Summary

• When `rli bmj watch` is called on a job that has already completed, it now displays the total elapsed time before printing the results table
• Elapsed time is inferred by fetching each `BenchmarkRun` associated with the job's `benchmark_outcomes` and computing `max(start_time_ms + duration_ms) - job.create_time_ms`
• The existing live-monitoring path (where the job completes while watch is running) is unchanged

How it works

`BenchmarkJobView` has `create_time_ms` (start) but no end time field. `BenchmarkRunView` has both `start_time_ms` and `duration_ms`, so `end_time = start_time_ms + duration_ms`. The job is considered complete when the last benchmark run finishes, so total elapsed = `max(run end times) - job.create_time_ms`.

Note: the `BenchmarkJobView` API has no update endpoint and no metadata field, so caching the computed end time server-side is not currently feasible.

Test plan

• Run `rli bmj watch ` and verify total elapsed time appears before the results table
• Run `rli bmj watch ` and verify the live monitoring path is unchanged
• Verify graceful handling when benchmark run data is unavailable (no elapsed time shown, no crash)

🤖 Generated with Claude Code