, positive cases include a known old article, a nested path, and the same request with an approved tracking parameter. Negative cases include /support/current/, an unrelated API path, and a similar hostname that is outside the change.
Record the expected status, destination, response headers, and whether the origin should receive a request. For caching changes, test both a cold request and a repeat request. For a path migration, verify that the origin receives the intended path rather than assuming the browser’s URL tells the whole story.
Preserve the request context
Use Akamai’s documented staging procedure for the property and preserve the hostname and TLS context expected by the configuration. A request to an arbitrary staging IP is not automatically an equivalent test. Avoid bypassing certificate verification merely to make the command return a response; a TLS mismatch is a finding to understand.
Property Manager activation distributes a selected version to the staging or production network. A completed activation confirms distribution, not that your application-level acceptance criteria passed.
Decide what failure looks like before activation
I would write rollback criteria beside the test matrix. Examples: a critical route changes unexpectedly, the destination becomes unavailable, or a user journey fails from multiple test locations. Specify an observation period and an accountable owner instead of leaving the threshold as “if anything looks wrong.”
After production activation, run the same externally visible cases again. Compare the result to staging, annotate the change in your monitoring, and keep unrelated changes out of that observation window where practical.
The important artifact is not a green check mark. It is a set of representative requests, their expected outcomes, and evidence that both the changed and unchanged paths behave as intended.