A latency number is a good place to start asking questions. It is a poor place to stop.
The number and its boundary
MICROJEV’s reported p50 is 7.4 microseconds, with zero hot-path allocations. Before comparing that number with another runtime, the important question is what the timer includes.
Input preparation, execution and output handling can each live inside or outside a benchmark boundary. Move the boundary and the number changes, even if the work performed by the whole application does not.
A median is one view
The median describes the middle observation in a set. It does not describe the slowest request, the distribution under contention, or the behaviour of a different workload. Those are separate measurements.
Likewise, an allocation-free hot path establishes one property of execution. It does not establish overall memory use or make all other sources of latency disappear.
What belongs beside the result
A reproducible performance write-up needs the workload, hardware, build configuration, measurement method and a description of what is timed. A breakdown of where MICROJEV spends its time belongs alongside those details.
That breakdown is not published here yet. For now, the useful habit is to keep the reported result attached to its limits, and to resist filling in the missing context with assumptions.