The 'reference.json' file hold information about the reference values
for benchmark measurements.

The file specifies a set of measurement names and units.  It is used
to indicate the set of expected measurements that should be returned
by a benchmark, as well as to indicate the units for each measurement name.

The file is specified in "split-out" format, using nested json structures
to indicate a measurement's test_set name, test_case name, and measurement_name.

Here is an example from Benchmark.cyclictest.  In this example, the
fully qualified measurement names in tguid form would be:
 * default.latencies.max_latency
 * default.latencies.min_latency
 * default.latencies.avg_latency

Example:
{{{#!YellowBox
{
    "test_sets":[
        {
            "name":"default",
            "test_cases":[
                {
                    "name":"latencies",
                    "measurements":[
                        {
                            "name":"max_latency",
                            "unit":"us"
                        },
                        {
                            "name":"min_latency",
                            "unit":"us"
                        },
                        {
                            "name":"avg_latency",
                            "unit":"us"
                        }
                    ]
                }
            ]
        }
    ]
}
}}}

For each of these, the 'units' for the measurement are indicated.
In this particular case, the units are micro-seconds (abbreviated to 'us').

This file is read by the Fuego results parser to provide the units for each measurement, so they can be displayed in results visualization output.

See also [[Benchmark parser notes]], [[parser_func_process|process]], and [[function_bench_processing|bench_processing]]