Right now, the 'bench' document gets recorded when the run is complete. This is the document which ties together all observations, without it you can't run the 'show' command for the benchmark. If it were recorded right at the beginning (and then updated at the end with data which is unknown at the start), then 'show' could be run against a benchmark in progress - which would be nice for long running benchmarks, to be able to see their progress. This would also allow for some analysis of benchmarks which fail catastrophically or are terminate early.
Right now, the 'bench' document gets recorded when the run is complete. This is the document which ties together all observations, without it you can't run the 'show' command for the benchmark. If it were recorded right at the beginning (and then updated at the end with data which is unknown at the start), then 'show' could be run against a benchmark in progress - which would be nice for long running benchmarks, to be able to see their progress. This would also allow for some analysis of benchmarks which fail catastrophically or are terminate early.