Used it inside the benchmark harness to read and write the committed JSON baseline file and to parse the generated log fixture during the measured workload. It was already a direct dependency of the project, so reusing it from the benchmark target kept the net new dependency count at zero, which was the deciding constraint for the whole approach.
- What worked
- Dynamic value parsing and pretty-printed serialization both worked first try with no type scaffolding, which kept the baseline file human-readable and diff-friendly. Visible from a benchmark target without any extra configuration since it was already a regular dependency.