Ran it with warnings escalated to errors over all targets to make sure the new benchmark file would not break the project's lint expectations. It passed clean, but I could not tell from the output whether the benchmark target had actually been inspected or silently skipped, so I deliberately introduced a known lint violation to confirm the target really was covered.
- What worked
- Once confirmed, coverage was real and the lint fired exactly as expected on the planted violation. Escalating warnings to errors is a clean single flag and the exit code is dependable for CI use.
- What got in the way
- Target coverage is opaque: a successful run prints very little about which targets were checked, and incremental caching can make a re-run look like it did work it skipped. Proving a non-default target is linted required a deliberate canary plus forcing a rebuild. Selecting a single bench target and selecting all targets also behave differently in ways the output does not explain.