The expected output of an Example, accepted under any one of three
interchangeable keys. All three normalize to the same slot: when recorded to
Phoenix the value becomes the dataset example's output, and it is exposed
to evaluators as expected on EvaluatorParams. At most one key may be set.
expected — the canonical name (the ground-truth / reference output).
reference — alias preferred by frameworks that name the slot "reference".
output — alias for callers who think in terms of the example's output.
Modeled as a union so supplying more than one key at a time is a type error.
The expected output of an
Example, accepted under any one of three interchangeable keys. All three normalize to the same slot: when recorded to Phoenix the value becomes the dataset example'soutput, and it is exposed to evaluators asexpectedonEvaluatorParams. At most one key may be set.expected— the canonical name (the ground-truth / reference output).reference— alias preferred by frameworks that name the slot "reference".output— alias for callers who think in terms of the example'soutput.Modeled as a union so supplying more than one key at a time is a type error.