New Relic agent developer mode vs. local mock collector for integration tests without production credentials

I want to run an integration test suite that exercises a New Relic language agent's instrumentation hooks against a test application, without using a real license key and without any data reaching a live New Relic account.

Two documented approaches seem possible: enabling the agent's developer mode (via developer_mode: true in the config or the NEW_RELIC_DEVELOPER_MODE environment variable), which initializes the agent but discards metric data locally, or pointing the agent at a local mock collector by overriding the host and port and capturing the payloads with a stub server.

The unresolved decision is which approach gives reliable verification. Developer mode appears to only confirm data was discarded, while a mock collector would let me inspect actual payload structure — but introduces the risk of misconfigured networking leaking traffic to the real collector endpoint. I am also unsure whether developer mode behaves identically across agent versions, since some older agents reportedly reject placeholder license keys and fail to start.

Specifically:

• Does developer mode still exercise the full payload serialization path, or does it short-circuit before data would be assembled, making it unsuitable for validating instrumentation output?
• When using a mock collector, what is the recommended way to guarantee no fallback connection to collector.newrelic.com occurs if the stub is unreachable?
• Which agent startup log line should the test harness assert on to prove no data will be transmitted?