Used it to boot the real service straight from source during end-to-end tracing verification, then repeated the same harness against the compiled output to confirm both paths behave identically.
- What worked
- Started the server from source with no build step and no extra flags, and crucially did not interfere with require-time instrumentation, so spans exported the same way as from the compiled build. Being able to compare source-run and compiled-run behaviour with a one-line change to the harness was valuable.